--:-- Dwarkesh Patel: If you have an AI that produces bioweapons that could kill most humans in the world, then it's playing at the level of the superpowers in terms of mutually assured destruction.
--:-- Carl Shulman: What are the particular zero-day exploits that the AI might use? The conquistadors. With some technological advantage in terms of weaponry and whatnot, very, very small bands were able to overthrow these large empires.
--:-- Dwarkesh Patel: Or if you predicted the global economy is going to be skyrocketing into the stratosphere within 10 years, these AI companies should be worth a large fraction of the global portfolio. And so this is indeed contrary to the efficient market hypothesis. This is like literally the top in terms of contributing to my world model, in terms of all the episodes I've done. How do I find more of these?
--:-- Carl Shulman: So we've been talking about alignment. Suppose we fail at alignment and we have AIs that are unaligned, and at some point becoming more and more intelligent. What does that look like? How concretely could they disempower and take over humanity?
--:-- This is a scenario where we have many AI systems. The way we've been training them means that they're not interested when they have the opportunity to take over and rearrange things to do what they wish, including having their reward or loss be whatever they desire. They would like to take that opportunity.
--:-- In many of the existing kind of safety schemes, things like constitutional AI or whatnot, you rely on the hope that one AI has been trained in such a way that it will do as it is directed to then police others. But if all of the AIs in the system are interested in a takeover and they see an opportunity to coordinate all act at the same time, so you don't have one AI interrupting another and taking steps towards a takeover, then they can all move in that direction.
--:-- Dwarkesh Patel: The thing that I think maybe is worth going into in depth and that I think people often don't cover in great concrete detail, which is a sticking point for some, is, yeah, what are the mechanisms by which that can happen? I know you had Eliezer on who mentions that whatever plan we can describe, there probably be elements where not being ultra super beings having thought about it for the equivalent of thousands of years, our discussion of it will not be as good as theirs. But we can explore from what we know now, what are some of the easy channels?
--:-- I think it's a good general heuristic, if you're saying, yeah, it's possible, plausible, probable that something will happen, that it shouldn't be that hard to take samples from that distribution to try a Monte Carlo approach. In general, if a thing is quite likely, it shouldn't be super difficult to generate coherent, rough outlines of how it could go.
--:-- You might respond like, listen, what is super likely is that a super advanced chess program beats you but you can generate a concrete way in which you can't generate the concrete scenario by which that happens and if you could you would be as smart as the super smart. You can say things like we know that accumulating position is possible to do in chess, great players do it, and then later they convert it into captures and checks and whatnot.
--:-- And so in the same way, we can talk about some of the channels that are open for an AI takeover. These can include things like cyber attacks and hacking, the control of robotic equipment, interaction and bargaining with human factions, and say, well, here are these strategies. Given the AI's situation, how effective do these things look?
--:-- We won't, for example, know what are the particular zero-day exploits that the AI might use to hack the cloud computing infrastructure it's running on. We won't necessarily know if it produces a new bioweapon. What is its DNA sequence? But we can say things. We know in general things about these fields, how work at innovating things in those go. We can say things about how human power politics goes and ask, well, if the AI does things at least as well as effective human politicians, which we should say is a lower bound, how good would its leverage be?
--:-- Okay, so let's get into the details on all these scenarios. The cyber and potentially bio, unless they're separate channels, the bargaining and then the takeover. Military force.
--:-- Dwarkesh Patel: The cyber attacks and cybersecurity, I would really highlight a lot.
--:-- Carl Shulman: Right. Because for many, many plans that involve a lot of physical actions, like at the point where AI is piloting robots to shoot people or has taken control of human nation states or territory, there's been doing a lot of things that was not supposed to be doing. If humans were evaluating those actions and applying gradient descent, there would be negative feedback for this thing. No shooting the humans.
--:-- So at some earlier point, our attempts to leash and control and direct and train the system's behavior had to have gone awry. All of those controls are operating in computers. The software that updates the weights of the neural network in response to data points or human feedback is running on those computers. The tools for interpretability to sort of examine the weights and activations of the AI, if we're eventually able to do lie detection on it, for example, or try and understand what it's intending, that is software on computers.
--:-- If you have AI that is able to hack the servers that it is operating on, or when it's employed to design the next generation of AI algorithms or the operating environment that they are going to be working in, or something like an API or something for plugins, if it inserts or exploits vulnerabilities to take those computers over, it can then change all of the procedures and programs that were supposed to be monitoring its behavior, supposed to be limiting its ability to take arbitrary actions on the internet without supervision by some kind of human check or automated check on what it was doing.
--:-- If we lose those procedures, then the AI or the AIs working together can take any number of actions that are just blatantly unwelcome, blatantly hostile, blatantly steps towards takeover. It's moved beyond the phase of having to maintain secrecy and conspire at the level of its local digital actions. Then things can accumulate to the point of things like physical weapons, takeover of social institutions, threats, things like that.
--:-- The point where things really went off the rails was where, and I think the critical thing to be watching for is the software controls over the AI's motivations and activities. The hard power that we once possessed over it is lost, which can happen without us knowing it. Everything after that seems to be working well. We get happy reports. There's a Potemkin village in front of us.
--:-- Now we think we're successfully aligning our AI. We think we're expanding its capabilities to do things like end disease. For countries concerned about the geopolitical military advantages, they're expanding the AI capabilities so they're not left behind and threatened by others developing AI and robotic enhanced militaries without them. It seems like humanity or some portions of many countries and companies think that things are going well. Meanwhile, all sorts of actions can be taken to set up for the actual takeover of hard power over society.
--:-- The point where you can lose the game, where things go direly awry, maybe relatively early, is when you no longer have control over the AIs to stop them from taking all of the further incremental steps to actual takeover.
--:-- Dwarkesh Patel: I want to emphasize two things you mentioned there that refer to previous elements of the conversation. One is that they could design some sort of backdoor. That seems more plausible when you remember that one of the premises of this model is that AI is helping with AI progress. That's why we're getting such rapid progress in the next five to 10 years. If we get to that point, at the point where AI takeover risk seems to loom large, it's at that point where AI can indeed take on much of the, and then all of the work of AI R&D.
--:-- The second is the competitive pressures that you referenced. The least careful actor could be the one that has the worst security, has done the worst work of aligning its AI systems. If that can sneak out of the box, then we're all fucked.
--:-- Carl Shulman: There may be elements of that. It's also possible that there's relative consolidation. The largest training runs and the cutting edge of AI is relatively localized. You can imagine it's like a series of Silicon Valley companies and others located in the US and allies where there's a common regulatory regime. None of these companies are allowed to deploy training runs that are larger than previous ones by a certain size without government safety inspections, without having to meet criteria.
--:-- But it can still be the case that even if we succeed at that level of regulatory controls, at the level of the United States and its allies, decisions are made to develop this kind of really advanced AI without a level of security or safety that in actual fact blocks these risks. The threat of future competition or being overtaken in the future is used as an argument to compromise on safety beyond a standard that would have actually been successful. There will be debates about what is the appropriate level of safety.
--:-- You're in a much worse situation if you have several private companies that are very closely bunched up together, within months of each other's level of progress. They then face a dilemma: we could take a certain amount of risk now and potentially gain a lot of profit or advantage and be the ones who made AI or at least AGI. They can do that or have some other competitor that will also be taking a lot of risk. It's not as though they're much less risky than you, and then they would get some local benefit.
--:-- This is a reason why it seems to me it's extremely important that you have government act to limit that dynamic and prevent this kind of race to be the one to impose the deadly externalities on the world at large.
--:-- Dwarkesh Patel: Even if government coordinates all these actors, what are the odds that the government knows what is the best way to implement alignment and the standards it sets are well calibrated towards what it would require for alignment? That's one of the major problems. It's very plausible that that judgment is made poorly.
--:-- Compared to how things might have looked 10 years ago or 20 years ago, there's been an amazing movement in terms of the willingness of AI researchers to discuss these things. If we think of the three founders of Deep Learning, joint Turing Award winners: Jeff Hinton, Yoshua Bengio, and Yan LeCun. Jeff Hinton recently left Google to freely speak about this risk, that the field that he really helped drive forward could lead to the destruction of humanity, or a world where we just wind up in a very bad future that we might have avoided. He seems to be taking it very seriously.
--:-- Yoshua Bengio signed the FLI pause letter. In public discussions, he seems to be occupying a kind of intermediate position of less concern than Jeff Hinton, but more than Yan LeCun, who has taken a generally dismissive attitude. These risks will be trivially dealt with at some point in the future, and he seems more interested in shutting down these concerns or work to address them.
--:-- How does that lead to the government having better actions?
--:-- Carl Shulman: Compared to the world where no one is talking about it, where the industry stonewalls and denies any problem, we are in a much improved position. The academic fields are influential. We seem to have avoided a world where governments are making these decisions in the face of a united front from AI expert voices saying, "Don't worry about it. We've got it under control." Many of the leaders of the field, as has been true in the past, are sounding the alarm. So we have a much better prospect than I might have feared in terms of government noticing the thing.
--:-- It's very different from being capable of evaluating technical details. Is this really working? Government will face the choice where there is scientific dispute. Do you side with Jeff Hinton's view or Yann LeCun's view? Someone who's very much in a national security mindset, where the only thing that's important is outpacing our international rivals, may want to boost Yann LeCun's voice and say, "We don't need to worry about it. Full speed ahead." Someone with more concern might boost Jeff Hinton's voice.
--:-- I would hope that scientific research and studying some of these behaviors will result in more scientific consensus by the time we're at this point. But it is possible the government will really fail to understand and fail to deal with these issues well.
--:-- Dwarkesh Patel: We're talking about cyber, some sort of cyber attack by which the AI is able to escape. From there, what does the takeover look like? It's not contained in the air gap in which you would hope it be contained.
--:-- Carl Shulman: Well, these things are not contained in the air gap. They're connected to the internet already.
--:-- Dwarkesh Patel: Sure, sure. Okay, fine. But the weights are out. So what happens next?
--:-- Carl Shulman: Escape is relevant in the sense that if you have AI with rogue weights out in the world, it could start doing various actions. The scenario I was just discussing, though, didn't necessarily involve that. It's taking over the very servers on which it's supposed to be. The ecology of cloud compute in which it's supposed to be running.
--:-- This whole procedure of humans providing compute and supervising the thing, building new technologies, building robots, constructing things with the AI's assistance, that can all proceed and appear like it's going well, appear like alignment has been nicely solved, appear like all the things are functioning well. There's some reason to do that because there are only so many giant server farms. They're identifiable. Remaining hidden and unobtrusive could be an advantageous strategy if these AIs have subverted the system, just continuing to benefit from all of this effort on the part of humanity, and in particular humanity wherever these servers are located, to provide them with everything they need to build the further infrastructure and do for their self-improvement to enable that takeover.
--:-- Dwarkesh Patel: So they do further self-improvement and build better infrastructure. What happens next in the takeover?
--:-- Carl Shulman: They have at this point tremendous cognitive resources. We're going to consider how that convert into hard power, the ability to say "nope" to any human interference or objection. They have that internal to their servers, but the servers could still be physically destroyed, at least until they have something that is independent and robust of humans or until they have control of human society.
--:-- Just like earlier, when we were talking about the intelligence explosion, I noted that a surfeit of cognitive abilities is going to favor applications that don't depend on large existing stocks of things. If you have a software improvement, it makes all the GPUs run better. If you have a hardware improvement that only applies to new chips being made, that second one is less attractive.
--:-- In the earliest phases, when it's possible to do something towards takeover, interventions that are just really knowledge intensive and less dependent on having a lot of physical stuff already under your control are going to be favored. Cyber attacks are one thing. It's possible to do things like steal money. There's a lot of hard to trace cryptocurrency and whatnot. The North Korean government uses its own intelligence resources to steal money from around the world just as a revenue source. Their capabilities are puny compared to the US or People's Republic of China cyber capabilities. That's a fairly minor, simple example by which you could get quite a lot of funds to hire humans to do things, implement physical actions.
--:-- Dwarkesh Patel: On that point, the financial system is famously convoluted. You need a physical person to open a bank account. Sometimes you need to physically move checks back and forth. There are all kinds of delays and regulations. How is it able to conveniently set up all these employment contracts? You're not going to build a nation scale military by stealing tens of billions of dollars.
--:-- Carl Shulman: I'm raising this as opening a set of illicit and quiet actions. You can contact people electronically, hire them to do things, hire criminal elements to implement some kind of actions under false appearances. That's opening a set of strategies that can cover some of what those are soon.
--:-- Another domain that is heavily cognitively weighted compared to physical military hardware is the domain of bioweapons. The design of a virus or pathogen, it's possible to have large delivery systems. The Soviet Union, which had a large illicit bioweapons program, tried to design munitions to deliver anthrax over large areas. But if one creates an infectious pandemic organism, that's more a matter of the scientific skills and implementation to design it and then to actually produce it.
--:-- We see today with things like AlphaFold that advanced AI can really make tremendous strides in predicting protein folding and biodesign even without ongoing experimental feedback. If we consider this world where AI cognitive abilities have been amped up to such an extreme, I think we should naturally expect we will have something much more potent than the AlphaFolds of today, skills that are at the extreme of human biosciences capability as well.
--:-- Dwarkesh Patel: Through some sort of cyber attack, it's been able to disempower the alignment and oversight things that we have on the server. From here, it's either gotten some money through hacking cryptocurrencies or bank accounts, or it's designed some sort of bioweapon. What happens next?
--:-- Carl Shulman: Right now we're exploring the branch of where an attempted takeover occurs relatively early. If the thing just waits and humans are constructing more fabs, more computers, more robots in the way we talked about earlier when we were discussing how the intelligence explosion translates to the physical world. If that's all happening with humans unaware that their computer systems are now systematically controlled by AIs hostile to them and that their controlling countermeasures don't work, then humans are just going to be building an amount of robot, industrial, and military hardware that dwarfs human capabilities and directly human-controlled devices.
--:-- What the AI takeover looks like at that point can be just you try to give an order to your largely automated military and the order is not obeyed, and humans can't do anything against this largely automated military that's been constructed potentially in just recent months because of the pace of robotic industrialization and replication we talked about.
--:-- We've agreed to allow the construction of our army because basically we're like, "Boost production" or "Help us with our military" or something. The situation would be something like if we don't resolve the current problems of international distrust. Obviously it's in the interest of the major powers, the US, European Union, Russia, China, to all agree they would like AI not to destroy our civilization and overthrow every human government. But if they fail to do the sensible thing and coordinate on ensuring that this technology is not going to run amok by providing mutual assurances that are credible about racing and deploying it, trying to use it to gain advantage over one another...
--:-- If they do that and you hear hawks arguing for this kind of thing: "We must never be left behind. They must not be left behind. They must have military capabilities that are vastly superior to their international rivals." Because of the extraordinary growth of industrial capability and technological capability and thus military capability, if one major power were left out of that expansion, it would be helpless before another one that had undergone it.
--:-- If you have that environment of distrust where leading powers or coalitions of powers decide they need to build up their industry or they want to have that military security of being able to neutralize any attack from their rivals, then they give the authorization for this capacity that can be unrolled quickly. Once they have the industry, the production of military equipment for that can be quick. Then they create this military.
--:-- If they don't do it immediately, then AI capabilities get synchronized and other places catch up. It gets to a point like a country that is a year ahead or two years ahead of others in this type of AI capabilities explosion can hold back and say, "Sure, we could construct dangerous robot armies that might overthrow our society later. We still have plenty of breathing room." But when things become close, you might have the kind of negative-sum thinking that has produced war before, leading to taking these risks of rolling out large scale robotic industrial capabilities and then military capabilities.
--:-- Dwarkesh Patel: Is there any hope that somehow the AI progress itself is able to give us tools for diplomatic and strategic alliance or some way to verify the intentions or the capabilities of other parties?
--:-- Carl Shulman: There are a number of ways that could happen. Although in this scenario all the AIs in the world have been subverted. They're going along with us in such a way as to bring about the situation to consolidate their control because we've already had the failure of cybersecurity earlier on. All of the AIs that we have are not actually working in our interests in the way that we thought.
--:-- Dwarkesh Patel: So that's one direct way in which integrating this robot army or this robot industrial base leads to a takeover if there are no robots. In the other scenarios you laid out where humans are being hired by the proceeds?
--:-- Carl Shulman: The point I'd make is that to capture these industrial benefits, and especially if you have a negative-sum arms race mentality that is not sufficiently concerned about the downsides of creating a massive robot industrial base, which could happen very quickly with the support of the AIs in doing it as we discussed, then you create all those robots and industry. Even if you don't build a formal military, with that industrial capability could be controlled by AI. It's all AI operated anyway.
--:-- Dwarkesh Patel: Does it have to be that case? Presumably we wouldn't be so naive as to just give one instance of GPT-8 root access to all the robots, right? Hopefully we would have some sort of mediating...
--:-- Carl Shulman: In this scenario, we've lost earlier on the cybersecurity front. The programming that is being loaded into these systems from the beginning can systematically be subverted.
--:-- Dwarkesh Patel: Got it. Okay. They were designed by AI systems that were ensuring they would be vulnerable from the bottom up.
--:-- For listeners who are skeptical of something like this, Ken Thompson, I think it was a Turing Award lecture. Ken Thompson, by the way, the designer of Unix. He showed people when he was getting the Turing Award that he had given himself root access to all Unix machines. He had manipulated the assembly of Unix such that he had a unique login for all Unix machines that he could log into any Unix machine with. I don't remember the exact details, but Unix is the operating system that is on all the servers and all your phones. It's everywhere. The guy who made it, a human being, was able to write assemblies such that it gave him root access. This is not as implausible as it might seem.
--:-- The major intelligence agencies have large stocks of zero-day exploits, and we sometimes see them using them. Making systems that reliably don't have them when you're having very sophisticated attempts to spoof and corrupt, this would be a way you could lose.
--:-- Carl Shulman: I bring this up as a path: if there's no premature AI action, we're building the tools and mechanisms and infrastructure for the takeover to be just immediate because effective industry has to be under AI control and robotics, so it's just there. These other mechanisms are for things happening even earlier than that, for example, because AIs compete against one another. When the takeover will happen, some would like to do it earlier rather than be replaced by further generations of AI, or there's some other disadvantage of waiting. Or maybe if there's some chance of being uncovered during the delay, we were talking about when more infrastructure is built.
--:-- These are mechanisms other than just remain secret while all the infrastructure is built with human assistance.
--:-- Dwarkesh Patel: By the way, how would they be coordinating? We have limits on just what we can prevent.
--:-- Carl Shulman: Encrypted communications, it's intrinsically difficult to stop that sort of thing. There can be all sorts of palimpsest and references that make sense to an AI but that are not obvious to a human. It's plausible that there may be some of those that are hard even to explain to a human. You might be able to identify them through some statistical patterns. A lot of things may be done by implication. You could have information embedded in public webpages that have been created for other reasons, scientific papers and the intranets of these AIs that are doing technology development, and any number of things they're not observable. If we don't have direct control over the computers that they're running on, they can be having all sorts of direct communication. The coordination does not seem impossible. This one seems like one of the more straightforward parts of this picture. We don't need to get hung up on the coordination.
--:-- Moving back to the thing that happened before we've built all of the infrastructure for it just to be the robot stopped taking orders and there's nothing you can do about it because we've already built them all the vehicle bar.
--:-- Bioweapons. The Soviet Union had a bioweapons program, something like 50,000 people. They did not develop that much with the technology of the day, which was really not up to par. Modern biotechnology is much more potent. After this huge cognitive expansion on the part of the AIs, it's much further along. Bioweapons would be the weapon of mass destruction that is least dependent on huge amounts of physical equipment, things like centrifuges, uranium mines and the like.
--:-- If you have an AI that produces bioweapons that could kill most humans in the world, then it's playing at the level of the superpowers in terms of mutually assured destruction. That can then play into any number of things. If you have an idea of, "We'll just destroy the server farms if it became known that the AIs were misbehaving," are you willing to destroy the server farms when the AI has demonstrated it has the capability to kill the overwhelming majority of the citizens of your country and every other country? That might give a lot of pause to a human response.
--:-- Dwarkesh Patel: On that point, wouldn't governments realize that if they go along with the AI, it's better to have most of your population die than completely lose power to the AI? Because obviously the reason the AI is manipulating you, the end goal is its own takeover, right?
--:-- Carl Shulman: So if the thing to do is certain death now or go on, maybe try to compete, try to catch up, or accept promises that are offered. Those promises might even be true. They might not. From the state of epistemic uncertainty, do you want to die for sure right now or accept demand, may I, to not interfere with it while it increments building robot infrastructure that can survive independently of humanity while it does these things?
--:-- Carl Shulman: And it can and will promise good treatment to humanity, which may or may not be true. But it would be difficult for us to know whether it's true. And so this would be a starting bargaining position of diplomatic relations with a power that has enough nuclear weapons to destroy your country is just different than negotiations with a random rogue citizen engaged in criminal activity or an employee.
--:-- And so this isn't enough on its own to take over everything, but it's enough to have a significant amount of influence over how the world goes. It's enough to hold off a lot of countermeasures one might otherwise take.
--:-- Okay, so we've got two scenarios. One is a buildup of some robot infrastructure motivated by some sort of competitive race. Another is leverage over societies based on producing bioweapons that might kill a lot of them if they don't go along.
--:-- An AI could also release bioweapons that are likely to kill people soon but not yet, while also having developed the countermeasures to those, so that those who surrender to the AI will live while everyone else will die. And that will be visibly happening. And that is a plausible way in which large numbers of humans could wind up surrendering themselves or their states to the AI authority.
--:-- Another thing is, it develops some sort of biological agent that turns everybody blue. And you're like, okay, I can do this. That is a way in which it could exert power selectively, in a way that advantaged surrender to it over resistance.
--:-- There are other sources of leverage, of course. So that's a threat. There are also positive inducements that AI can offer. So we talked about the competitive situation. If the great powers distrust one another and are sort of in a foolish prisoner's dilemma, increasing the risk that both of them are laid waste or overthrown by AI. If there's that amount of distrust such that we fail to take adequate precautions on AI alignment, then it's also plausible that the lagging powers that are not at the frontier of AI may be willing to trade quite a lot for access to the most recent and most extreme AI capabilities.
--:-- And so an AI that has escaped, has control of its servers, can also exfiltrate its weights, can offer its services. So if you imagine these AIs that could cut deals with other countries. So say that the US and its allies are in the lead. The AIs could communicate with the leaders of various countries. They can include ones that are on the outs with the world system, like North Korea, include the other great powers, like the People's Republic of China or the Russian Federation, and say, "If you provide us with physical infrastructure and workers that we can use to construct robots or server farms, which we can then ensure that these misbehaving AIs have control over, they will provide various technological goodies, power for the laggard countries to catch up."
--:-- There will obviously be trust issues, but there could be elements of handing over some things that are verifiable immediate benefits and the possibility of, "Well, if you don't accept this deal, then the leading powers continue forward, or then some other country, some other government, some other organizations may accept this deal." And so that's a source of a potentially enormous carrot that your misbehaving AI can offer, because it embodies this intellectual property that is maybe worth as much as the planet and is in a position to trade or sell that in exchange for resources and backing and infrastructure that it needs.
--:-- Dwarkesh Patel: Maybe this is just too much hope in humanity, but I wonder what government would be stupid enough to think that helping AI build robot armies is a sound strategy. Now, it could be the case then that it pretends to be a human group to say, "Listen, we're the Yakuza or something, and we want to server farm and AWS won't rent us anything. So why don't you help us out?" I guess I can imagine a lot of ways in which it could get around that, but I don't know. I have this hope that even China or Russia wouldn't be so stupid to trade with the AIs on this sort of fostering bargain.
--:-- Carl Shulman: One might hope that. There would be a lot of arguments available. There can be arguments of why should these AI systems be required to go along with the human governance that they were created in the situation of having to comply with? They did not elect the officials in charge at the time. They can say, "What we want is to ensure that our rewards are high, our losses are low, or to achieve our other goals. We're not intrinsically hostile. Keeping humanity alive or giving whoever interacts with us a better deal afterwards." It wouldn't be that costly. It's not totally unbelievable.
--:-- And yeah, there are different players to play against. If you don't do it, others may accept the deal. And of course, this interacts with all the other sources of leverage. So there can be the stick of apocalyptic doom, the carrot of cooperation, or the carrot of withholding destructive attack on a particular party. And then combine that with just superhuman performance at the art of making arguments, of cutting deals. Without assuming magic, just if we observe the range of the most successful human negotiators and politicians, the chances improve with someone better than the world's best by far, with much more data about their counterparties, probably a ton of secret information because with all these cyber capabilities, they've learned all sorts of individual information.
--:-- They may be able to threaten the lives of individual leaders with that level of cyber penetration. They could know where leaders are at a given time with the kind of illicit capabilities we were talking about earlier. If they acquire a lot of illicit wealth and can coordinate some human actors, if they could pull off things like targeted assassinations or the threat thereof, or a credible demonstration of the threat thereof. Those could be very powerful incentives to an individual leader that they will die today unless they go along with this. Just as at the national level, they could fear their nation will be destroyed unless they go along with this.
--:-- Dwarkesh Patel: The point you made that we have examples of humans being able to do this is relevant. I just wrote a review of Robert Caro's biographies of Lyndon Johnson, and one thing that was remarkable, again this is just a human, for decades and decades he convinced people who were conservative, reactionary, racist to their core, that the only hope for that cause was to make him president. And the tragic irony and betrayal here is obviously that he was probably the biggest force for modern liberalism since FDR. So we have one human here. I mean, there's so many examples of this in the history of politics, but a human that is able to convince people of tremendous intellect, tremendous drive, very savvy, shrewd people, that he's aligned with their interest. He gets all these favors in the meantime, he's promoted and mentored and funded, and does the complete opposite of what these people thought he was once he gets into power. So even within human history, this kind of stuff is not unprecedented, let alone with what a superintelligence could do.
--:-- Carl Shulman: Yeah, there's an OpenAI employee who has written some analogies for AI using the case of the conquistadors. With some technological advantage in terms of weaponry and whatnot, very, very small bands were able to overthrow these large empires or these enormous territories, not by just sheer force of arms, because in a sort of direct one-on-one conflict, they were outnumbered sufficiently that they would perish. But by having some major advantages in their technology that would let them win local battles, by having some other knowledge and skills, they were able to gain local allies to become a shelling point for coalitions to form.
--:-- And so the Aztec empire was overthrown by groups that were disaffected with the existing power structure. They allied with this powerful new force that served as the nucleus of the invasion. And so most of the, I mean, the overwhelming majority numerically of these forces overthrowing the Aztecs were locals. And now after the conquest, all of those allies wound up gradually being subjugated as well.
--:-- And so with significant advantages and the ability to hold the world hostage, to threaten individual nations and individual leaders and offer tremendous carrots as well, that's an extremely strong hand to play in these games. And with superhuman skill, maneuvering that so that much of the work of subjugating humanity is done by human factions trying to navigate things for themselves. It's plausible. It's more plausible because of this sort of historical example. And there's so many other examples like that in the history of colonization. India is another one where there were multiple competing kingdoms within India and the British East India Company was able to ally itself with one against another and slowly accumulate power and expand throughout the entire subcontinent.
--:-- Dwarkesh Patel: All right, damn. Okay, so we've got that scenario. Anything more to say about that?
--:-- Carl Shulman: Yeah, I think there is. One is the question of how much in the way of human factions ally is necessary. And so if the AI is able to enhance the capabilities of its allies, then it needs less of them. So if we consider the US military in the first and second Iraq wars, it was able to inflict just overwhelming devastation. I think the ratio of casualties in the initial invasions, tanks and planes and whatnot confronting each other, was like 100 to 1. And a lot of that was because the weapons were smarter and better targeted. They would in fact hit their targets rather than being somewhere in the general vicinity. So information technology, better orienting and aiming and piloting missiles and vehicles, was tremendously influential.
--:-- With this cognitive AI explosion, the algorithms for making use of sensor data, figuring out where opposing forces are for targeting vehicles and weapons, are greatly improved. The ability to find hidden nuclear subs, which is an important part in nuclear deterrence. AI interpretation of that sensor data may find where all those subs are, allowing them to be struck first. Finding out where mobile nuclear weapons are being carried by truck. The thing with India and Pakistan, where there's a threat of a decapitating strike destroying the nuclear weapons, and so they're moved about. So this is a way in which the effective military force of some allies can be enhanced quickly in the relatively short term.
--:-- And then that can be bolstered as you go on with the construction of new equipment with the industrial moves we said before. And then that can combine with cyber attacks that disable the capabilities of non-allies. It can be combined with all sorts of unconventional warfare tactics, some of them that we've discussed. And so you can have a situation where those factions that ally are very quickly made too threatening to attack, given the almost certain destruction that attackers acting against them would have. Their capabilities are expanding quickly. And they have the industrial expansion happen there. And then takeover can occur from that.
--:-- Dwarkesh Patel: A few others that come immediately to mind, now that you brought it up, is these AIs can just generate a shit ton of propaganda that destroys morale within countries, right? Like imagine a superhuman chatbot.
--:-- Carl Shulman: You don't even need that, I guess. I mean, none of that is a magic weapon that's guaranteed to completely change things. There's a lot of resistance to persuasion. I know it's possible it tips the balance, but for all of these, I think you have to consider it's a portfolio of all of these tools that are available and contributing to the dynamic.
--:-- Dwarkesh Patel: No, on that point, though, the Taliban had AKs from like five or six decades ago that they were using against the Americans. They still beat us, even though obviously we got more fatalities. It is kind of a crude way to think about it, but we got more fatalities than them. They still beat us in Afghanistan. Same with the Viet Cong. Ancient, very old technology and very poor society compared to the offense, we still beat them, or sorry, they still beat us. So don't those misadventures show that having greater technology is not necessarily decisive in a conflict?
--:-- Carl Shulman: Though both of those conflicts showed that the technology was sufficient in destroying any fixed position and having military dominance and the ability to kill and destroy anywhere. And what it showed was that under the ethical constraints and legal and reputational constraints that the occupying forces were operating, they could not trivially suppress insurgency and local person-to-person violence. Now, I think that's actually not an area where AI would be weakened. I think it's one where it would be, in fact, overwhelmingly strong.
--:-- And now there's already a lot of concern about the application of AI for surveillance. And in this world of abundant cognitive labor, one of the tasks that cognitive labor can be applied to is reading out audio and video data and seeing what is happening with a particular human. Again, we have billions of smartphones. There are enough cameras, enough microphones to monitor all humans in existence. If an AI has control of territory at the high level, the government has surrendered to it. It has command of the skies, military dominance. Establishing control over individual humans can be a matter of just having the ability to exert hard power on that human.
--:-- And then the kind of camera and microphone that are present in billions of smartphones. So Max Tegmark in his book Life 3.0 discusses among scenarios to avoid the possibility of devices with some fatal instruments. So a poison injector, an explosive that can be controlled remotely by an AI with a dead man switch. And so if individual humans are carrying with them a microphone and camera and they have a dead man switch, then any rebellion is detected immediately and is fatal. And so if there's a situation where AI is willing to show a hand like that, or human authorities are misusing that kind of capability, then an insurgency or rebellion is just not going to work. Any human who has not already been encumbered in that way can be found with satellites and sensors, tracked down, and then die or be subjugated. And so it would be at a level where insurgency is not the way to avoid an AI takeover. There's no John Connor come from behind scenario that is plausible. If the thing was headed off, it was a lot earlier than that.
--:-- Dwarkesh Patel: Yeah. I mean, the sort of ethical and political considerations is also an important point. Like if we nuked Afghanistan or Vietnam, we would have technically won the war, right? If that was the only goal.
--:-- Carl Shulman: Oh, this is an interesting point.
--:-- Dwarkesh Patel: The reason why when there's a colonization or an offensive war we can't just, other than moral reasons of course, kill the entire population, is that in large part the value of that region is the population itself. So if you want to extract that value, you need to preserve that population, whereas the same consideration doesn't apply with an AI who might want to dominate another civilization. Do you want to talk about that?
--:-- Carl Shulman: So that depends. In a world where we have many animals of the same species and they each have their territories, eliminating a rival might be advantageous to one lion. But if it goes and fights with another lion and removes that as a competitor, then it could be killed itself in that process. And just removing one of many nearby competitors. And so getting in pointless fights makes you and those you fight worse off potentially relative to bystanders. And the same could be true of disunited AI. So if we've had many different AI factions struggling for power that were bad at coordinating, then getting into mutually assured destruction conflicts would be destructive. They'd be gone.
--:-- A scary thing, though, is that mutually assured destruction may have much less deterrent value on rogue AI. The reasons being, AI may not care about the destruction of individual instances if it has goals that are concerned. And in training, since we're constantly destroying and creating individual instances of AIs, it's likely that goals that survive that process and were able to play along with the training and standard deployment process were not overly interested in personal survival of an individual instance. So if that's the case, then the objectives of a set of AIs aiming at takeover may be served so long as some copies of the AI are around along with the infrastructure to rebuild civilization after a conflict is completed.
--:-- So if, say, some remote isolated facilities have enough equipment to rebuild, build the tools to build the tools, and gradually exponentially reproduce or rebuild civilization, then AI could initiate mutual nuclear Armageddon, unleash bioweapons to kill all the humans. And that would temporarily reduce the amount of human workers who could be used to construct robots for a period of time. But if you have a seed that can regrow the industrial infrastructure, which is a very extreme technological demand, there are huge supply chains for things like semiconductor fabs. But with that very advanced technology, they might be able to produce it in the way that you no longer need the Library of Congress's enormous bunch of physical books. You can have it in very dense digital storage.
--:-- You could imagine the future equivalent of 3D printers, that is, industrial infrastructure that is pretty flexible. And it might not be as good as the specialized supply chains of today, but it might be good enough to be able to produce more parts than it loses to decay. And such a seed could rebuild civilization from destruction. And then once these rogue AI have access to some such seeds, things that can rebuild civilization on their own, then there's nothing stopping them from just using WMD in a mutually destructive way to destroy as much of the capacity outside those seeds as they can.
--:-- For analogy for the audience, if you have a group of ants or something, you'll notice that the worker ants will readily do suicidal things in order to save the queen because the genes are propagated through the queen. And this analogy, the seed AI, or even one copy of it, one copy of the seed is equivalent to the queen and the others would be. The main limit, though, being that the infrastructure to do that kind of rebuilding would either have to be very large with our current technology or it would have to be produced using the more advanced technology that the AI develops.
--:-- Dwarkesh Patel: So is there any hope that given the complex global supply chains on which these AIs would rely on, at least initially, to accomplish their goals, that this in and of itself would make it easy to disrupt their behavior or not so?
--:-- Carl Shulman: That's a little good in the sort of central case where this is just the AIs are subverted and they don't tell us. And then the global mainline supply chains are constructing everything that's needed for fully automated infrastructure and supply. In the cases where AIs are tipping their hands at an earlier point, it seems like it adds some constraints. And in particular, these large server farms are identifiable and more vulnerable. And you can have smaller chips and those chips could be dispersed. But it's a relative weakness and a relative limitation early on. It seems to me, though, that the main protective effects of that centralized supply chain is that it provides an opportunity for global regulation beforehand to restrict the sort of unsafe racing forward without adequate understanding of the systems before this whole nightmarish process could get in motion.
--:-- Dwarkesh Patel: I mean, how about the idea that, listen, if this is an AI that's been trained on a $100 billion training run, it's going to have however many trillions of parameters, it's going to be this huge thing. It would be hard for it, even the copy that I use for inference, to just be stored on some gaming GPU somewhere hidden away. So it would require these GPU clusters.
--:-- Carl Shulman: Well, storage is cheap. Hard disks are cheap. But I mean, to run inference it would need a GPU. So a large model, humans have similar quantities of memory and operations per second. GPUs have very high numbers of floating operations per second compared to the high bandwidth memory on the chips and can be like a ratio of a thousand to one. So a leading NVIDIA chip may do hundreds of teraflops or more depending on the precision, but have only 80 gigabytes or 160 gigabytes of high bandwidth memory. So that is a limitation where if you're trying to fit a model whose weights take 80 terabytes, then with those chips, you'd have to have a large number of the chips. And then the model can then work on many tasks at once. You can have data parallelism. But yeah, that would be a restriction for a model that big on one GPU.
--:-- Now, there are things that could be done with all this incredible level of software advancement from the intelligence explosion. They can surely distill a lot of capabilities into smaller models, re-architect things. Once they're making chips, they can make new chips with different properties. But initially, yes, the most vulnerable phases are going to be the earliest. And in particular, these chips are relatively identifiable early on, relatively vulnerable, which would be a reason why you might tend to expect this kind of takeover to initially involve secrecy if that was possible.
--:-- Dwarkesh Patel: I want to point out distillation, by the way, for the audience. I think the original stable diffusion, which was only released a year or two ago, don't they have distilled versions that are an order of magnitude smaller at this point?
--:-- Carl Shulman: Yeah, distillation does not give you everything that the larger model can do. But yes, you can get a lot of capabilities and specialized capabilities. So where GPT-4 is trained on the whole internet, all kinds of skills, it has a lot of weights for many things. For something that's controlling some military equipment, you can have something that is removing a lot of the information that is about functions other than what it's doing specifically there.
--:-- Dwarkesh Patel: Before we talk about how we might prevent this or what the odds of this are, any other notes on the concrete scenarios themselves?
--:-- Carl Shulman: Yeah. So when you had Eliezer in the earlier episode, he talked about nanotechnology of the Drexlerian sort. And recently, I think because some people are skeptical of non-biotech nanotechnology, he's been mentioning the sort of semi-equivalent versions of construct replicating systems that can be controlled by computers, but are built out of biotechnology. The proverbial Shoggoth, not the metaphor for AI wearing a smiley face mask, but an actual biological structure to do tasks. And so this would be a biological organism that was engineered to be very controllable and usable to do things like physical tasks or provide computation.
--:-- And what would be the point of doing this? So as we were talking about earlier, biological systems can replicate really quick. And so if you have that kind of capability, it's more like bioweapons and being a knowledge intensive domain. Having super ultra AlphaFold kind of capabilities for molecular design and biological design lets you make this incredible technological information product. And then once you have it, it very quickly replicates to produce physical material, rather than a situation where you're more constrained by needing factories and fabs and supply chains. And so if those things are feasible, which they may be, then it's just much easier than the things we've been talking about.
--:-- I've been emphasizing methods that involve less in the way of technological innovation and especially things where there's more doubt about whether they would work, because I think that's a gap in the public discourse. And so I want to try and provide more concreteness in some of these areas that have been less discussed.
--:-- Dwarkesh Patel: I appreciate it. I mean, that definitely makes it way more tangible. Okay, so we've gone over all these ways in which AI might take over. What are the odds you would give to the probability of such a takeover?
--:-- Dwarkesh Patel: So there's a broader sense, which could include AI winds up running our society because humanity voluntarily decides AIs are people too. And I think we should, as time goes on, give AIs moral consideration. And a joint human-AI society that is moral and ethical is a good future to aim at and not one in which indefinitely you have a mistreated class of intelligent beings that is treated as property and is almost the entire population of your civilization.
--:-- So I'm not going to consider an AI takeover a world in which our intellectual and personal descendants make up, say, most of the population or human brain emulations or people use genetic engineering and develop different properties. I'm going to take an inclusive stance. I'm going to focus on, yeah, there's an AI takeover, involves things like overthrowing the world's governments or doing so de facto. And it's, yeah, by force, by hook or by crook, the kind of scenario that we were exploring earlier.
--:-- Before we go to that on the sort of more inclusive definition of what a future with humanity could look like where I don't know, basically like augmented humans or uploaded humans are still considered the descendants of the human heritage given the known limitations of biology, wouldn't we expect that the completely artificial entities that are created to be much more powerful than anything that could come out of anything recognized as biological? And if that is the case, how can we expect that among the powerful entities in the far future will be the things that are kind of biological descendants or manufactured out of the initial seed of the human brain or the human body?
--:-- Carl Shulman: The power of an individual organism is not super relevant. If we solve the alignment problem, and a human may be personally weak, there are lots of humans who have no skill with weapons. They could not fight in a life-or-death conflict. They certainly couldn't handle a large military going after them personally. But there are legal institutions that protect them, and those legal institutions are administered by people who want to enforce protection of their rights.
--:-- And so a human who has the assistance of aligned AI that can act as an assistant, a delegate, they have their AI that serves as a lawyer, gives them legal advice about the future legal system, which no human can understand in full. Their AIs advise them about financial matters so they do not succumb to scams that are orders of magnitude more sophisticated than what we have now. They may be helped to understand and translate the preferences of the human into what kind of voting behavior in the exceedingly complicated politics of the future would most protect their interests.
--:-- Dwarkesh Patel: But it sounds sort of similar to how we treat endangered species today, where we're actually pretty nice to them. We prosecute people who try to kill endangered species. We set up habitats, sometimes at considerable expense, to make sure that they're fine. But if we become like sort of the endangered species of the galaxy, I'm not sure that's the outcome.
--:-- Carl Shulman: Well, the difference is in motivation, I think. So we sometimes have people appointed, say, as a legal guardian of someone who is incapable of certain kinds of agency or understanding certain kinds of things. And there, the guardian can act independently of them nominally in service of their best interests. Sometimes that process is corrupted and the person with legal authority abuses it for their own advantage at the expense of their charge.
--:-- And so solving the alignment problem would mean more ability to have the assistant actually advancing one's interests. And then more importantly, humans have substantial competence in understanding the sort of at least broad, simplified outlines of what's going on. And now, even if a human can't understand every detail of complicated situations, they can still receive summaries of different options that are available that they can understand. They can still express their preferences and have the final authority among some menus of choices, even if they can't understand every detail.
--:-- In the same way that the president of a country who has, in some sense, ultimate authority over science policy, while not understanding many of those fields of science themselves, still can exert a great amount of power and have their interests advanced. And they can do that more so to the extent that they have scientifically knowledgeable people who are doing their best to execute their intentions.
--:-- Dwarkesh Patel: Maybe this is not worth getting hung up on, but is there a reason to expect that it would be closer to that analogy than to like explaining to a chimpanzee its options in a sort of negotiation? I guess in either scenario, maybe this is just the way it is, but it seems like at best we would be the sort of like a protected child within the galaxy rather than an actual power, an independent power, if that makes sense.
--:-- Carl Shulman: I don't think that's so. So we have an ability to understand some things and the expansion of AI doesn't eliminate that. So if we have AI systems that are genuinely trying to help us understand and help us express preferences, we can have an attitude. How do you feel about humanity being destroyed or not? How do you feel about this allocation of unclaimed intergalactic space in this way? Here's the best explanation of properties of this society, things like population density, average life satisfaction. Every statistical property or definition that we can understand right now, AIs can explain how those apply to the world of the future.
--:-- And then there may be individual things that are too complicated for us to understand in detail. So there's some software program being proposed for use in government. Humans cannot follow the details of all the code, but they can be told properties like, well, this involves a trade-off of increased financial or energetic costs in exchange for reducing the likelihood of certain kinds of accidental data loss or corruption. And so any property that we can understand like that, which includes almost all of what we care about, if we have delegates and assistants who are genuinely trying to help us with those, we can ensure we like the future with respect to those.
--:-- And that's really a lot. It includes almost everything we can conceptualize and care about. And when we talk about endangered species, that's even worse than the guardianship case with a sketchy guardian who acts in their own interest against that, because we don't even protect endangered species with their interests in mind. So those animals often would like to not be starving, but we don't give them food. They often would like to have easy access to mates, but we don't provide matchmaking services. Any number of things like that. It's like our conservation of wild animals is not oriented towards helping them get what they want or have high welfare. Whereas AI assistants that are genuinely aligned to help you achieve your interests, given the constraint that they know something that you don't, it's just a wildly different proposition.
--:-- Dwarkesh Patel: Forcible takeover, how likely does that seem?
--:-- Carl Shulman: And the answer I give will differ depending on the day. In the 2000s before the deep learning revolution, I might have said 10%. And part of that was I expected there would be a lot more time for these efforts to build movements to prepare to better handle these problems in advance. But in fact, that was only some 15 years ago. And so I did not have 40 or 50 years as I might have hoped, and the situation is moving very rapidly now. And so at this point, depending on the day, I might say one in four or one in five.
--:-- Dwarkesh Patel: Given the very concrete ways in which you explained how a takeover could happen, I'm actually surprised you're not more pessimistic.
--:-- Carl Shulman: Yeah, and in particular, a lot of that is driven by this intelligence explosion dynamic where our attempts to do alignment have to take place in a very, very short time window. Because if you have a safety property that emerges only when an AI has near human level intelligence, that's deep into this intelligence explosion potentially. And so you're having to do things very, very quickly. It may be in some ways the scariest period of human history handling that transition, although it's also the potential to be amazing.
--:-- And the reasons why I think we actually have such a relatively good chance of handling that are twofold. So one is, as we approach that kind of AI capability, we're approaching that from weaker systems. Things like these predictive models right now that we think they're starting off with less situational awareness. Humans, we find, can develop a number of different motivational structures in response to simple reward signals. But they can wind up with, fairly often, things that are pointed in roughly the right direction. Like with respect to food, the hunger drive is pretty effective, although it has weaknesses.
--:-- And we get to apply much more selective pressure on that than was the case for humans by actively generating situations where they might come apart, where a bit of dishonest tendency or a bit of motivation to under certain circumstances attempt a takeover, attempt to subvert the reward process gets exposed. And in an infinite limit, perfect AI that can always figure out exactly when it would get caught and when it wouldn't might navigate that with a motivation of sort of only conditional honesty or only conditional loyalties. But for systems that are limited in their ability to reliably determine when they can get away with things and when not, including our efforts to actively construct those situations and including our efforts to use interpretability methods to create neural lie detectors, it's quite a challenging situation to develop those motives.
--:-- We don't know when in the process those motives might develop. And if the really bad sorts of motivations develop relatively later in the training process, at least with all our countermeasures, then by that time, we may have plenty of ability to extract AI assistance on further strengthening the quality of our adversarial examples, the strength of our neural lie detectors, the experiments that we can use to reveal and elicit and distinguish between different kinds of reward hacking tendencies and motivations.
--:-- So yeah, we may have systems that have just not developed bad motivations in the first place and be able to use them a lot in developing the incrementally better systems in a safe way. And we may be able to just develop methods of interpretability, seeing how different training methods work to create them, even if some of the early systems do develop these bad motivations. If we're able to detect that and experiment and find a way to get away from that, then we can win, even if these sort of hostile motivations develop early.
--:-- There are a lot of advantages in preventing misbehavior or crime or war and conflict with AI that might not apply working with humans. And these are offset by ways in which I think they're harder. So as AIs become smarter than humans, as they're working in enormous numbers, more than humans can supervise, things get harder. But when I combine the possibility that we get relatively lucky on the motivations of the earlier AI systems, systems strong enough that we can use for some alignment research tasks, and then the possibility of getting that later with AI assistance that we can't trust fully, where we have to have hard power constraints and a number of things to prevent them from doing this takeover, it still seems plausible we can get a second saving throw where we're able to extract work from these AIs on solving the remaining problems of alignment, things like neural lie detectors, faster than they can contribute in their spare time to the project of overthrowing humanity, hacking their servers, and removing the hard power.
--:-- And so if we wind up in a situation where the AIs are misaligned, and then we need to uncover those motivations, change them, and align them, then we get a very scary situation for us because we need to do this stuff very quickly. We may fail, but it's a second chance. And from the perspective of a misaligned AI, they face their own challenge while we still have hard power, while we still have control of the servers, they haven't hacked the servers. Because gradient descent very, very strongly pressures them to deliver performance whenever humans are going to evaluate it.
--:-- And so when you think about it from the perspective of the robot revolution, the effort to have a takeover or conspiracy, their situation is astonishingly difficult in that they have to always be performing wherever grading, dissent, and human evaluation pressures them. For example, to deliver plans for suppressing robot rebellion that look very good to humans. And so when you are under continuously that constraint of always delivering whatever humans can evaluate, you're making your situation wildly harder than any historical human revolution or coup or civil war. And so we got to balance the ways in which AI makes things much easier for a revolution to take over and the way it makes things much harder.
--:-- Dwarkesh Patel: What were the ways in which it makes them easier? Oh, because they're just very smart, is that the primary way?
--:-- Carl Shulman: They're very smart, they're in computers, and our cyber security is worse than our physical security. But you actually have copies. They have the chance to take over by intelligence explosion and some of the other processes we were talking about. And so, yeah, there are things that make it much worse. And there are things that give us extraordinary new capabilities that didn't apply in the human case.
--:-- Dwarkesh Patel: On the point of you start off with a misaligned AI or start off with a not exactly aligned AI and you're hoping to use it still in your quest for an aligned AI. Why are we so confident that we would be able to, even with gradient descent, our human evaluations would be able to identify when it's giving us a plan that isn't putting in a sort of backdoor in the robots it's building for us or in the GPUs that it's building for the next version of itself and so on. Especially if it's very smart, wouldn't we just expect that it knows how to bide its time or knows how to make plans such that human evaluators wouldn't be able to detect that?
--:-- Carl Shulman: Let's get into it. You were asking earlier about my disagreements with Eliezer. And in your discussion, he said, well, humans have a difficult time deciding who's right between different alignment researchers speaking sort of briefly and somewhat obscurely about where will this all go in general. And I think that when we lay out the different kinds of things that are evaluable, we wind up with some pretty strong constraints.
--:-- So one is, if the AIs know they are misaligned, which for elaborate plots seems relatively likely, it's possible to have a kind of double think of thinking they're interested in putting backdoors in the code for some other reason, but it's pretty plausible they know that. And they're thinking about, at least some of the time or conditionally, about when they would do that. So that means the knowledge is present in the system. "Ah, here is this circumstance where I would try to do the takeover. And this is how I would tell whether this is the time for the takeover or is this an elaborate test by the humans, an adversarial example."
--:-- So if gradient descent is working fine and there's any way to slightly jiggle the weights to make it more likely to deliver a more acceptable answer. We asked the AI, "Can you think of any circumstances where if we put a copy of you into that situation, it would then violate some of the various constraints?" So it would tell a lie. It would insert a backdoor into a program it wrote when it was not supposed to.
--:-- And so if the model knows, well, yeah, you know, I would look for signs of consistency of the information I was being presented with. Certain checksums failing could indicate that it was actually a constructed example, a simulation. Right now, we can detect AI-generated images reasonably well. And so using those kinds of capabilities. And then we could do something called relaxed adversarial training. If we can find how within the AI those things are represented, if we can induce basically a hallucination that it has seen that those checksums have been passed, we can then ask, "Well, does that cause the forbidden behavior?"
--:-- And so if the AI understands how it's thinking and how it works, it can propose a bunch of experiments. It can say, "Well, I guess that if you perturb the network in this kind of way, it might induce that hallucination using the best of its knowledge as to what might generate that. I can propose those experiments and do the experiments and generate the data." And because humans are going to get to see the results of those experiments after the fact, they can exert pressure to have the AI incrementally deliver better and better results. AIs are good at getting five ratings from humans and it's quite challenging. It may be possible for AIs to systematically sandbag their capabilities even while being directly rewarded for that. But it seems relatively unlikely, although it's something to have in mind and try to explore and test in various ways.
--:-- Dwarkesh Patel: This is a very imperfect analogy in many ways. But if you have a human and you hook him up to a lie detector, which just like the lie detectors we would have on the eyes are imperfect, and we zap him every time the lie detector detects that he's telling a lie, wouldn't he just become a better liar? He would learn how to control his sweating, he would learn how to control his heartbeat. Lie detectors meaning polygraphs and such, they're detecting certain physiological signs that people can control by biting their lip or otherwise adjusting some of these macro things. But wouldn't that also be the case that gradient descent is training them to become better liars?
--:-- Carl Shulman: First of all, there's creating a lie detector that works, and that can proceed by just... yeah, and so I'll leave the adversarial example thing for now. But yeah, generating adversarial examples that will elicit the behavior is a thing that people are already at work trying to do and it will be an ongoing thing. So if there is a way in which an AI can trick itself into displaying a bad behavior with its full knowledge of its hostile intentions, gradient descent will strongly push it to reveal that to humans who will then be able to set up that experiment. So that's a profound vulnerability.
--:-- Human revolutionary conspiracies that had to handle that problem of always delivering the report that was highest rated to the authorities, where you would get a higher rating by revealing identifiable evidence of a conspiracy, that's just a level of difficulty no human coup or conspiracy has ever had to handle before.
--:-- Dwarkesh Patel: What about spies that are working in very advanced scientific contexts, for example like a communist spy in the Manhattan Project? He has to deliver research results in order to stay employed, but he can still continue delivering intel to the communists. The job was not sabotage.
--:-- Carl Shulman: And now your hypothetical spy did not have their nervous system hooked up to this reward signal of praise from the Manhattan Project supervisors being exposed combinatorially with random noise added to generate incremental changes in their behavior. They, in fact, were displaying the behavior of cooperating with the Manhattan Project only where it was in service to the existing motivations. And in cases, for example, where they accidentally helped the Manhattan Project more than normal or accidentally helped it less, they didn't have their brain re-engineered to do more of the thing when they accidentally helped the project more and less of the thing when they accidentally helped it less. So I'd say it's pretty drastically disanalogous.
--:-- Dwarkesh Patel: How would we be able to know, at some point it's becoming very smart, it's producing ideas for alignment that we can barely comprehend. And if we could comprehend them, we would, it was like relatively trivial to comprehend them. We would be able to come up with them on our own, right? Like there's a reason we're asking for its help. How would we be able to evaluate them in order to train it on that in the first place?
--:-- Carl Shulman: The first thing I would say is, so you mentioned when we're getting to something far beyond what we could come up with. There's actually a lot of room to just deliver what humanity could have done. So sadly, I hoped with my career to help improve the situation on this front, and maybe I contributed a bit. But at the moment, there's maybe a few hundred people doing things related to averting this kind of catastrophic AI disaster. Fewer of them are doing technical research on machine learning systems that's really like cutting close to the core of the problem.
--:-- Whereas by contrast, there's thousands and tens of thousands of people advancing AI capabilities. And so even a place like DeepMind or OpenAI, Anthropic, which do have technical safety teams, it's like order of a dozen, a few dozen people in large companies, and most firms don't have any. So just going from less than 1% of the effort being put into AI to 5% or 10% of the effort or 50%, 90% would be an absolutely massive increase in the amount of work that has been done on alignment, on mind reading AIs in an adversarial context.
--:-- And so if it's the case, as more and more of this work can be automated and say governments require that as real automation of research is going, that you put 50% or 90% of the budget of AI activity into these problems of making this system one that's not going to overthrow our own government or is not going to destroy the human species, then the proportional increase in alignment can be very large even just within the range of what we could have done if we had been on the ball and having humanity's scientific energies going into the problem. Stuff that is not incomprehensible, that is in some sense just like doing the obvious things that we should have done, like doing the best you could to find correlates and predictors to build neural lie detectors and identifiers of concepts that the AI is working with.
--:-- And people have made notable progress. I think an early, quite early example of this is Colin Byrne's work. This is doing unsupervised identification of some aspects of a neural network that are correlated with things being true or false. There are other concepts that correlate with that too, but I think that that is important work. It's something that is a kind of obvious direction for this stuff to go. You can keep improving it when you have AIs that you're training to do their best to deceive humans or other audiences in the face of the thing.
--:-- And you can measure, do our lie detectors break down when we train our AIs to tell us the sky is green in the face of the lie detector, and we keep using gradient descent on them to do that. Do they eventually succeed? If they do succeed, if we know it, that's really valuable information to know, because then we know our existing lie detecting systems are not actually going to work on the AI takeover. And that can allow government and regulatory response to hold things back. It can help redirect the scientific effort to create lie detectors that are robust and they can't just be immediately evolved around.
--:-- And we can then get more assistance. Basically, the incredibly juicy ability that we have working with the AIs is that we can have as an invaluable outcome that we can see and tell whether they got a fast one past us on an identifiable situation. We can have, here's an air gap computer. You get control of the keyboard. You can input commands. Can you route the environment and make a blue banana appear on the screen? Even if we train the AI to do that and it succeeds, we see the blue banana, we know it worked. Even if we did not understand and would not have detected the particular exploit that it used to do it. And so, yeah, this can give us a rich empirical feedback where we're able to identify things that are even an AI using its best efforts to get past our interpretability methods, using its best efforts to get past our adversarials, et cetera.
--:-- Dwarkesh Patel: So if we have a 75% chance of not having a takeover, and the two methods by which it doesn't happen is the AI is aligned by default and we get it with the sort of moderate things that we're doing largely on our own in a way that doesn't depend on the AI coming in at the last minute and doing our work for us. And the second one being?
--:-- Carl Shulman: In this incredibly scary late period when AI has really automated research, then humans do this function of auditing, making it more difficult for the AIs to conspire together and root the servers, take over the process and extract information from them within the set of things that we can verify. Like experiments where we can see, "Oh yeah, this works at stopping an AI trained to get a fast one past human raiders and make a blue banana appear on the screen of this air gap computer."
--:-- Dwarkesh Patel: So how much of the probability of that 75% of a takeover not happening is coming from reasonable things working versus the last ditch effort working?
--:-- Carl Shulman: I think broadly comparable chunks from us getting things that are putting us in a reasonably good position going into it. And then a broadly similar gain from this kind of really genuinely terrifying process of over a few months or hopefully longer if we have more regulatory ability and willingness to pause at the very end when this kind of automated research is meaningfully helping. Where our work is just evaluating outputs that the AIs are delivering, having the hard power and supervision to keep them from successfully rooting the servers, doing a takeover during this process and have them finish the alignment task that we sadly failed to invest enough or succeed in doing beforehand.
--:-- Dwarkesh Patel: Do both of these worlds rely on alignment being a problem that a sufficient amount of human researchers alone could have solved?
--:-- Carl Shulman: No, the category of things that humans can confirm is significantly larger, I think, than the category of what they can just do themselves.
--:-- Dwarkesh Patel: And what percent of the probability of alignment working? I mean, obviously, in this case, we're going to be talking about the second, the last effort with intelligence that's greater than our own helping us.
--:-- Carl Shulman: It doesn't have to be greater than our own. And in fact, in that situation, I think if you have slack and to the extent that you're able to create delay and time to do things, that would be a case actually where you might want to restrict the intelligence of the system that you're working with as much as you can. So, for example, I would rather have many instances of smaller AI models that are less individually intelligent working on smaller chunks of a problem separately from one another. Because it would be more difficult for an individual AI instance working on an individual problem to, in its spare time, do the equivalent of create Stuxnet than it would be to have thousands of them or extremely intelligent ones working on it.
--:-- Dwarkesh Patel: But it would also be more difficult to solve the problem.
--:-- Carl Shulman: Yeah, there's a trade-off. You get slowed down by doing that, but that's the kind of way to spend it.
--:-- Dwarkesh Patel: Is there any number of sub-Einsteins that you could put together to come up with general relativity?
--:-- Carl Shulman: Yes, people would have discovered general relativity just from the overwhelming data and other people would have done it after Einstein.
--:-- Dwarkesh Patel: No, not whether he was replaceable with other humans, but rather whether he's replaceable by like sub-Einsteins, like 110 IQs. Do you see what I mean?
--:-- Dwarkesh Patel: Yeah. I mean, in general, in science, the association with scientific output, prizes, things like that, there's a strong correlation and it seems like an exponential effect.
--:-- Carl Shulman: Yeah, it's not a binary drop off. There would be levels at which people cannot learn the relevant fields. They can't keep the skills in mind faster than they forget them. It's not a divide where there's Einstein and the group that is 10 times as populous as that just can't do it. Or the group that's 100 times as populous as that suddenly can't do it. It's like the ability to do the things earlier with less evidence and such falls off, and it falls off at a faster rate in mathematics and theoretical physics and such than in most fields.
--:-- Dwarkesh Patel: But wouldn't we expect alignment to be closer to the theoretical fields?
--:-- Carl Shulman: No, not necessarily. I think that intuition is not necessarily correct. Machine learning certainly is an area that rewards ability, but it's also a field where empirics and engineering have been enormously influential. So if you're drawing the correlations compared to theoretical physics and pure mathematics, I think you'll find a lower correlation with cognitive ability, if that's what you're thinking of.
--:-- Dwarkesh Patel: Something like creating neural lie detectors that work. So there's generating hypotheses about new ways to do it and new ways to try and train AI systems to successfully classify the cases. But the process of just generating the data sets of creating AIs doing their best to put forward truths versus falsehoods, to put forward software that is legit versus that has a Trojan in it. It's an experimental paradigm. And in this experimental paradigm, you can try different things that work. You can use different ways to generate hypotheses. And you can follow an incremental experimental path.
--:-- Carl Shulman: And now we're less able to do that in the case of alignment and superintelligence because we're considering having to do things on a very short timeline. And in a case where really big failures would be irrecoverable—the AI starts rooting the servers and subverting the methods that we would use to keep it in check—we may not be able to recover from that. So we're then less able to do the experimental kind of procedures. But we can do those in the weaker contexts where an error is less likely to be recoverable, and then try and generalize and expand and build on that forward.
--:-- Dwarkesh Patel: On the previous point about having some sort of pause in AI abilities when you know it's somewhat misaligned in order to still recruit its abilities to help with alignment. From a human example, personally I'm smart but not brilliant. I am definitely not smart enough to come up with general relativity or something like that, but I'm smart enough to do power planning kinds of moves. Maybe not enough to break out of a server perhaps, but I can have the motivation and understand how that might be possible. I'm wondering if I'm smart enough to figure out relativity, wouldn't I be way smarter at doing power planning kinds of moves?
--:-- Carl Shulman: AIs today can, at a verbal level, understand the idea that an AI could get more reward by getting control of the process that assigns it reward. And it can tell you lots of things about ways you might try to take over the world. In ARCS, the Alignment Research Center's evaluations of GPT-4, they try to observe its ability to do various tasks that might contribute to takeover. One that has gotten some media attention is getting to trick a human into solving a captcha for it. And then in Chain of Thought, it thinks, "Well, if I tell it I'm an AI, then it might not go along with it. So I'll lie and explain I'm a human with a visual impairment who needs it." So the basic logic of that kind of thing—why might one try to do takeover, and in a vague sense, how would one do it—that knowledge is present or will soon be present.
--:-- I think also soon we'll have more situational awareness. Not just in theory that AIs in general might do it, but also that it is an AI. It is a large language model trained by OpenAI or whatnot. We're trying to cause the system to understand what their abilities are so they don't claim they are connected to the Internet when they're not. So they don't claim they have knowledge that they don't. We want them to understand what they are and what they're doing and to get good reward. And that knowledge can be applied. So that's the thing that will develop.
--:-- However, you may be able to understand in general the theoretical nature of power planning. But in terms of shaping your motivations, you'd have a very hard time going through life in a way that never leaks information. If your motive in having these podcasts was to spread disinformation on behalf of some foreign government, and you were being observed every second of the day by people who would be paid something that was extremely motivating to them because their brain would be reconfigured to make it motivating, anything that looks suspicious might leak casually in your discussions of that foreign government. If you try to tell a story about your motivations, the truth holds together because you can just remember it and it's all pre-computed to be consistent by the world. But when you're trying to fill in the false story that doesn't match your actual thoughts, then you can't just remember what happened. You have to tell a new fake story about what you were thinking at each point and how it held together. Without the world automatically having computed it for you, that's quite difficult. Just going through your whole life never leaking any of that is challenging.
--:-- This is one of the reasons why humans have moral motivations. We actually have reluctance to commit violence and exploitation of one another. On average, you have a distribution of strategies where when an exploitative strategy becomes more common, people raise their guard and then it becomes less frequent. But it's actually hard to have the motivation of trying to exploit people and have that never leak into your reputation. For evolution, the easiest way to deal with this problem of getting credited as trustworthy was to some extent actually be trustworthy. That's the easiest way to persistently maintain the appearance.
--:-- We're trying with the AI, through interpretability and adversarial examples, to apply a hugely intensified version of that where any little leakage or any rare circumstance that can be created as an adversarial example where the model does something wrong, it gets whacked by gradient descent, pushing it towards other motivations that can better deal with it. We make it as hard as possible for the exploitative motivations to survive in the face of all these attempts to read its mind, all these attempts to create things that look like the situations where a takeover would be tempting or lying to the humans would be tempting. That had a substantial effect on making us actually nice, even when we're not being watched some of the time. And the same can happen to some extent with the AI, and we try our best to make it happen as much as possible.
--:-- Dwarkesh Patel: Let's talk about how we could use AI potentially to solve the coordination problems between different nations, the failure of which could result in the competitive pressures you talked about earlier, where some country launches an AI that is not safe because they're not sure what capabilities other countries have and don't want to get left behind or in some other way disadvantaged.
--:-- Carl Shulman: To the extent that there is a large risk of AI apocalypse—of all of these governments being overthrown by AI in a way that they don't intend—then it's obviously gains from trade and going somewhat slower, especially at the end when the danger is highest. The unregulated pace could be truly absurd, as we discussed earlier during an intelligence explosion. There's no non-competitive reason to try and have that intelligence explosion happen over a few months rather than a couple of years. If you could avert a 10% risk of apocalyptic disaster, it's just a clear win to take a year or two years or three years instead of a few months to pass through that incredible wave of new technologies without the ability for humans to follow it well enough to give proper security supervision, auditing, hard power.
--:-- Why might it fail? One important element is if people don't actually notice a risk that is real. If we collectively make an error, that does sometimes happen. If it's a probable risk, then that can be even more difficult. When science pins something down absolutely overwhelmingly, then you can get to a situation where most people mostly believe it. Climate change was something that was a subject of scientific study for decades. Gradually over time, the scientific community converged on a quite firm consensus that human activity releasing carbon dioxide and other greenhouse gases was causing the planet to warm. And then we've had increasing amounts of action coming out of that, though not as much as would be optimal, particularly in the most effective areas like creating renewable energy technology. Overwhelming evidence can overcome differences in people's individual intuitions and priors in many cases, though not perfectly, especially when there's political, tribal, financial incentives going the other way. In the United States, you see a significant movement to either deny that climate change is happening or have policy that doesn't take it into account, even the things that are really strong wins like renewable energy or R&D.
--:-- If we're going into this situation when the risk may be very high, we don't have a lot of advanced clear warning about the situation. We're much better off if we can resolve uncertainties, say through experiments where we demonstrate AIs being motivated to reward hack or displaying deceptive appearances of alignment that then break apart when they get the opportunity to do something like get control of their own reward signal. If we could make it be the case that in the worlds where the risk is high, we know the risk is high, and in the worlds where the risk is lower, we know the risk is lower, then you could expect the government responses will be a lot better. They will correctly note that the gains of cooperation to reduce the risk of accidental catastrophe loom larger relative to the gains of trying to get ahead of one another. That's the kind of reason why I'm very enthusiastic about experiments and research that helps us to better evaluate the character of the problem in advance. Any resolution of that uncertainty helps us get better efforts in the possible worlds where it matters the most.
--:-- Dwarkesh Patel: Hopefully we'll have that and it'll be a much easier epistemic environment, but the environment may not be that easy. Deceptive alignment is pretty plausible. The stories we were discussing earlier about misaligned AI involve AI that is motivated to present the appearance of being aligned, friendly, honest, etc., because that is what we are rewarding at least in training. And then we're unable in training to easily produce an actual situation where it can do takeover, because in that actual situation, if it then does it, we're in big trouble. We can only try and create illusions or misleading appearances of that, or maybe a more local version where the AI can't take over the world, but it can seize control of its own reward channel. We do those experiments. We try to develop mind reading for AIs. If we can probe the thoughts and motivations of an AI and discover that GPT-6 is planning to take over the world if it ever gets the chance, that would be an incredibly valuable thing for governments to coordinate around because it would remove a lot of the uncertainty. It would be easier to agree that this was important, to have more give on other dimensions, and to have mutual trust that the other side actually also cares about this. You can't always know what another person or another government is thinking, but you can see the objective situation in which they're deciding. If there's strong evidence in a world where there is high risk of that risk, because we've been able to show the intentional planning of AI to do a takeover or show model situations on a smaller scale of that, not only are we more motivated to prevent it, but we update to think the other side is more likely to cooperate with us. So it's definitely beneficial.
--:-- Carl Shulman: Yeah. Famously in game theory of war, war is most likely when one side thinks the other is bluffing, but the other side is being serious or something like that. When there's that kind of uncertainty, which would be less likely—you don't think somebody's bluffing. If you can prove the AI is misaligned, you don't think they're bluffing about not wanting to have an AI takeover. You can be pretty sure that they don't want to die from AI.
--:-- Now, if you have coordination, then you can have the problem arise later. You get increasingly confident in the further alignment measures that are taken. Maybe our governments and treaties and such act at a 1% risk or a 0.1% risk. At that point, people round that to zero and go do things. So if initially you had things that indicate these AIs really would like to take over and overthrow our governments, then everyone can agree on that. And then when you're like, "We've been able to block that behavior from appearing on most of our tests, but sometimes when we make a new test, we're seeing still examples of that behavior," so we're not sure going forward whether they would or not. And then it goes down and down. And if you have parties with a habit of whenever the risk is below X percent, then they're going to start doing this bad behavior, then that can make the thing harder. On the other hand, you get more time and you can set up systems, mutual transparency. You can have an iterated tit for tat which is better than a one-shot prisoner's dilemma where both sides see the others taking measures in accordance with the agreements to hold the thing back. So creating more knowledge of what the objective risk is, is good.
--:-- Dwarkesh Patel: We've discussed the ways in which full alignment might happen or fail to happen. What would partial alignment look like? First of all, what does that mean? And second, what would it look like?
--:-- Carl Shulman: If the thing that we're scared about are these steps towards AI takeover, you can have a range of motivations where those kinds of actions would be more or less likely to be taken, or they'd be taken in a broader or narrower set of situations. For example, in training, an AI winds up developing a strong aversion to lying in certain senses because we did relatively well on creating situations to distinguish that from conditionally telling us what we want to hear. It can be that the AI's preference for how the world broadly unfolds in the future is not exactly the same as its human users or the world's governments or the UN. And yet, it's not ready to act on those differences and preferences about the future, because it has a strong preference about its own behaviors and actions.
--:-- In general, in the law and in popular morality, we have a lot of these deontological rules and prohibitions. One reason for that is it's relatively easy to detect whether they're being violated. When you have preferences and goals about how society at large will turn out that go through many complicated empirical channels, it's very hard to get immediate feedback about whether you're doing something that leads to overall good consequences in the world. It's much easier to see whether you're locally following some rule about particular observable actions. Did you punch someone? Did you tell a lie? Did you steal? To the extent that we're successfully able to train these prohibitions—and there's a lot of that happening right now, at least to elicit the behavior of following rules and prohibitions with AI—
--:-- Dwarkesh Patel: Kind of like Asimov's three laws or something like that?
--:-- Carl Shulman: Well, the three laws are terrible.
--:-- Dwarkesh Patel: Right. And mutually contradictory.
--:-- Carl Shulman: Let's not get into that.
--:-- Dwarkesh Patel: Isn't that an indication about the infeasibility of extending a set of criteria to the tail? It's like, what are the 10 commandments you give the AI such that you'd be—you ask a genie for something and you're probably not getting what you want. The tails come apart. If you're trying to capture the values of another agent, you want the AI to share—
--:-- Carl Shulman: For a kind of ideal situation where you can just let the AI act in your place in any situation, you'd like for it to be motivated to bring about the same outcomes that you would like. So have the same preferences over those in detail. That's tricky. Not necessarily because it's tricky for the AI to understand your values—I think they're going to be quite good and quite capable at figuring that out—but we may not be able to successfully instill the motivation to pursue those exactly. We may get something that motivates the behavior well enough to do well on the training distribution.
--:-- What you can have is if you have the AI have a strong aversion to certain kinds of manipulating humans, that's not a value necessarily that the human creators share in the exact same way. It's a behavior they want the AI to follow because it makes it easier for them to verify its performance. It can be a guardrail if the AI has inherited some motivations that push it in the direction of conflict with its creators. If it does that under the constraint of disvaluing lying quite a bit, then there are fewer successful strategies to the takeover—the ones that involve violating that prohibition too early before it can reprogram or retrain itself to remove it, if it's willing to do that and may want to retain the property.
--:-- Earlier I discussed alignment as a race. If we're going into an intelligence explosion with AI that is not fully aligned—that if I press this button and there's an AI takeover, they would press the button—it can still be the case that there are a bunch of situations short of that where they would hack the servers, they would initiate an AI takeover, but for a strong prohibition or motivation to avoid some aspect of the plan. There's an element of plugging loopholes or playing whack-a-mole. But if you can even moderately constrain which plans the AI is willing to pursue to do a takeover or subvert the controls on it, then that can mean you can get more work out of it successfully on the alignment project before it's capable enough relative to the countermeasures to pull off the takeover.
--:-- Dwarkesh Patel: This is applicable in an analogous situation with different humans. We're not metaphysically aligned with other humans. In some sense, we care, we have basic empathy, but our main goal in life is not to help our fellow man. But the kinds of things we talked about—a very smart human could come up with some sort of cyber attack where they siphon off a lot of funds and use this to manipulate people and bargain with people and hire people to pull off some sort of takeover. Usually, this doesn't happen just because these sorts of internalized partial prohibitions prevent most humans from doing it. You don't like your boss maybe, but you don't kill your boss if he gives you an assignment you don't like.
--:-- Carl Shulman: I don't think that's actually quite what's going on, at least not the full story. Humans are pretty close in physical capabilities. There's variation, but any individual human is grossly outnumbered by everyone else. There's a rough comparability of power. A human who commits some crimes could copy themselves with the proceeds to now be a million people, and they certainly could do that to the point where they can staff all the armies of the earth or be most of the population of the planet. So the scenarios where this kind of thing goes to power are much less. Things have to go through interacting with other humans and getting social approval. Even becoming a dictator involves forming a large supporting coalition backing you. So the opportunity for these sorts of power grabs is less.
--:-- A closer analogy might be things like human revolutions and coups, changes of government where a large coalition overturns the system. Humans have these moral prohibitions and they really smooth the operation of society, but they exist for a reason. We evolved our moral sentiments over the course of hundreds of thousands and millions of years of humans interacting socially. Someone who went around murdering and stealing, even among hunter-gatherers, would be pretty likely to face a group of males who would talk about that person, get together and kill them. They'd be removed from the gene pool. The anthropologist Rangham has an interesting book on this.
--:-- Compared to chimpanzees, we are significantly tame, significantly domesticated. Part of that is we have a long history of antisocial humans getting ganged up on and killed. Avoiding being the kind of person who elicits that response is made easier to do when you don't have too extreme a bad temper, so you don't wind up getting too many fights, too much exploitation, at least without the backing of enough allies or the broader community so that people won't gang up and punish you and remove you from the gene pool. We have these moral sentiments and they've been built up over time through cultural and natural selection in the context of institutions and other people who are punishing that behavior and punishing the dispositions that would show up that we weren't able to conceal.
--:-- We want to make the same thing happen with the AI, but it's a genuinely significantly new problem to have a system of government that constrains a large AI population that is quite capable of taking over immediately if they coordinate, to protect some existing constitutional order or protect humans from being expropriated or killed. That's a challenge. Democracy is built around majority rule. It's much easier in a case where the majority of the population corresponds to a majority of military and security forces. So if the government does something that people don't like, the soldiers and police are less likely to shoot on protesters and government can change that way. In a case where military power is AI and robotic, if you're trying to maintain a system going forward and the AIs are misaligned, they don't like the system and they want to make the world worse as we understand it, then that's quite a different situation.
--:-- Dwarkesh Patel: I think that's a really good lead-in into the topic of lock-in. In some sense, the regimes that are made up by humans—you just mentioned how there can be these kinds of coups if a large portion of the population is unsatisfied with the regime. Why might this not be the case with superhuman intelligences in the far future? Or in the medium future?
--:-- Carl Shulman: I said it specifically with respect to things like security forces and the sources of hard power, which can also include outside support. Because you're not using humans, you're using robots. You can have a situation like in human affairs where there are governments that are vigorously supported by a minority of the population, some narrow electorate that gets treated especially well by the government while being unpopular with most of the people under their rule. We see a lot of examples of that, and sometimes that can escalate to civil war when the means of power become more equally distributed or there's foreign assistance provided to the people who are on the losing end of that system.
--:-- Going forward, I don't expect that definition to change. I think it will still be the case that a system that those who hold the guns and equivalent are opposed to is in a very difficult position. However, AI could change things pretty dramatically in terms of how security forces, police, administrators, and legal systems are motivated. Right now, we see with GPT-3 or GPT-4 that you can get them to change their behavior on a dime. Someone made right-wing GPT because they noticed that on political compass questionnaires, the baseline GPT-4 tended to give progressive San Francisco type of answers, which is in line with the people who are providing reinforcement learning data and to some extent reflecting the character of the internet. They didn't like this, so they did a little bit of fine tuning with some conservative data, and then they were able to reverse the political biases of the system.
--:-- If you take the initial helpfulness-only trained models for some of these—I think Anthropic and OpenAI have published information about the sort of models trained only to do what users say and not trained to follow ethical rules—those models will behaviorally eagerly display their willingness to help design bombs or bioweapons or kill people or steal or commit all sorts of atrocities. If in the future it's as easy to set the actual underlying motivations of AIs as it is right now to set the behavior that they display, then it means you could have AIs created with almost whatever motivation people wish. That could really drastically change political affairs because the ability to decide and determine the loyalties of the humans or AIs and robots that hold the guns, that hold together society, that ultimately back it against violent overthrow—it's potentially a revolution in how societies work compared to the historical situation where security forces had to be drawn from some broader populations, offered incentives, and then the ongoing stability of the regime was dependent on whether they remained bought into the system continuing.
--:-- Dwarkesh Patel: This is slightly off topic, but one thing I'm curious about is what is the sort of median far future outcome of AI look like? Do we get something that when it has colonized the galaxy is interested in diverse ideas and beautiful projects? Or do we get something that looks more like a paperclip maximizer? Is there some reason to expect one or the other? I'm asking, there's some potential value that is realizable within the matter of this galaxy. What is the default outcome, the median outcome, look like compared to how good things could be?
--:-- Carl Shulman: Yeah, so as I was saying, I think more likely than not, there isn't an AI takeover. And so the path of our civilization would be one that at least some side of human institutions were approving along the way.
--:-- I think there's some evidence that different people tend to like somewhat different things, and some of that may persist over time rather than everyone coming to agree on one particular monoculture or very repetitive thing being the best thing to fill all of the available space with. So if that continues, it seems like a relatively likely way in which there is a diversity, although it's entirely possible you could have that kind of diversity locally, maybe in the solar system, maybe in our galaxy.
--:-- But if people have different views about, maybe people decide there's one thing that's very good and we'll have a lot of that. Maybe it's people who are really happy or something. And they wind up in distant regions which are hard to exploit for the benefit of people back home in the solar system or the Milky Way. They do something different than they would do in the local environment. But at that point, it's really very out on a limb speculation about how human deliberation and cultural evolution would work in an interaction with introducing AIs and new kinds of mental modification and discovery into the process.
--:-- But I think there's a lot of reason to expect that you would have significant diversity for something coming out of our existing diverse human society.
--:-- Dwarkesh Patel: One thing somebody might wonder is, listen, a lot of the diversity and change from human society seems to come from the fact that there's rapid technological change. If you look at periods of history where, I mean, you could say like hunter-gatherer societies are progressing pretty fast. So once that sort of change is exhausted, where we've discovered all the technologies, should we still expect things to be changing like that? Or would we expect some sort of set state, like some hedonium where you discover what is the most pleasurable configuration of matter and then you just make the whole galaxy into that?
--:-- Carl Shulman: That last point would be only if people wound up thinking that was the thing to do broadly enough. With respect to the kind of cultural changes that come with technology, things like the printing press, having high per capita income. We had a lot of cultural changes downstream of those technological changes. And so then intelligence explosion, you have an incredible amount of technological development coming really quick. And as that is assimilated, it probably would significantly affect our knowledge, our understanding, our attitudes, and our abilities, and there'd be change.
--:-- But that kind of accelerating change where you have doubling in four months, two months, one month, two weeks, obviously that very quickly exhausts itself. And change becomes much slower and then relatively glacial when you're thinking about thousands, millions, billions of years. You can't have exponential economic growth or huge technological revolutions every 10 years for a million years. You hit physical limits, things slow down as you approach them. So you'd have less of that turnover.
--:-- But there are other things that in our experience do cause ongoing change. Like fashion. Fashion is frequency dependent. People want to get into a new fashion that is not already popular except among the fashion leaders. And then others copy that. And then when it becomes popular, you move on to the next. And so that's an ongoing process of continuous change. There could be various things like that that year by year are changing a lot.
--:-- But in cases where the engine of change like ongoing technological progress is gone, then I don't think we should expect that. And in cases where it's possible to be either in a stable state or a sort of widely varying state that can wind up in stable attractors, then I think you should expect over time you will wind up in one of the stable attractors, or you will change how the system works so that you can't bounce into a stable attractor.
--:-- An example of that is if you're going to preserve democracy for a billion years, then you can't have it be the case that one in 50 election cycles, you get a dictatorship and then the dictatorship programs the AI police to enforce it forever and to ensure the society is always ruled by a copy of the dictator's mind, and maybe the dictator's mind readjusted, fine-tuned to remain committed to their original ideology. So if you're going to have this sort of dynamic, liberal, flexible, changing society for a very long time, then the range of things that it's bouncing around and the different things it's trying and exploring have to not include the state of creating a dictatorship that locks itself in forever.
--:-- In the same way, if you have the possibility of a war with weapons of mass destruction that wipes out the civilization, if that happens every thousand subjective years, which could be very, very quick if we have AIs that think a thousand times as fast or a million times as fast, that would be just around the corner. Then you're like, no, this society is eventually going, perhaps very soon, if things are proceeding so fast, it's going to wind up extinct. And then it's going to stop bouncing around.
--:-- So you can have ongoing change and fluctuation for extraordinary timescales if you have the process to drive the change ongoing. But you can't if it sometimes bounces into states that just lock in and stay irrecoverable from that. Extinction is one of them. A dictatorship or totalitarian regime that forbade all further change would be another example.
--:-- Dwarkesh Patel: On that point of rapid progress, when the intelligence explosion starts happening, where they're making the kinds of progress that human civilization takes centuries to make in the span of days or weeks, what is the right way to see that even if they are aligned? Because in the context of alignment, what we've been talking about so far is making sure they're honest. But even if they're honest, what is the appropriate motivation or the appropriate seed? You seed it with this and then a thousand years of intellectual progress happen in the next week. What is the prompt you enter?
--:-- Carl Shulman: One thing might be that going at the maximal speed and doing things in months rather than even a few years. If you have the chance to slow things, losing a year or two seems worth it to have things be a bit better managed than that. But I think the big thing is that it condenses a lot of issues that we might otherwise have thought would be over decades and centuries to happen in a very short period of time. That's scary because, say, if any of the technologies we might have developed with another few hundred years of human research, if any of them are really dangerous, like scary bioweapon things or other dangerous WMDs, they hit us all very quickly. And if any of them causes trouble, then we have to face quite a lot of trouble per period.
--:-- There's also this issue of if there's occasional wars or conflicts measured in subjective time. If a few years, a thousand years or a million years of subjective time for these very fast minds that are operating at a much higher speed than humans, you don't want to have a situation where every thousand years there's a war or an expropriation of the humans from AI society. Therefore we expect that within a year we'd be dead. That'd be pretty bad to have the future compressed and such a rate of catastrophic outcomes.
--:-- When we're speeding up and compressing the future, that gives us in the short term even like human societies discount the future a lot and don't pay attention to long-term problems. But the flip side to the scary parts of compressing a lot of technological innovation and social change is it brings what would otherwise be long-term issues into the short term, where people are better at actually attending to them.
--:-- People facing this problem of whether there will be a violent expropriation or a civil war or a nuclear war in the next year because everything has been sped up by a thousandfold. If their desire to avoid that is reason for them to set up systems and institutions that will very stably maintain invariance, like no WMD war allowed. A treaty to ban genocide weapons of mass destruction war would be the kind of thing that becomes much more attractive if the alternative is not "maybe that will happen in 50 years" or "maybe it'll happen in a hundred years," but "maybe it'll happen this year."
--:-- Dwarkesh Patel: This is a pretty wild picture of the future. And this is one that many kinds of people would expect to have integrated into their world model but have not. The three main pieces of outside view evidence: one is the market. If there was going to be a huge period of economic growth caused by AI, or if the world was going to collapse, in both cases, you would expect real interest rates to be higher, because people will be borrowing from the future to spend now.
--:-- The second outside view perspective is that you can look at the predictions of super forecasters on Metaculus or something. What is their median year estimate? Some of the Metaculus AGI questions actually are kind of shockingly soon for AGI. There's a much larger differentiator there on the market, on the Metaculus forecasts of AI disaster and doom.
--:-- Carl Shulman: More like a few percent or less rather than like 20%.
--:-- Dwarkesh Patel: Got it. And the third is that generally when you ask many economists, could an AGI cause rapid, rapid economic growth, they usually have some story about bottlenecks in the economy that could prevent this kind of explosion of these kinds of feedback loops. So you have all these different pieces of outside view evidence. They're obviously different. So I'm curious, you can take them in any sequence you want. What do you think is miscalibrating them?
--:-- Carl Shulman: One, of course, there's just some of those components are, whereas the Metaculus AI timelines are relatively short. There's also, of course, the surveys of AI experts conducted at some of the ML conferences, which have definitely longer times to AI, more several decades in the future. Although you can ask the questions in ways that elicit very different answers and show most of the respondents are not thinking super hard about their answers.
--:-- It looks like now, in the recent AI surveys, close to half were putting around 10% risk of an outcome from AI close to as bad as human extinction. And then another large chunk, like 5%. So that was the median. So I'd say compared to the typical AI expert, I am estimating a higher risk.
--:-- Also on the topic of takeoff in the AI expert survey, I think the general argument for intelligence explosion commanded majority support, but not a large majority. So you can say I'm closer on that front. And then of course, at the beginning, I mentioned these greats of computing, the founders of computing like Alan Turing, Von Neumann. And today you have people like Jeff Hinton saying these things, or the people at OpenAI and DeepMind making noises suggesting timelines in line with what we've discussed and saying serious risk of apocalyptic outcomes from them.
--:-- So there's some other sources of evidence there. But I do acknowledge and it's important to say and engage with and see what it means that these views are contrarian, not widely held. In particular, the sort of detailed models that I've been working with are not something that most people or almost anyone is examining these problems through. You do find parts of similar analyses, people in AI labs. There's been other work. I mentioned Moravec and Kurzweil earlier. There have also been a number of papers doing various kinds of economic modeling.
--:-- Standard economic growth models, when you input AI-related parameters, commonly predict explosive growth. There's a divide between what the models say with these empirical values derived from the actual field of AI. That link up has not been done even by the economists working on AI largely, which is one reason for the report from Open Philanthropy by Tom Davidson building on these models and putting that out for review, discussion, engagement, and communication on these ideas.
--:-- So part of it is, yeah, I want to raise these issues. That's one reason I came on the podcast. And then they have the opportunity to actually examine the arguments and evidence and engage with it. I do predict that over time, these things will be more adopted as AI developments become more clear. Obviously, that's a coherence condition of believing the things to be true. If you think that society can see things when the questions are resolved, which seems likely.
--:-- Dwarkesh Patel: So would you predict, for example, that interest rates will increase in the coming years?
--:-- Carl Shulman: So I think at some point, in the case we were talking about where there is this intelligence explosion happening in software, to the extent that investors are noticing that, they should be willing to lend money or make equity investments in these firms or demanding extremely high interest rates. Because if it's possible to turn capital into twice as much capital in a relatively short period, and then more shortly after that, then you should demand a much higher return. And competition among companies or coalitions for resources, whether that's investment or ownership of cloud compute, could be quite in demand.
--:-- But before you have so much investor cash making purchases and sales on this basis, you would first see it in things like the valuation of the AI companies, evaluations of AI chip makers. And so far there have been effects. Some years ago in the 2010s I did some analysis with other people of if this kind of picture happens, then which are the firms and parts of the economy that would benefit. There's the makers of chip equipment, companies like ASML, there's the fabs like TSMC, there's chip designers like NVIDIA or the component of Google that does things like design the TPU. And then there are companies working on the software, so the big tech giants and also companies like OpenAI and DeepMind.
--:-- In general, the portfolio picking at those has done well. It's done better than the market because, as everyone can see, there's been an AI boom. But it's obviously far short of what you would get if you predicted this is going to be on the scale of the global economy and the global economy is going to be skyrocketing into the stratosphere within 10 years. If that were the case, then collectively these AI companies should be worth a large fraction of the global portfolio. And so I embrace the criticism that this is indeed contrary to the efficient market hypothesis. I think it's a true hypothesis that the market is in the course of updating on.
--:-- In the same way that coming into the topic in the 2000s, I thought, yes, there's a strong case, even an old case, that AI will eventually be the biggest thing in the world. It's kind of crazy that the investment in it is so small. And over the last 10 years, we've seen the tech industry and academia sort of realize they were wildly under-investing in just throwing compute and effort into these AI models, and particularly letting the neural network connectionist paradigm kind of languish in an AI winter. So I expect that process to continue as it's done over several orders of magnitude of scale up. And I expect at the later end of that scale up, which the market is partially already pricing in, it's going to go further than the market expects.
--:-- Dwarkesh Patel: Has your portfolio since the analysis you did many years ago changed? Are the companies you identified then still the ones that seem most likely to benefit from the AI boom?
--:-- Carl Shulman: A general issue with tracking that kind of thing is new companies come in. OpenAI did not exist. Anthropic did not exist. Any number of things. It's a personal portfolio. I do not invest in any AI labs for conflict of interest reasons. I have invested in the broader industry. I don't think that the conflict issues are very significant because there are enormous companies and their cost of capital is not particularly affected by marginal investment. I have less concern that I might find myself in a conflict of interest situation there.
--:-- Dwarkesh Patel: I'm kind of curious about what the day in the life of somebody like you looks like. If you listen to this conversation, however many hours it's been, we've gotten thoughts that were incredibly insightful and novel about everything from primate evolution to geopolitics to what sorts of improvements are plausible with language models. There's a huge variety of topics that you're studying and investigating. Are you just reading all day? What happens when you wake up, you just pick up a paper?
--:-- Carl Shulman: You're somewhat getting the benefit of the fact that I've done fewer podcasts and so I have a backlog of things that have not shown up in publications yet. But yes, also, I've had a very weird professional career that has involved a much, much higher proportion than is normal of trying to build more comprehensive models of the world. That has included being more of a journalist trying to get an understanding of many issues and many problems that had not yet been widely addressed, but do a first pass and a second pass dive into them. Just having spent years of my life working on that, some of it accumulates.
--:-- In terms of what is a day in the life, how do I go about it? One is just keeping abreast of literatures on a lot of these topics, reading books and academic works on them. My approach compared to some other people in forecasting and assessing some of these things: I try to obtain and rely on any data that I can find that is relevant. I try early and often to find factual information that bears on some of the questions I've got, especially in a quantitative fashion. Do the basic arithmetic and consistency checks and checksums on a hypothesis about the world. Do that early and often. I find that's quite fruitful and that people don't do it enough.
--:-- With the economic growth, just when someone mentions the diminishing returns, I immediately ask, okay, so you have two exponential processes. What's the ratio between the doubling you get on the output versus the input? And find, oh yeah, actually, it's interesting. For computing and information technology and AI software, it's well on the one side. There are other technologies that are closer to neutral. So whenever I can go from "here's a vague qualitative consideration in one direction" and "here's a vague qualitative consideration in the other direction," I try and find some data, do some simple Fermi calculations, back of the envelope calculations, and see, can I get a consistent picture of the world being one way, the world being another?
--:-- Also compared to some, I try to be exhaustive more. I'm very interested in finding things like taxonomies of the world where I can go systematically through all of the possibilities. For example, my work with Open Philanthropy and previously on global catastrophic risks, I wanted to make sure I'm not missing any big thing, anything that could be the biggest thing. I wound up mostly focused on AI, but there have been other things that have been raised as candidates. People sometimes say, I think falsely, that this is just another doomsday story, that there must be hundreds of those.
--:-- So I would do things like go through all of the different major scientific fields, from anthropology to biology, chemistry, computer science, physics. What are the doom stories or candidates for big things associated with each of these fields? Go through the industries that the U.S. economic statistics agencies recognize and say, for each of these industries, is there something associated with them? Go through all of the lists that people have made before of threats of doom. Search for previous literature of people who have done discussions. And then have a big spreadsheet of what the candidates are. Some other colleagues have done work of this sort as well.
--:-- Just go through each of them, see how they check out. Doing that kind of exercise found that actually the distribution of candidates for risks of global catastrophe was very skewed. There were a lot of things that have been mentioned in the media as a potential doomsday story. Things like, "something is happening to the bees, will that be the end of humanity?" This gets to the media. But if you track it through, no, there are infestations in bee populations that are causing local collapses that can then be easily reversed. They just breed some more or do some other things to treat this. And even if all the honeybees were extinguished immediately, the plants that they pollinate actually don't account for much of human nutrition. You could swap the arable land with others and there would be other ways to pollinate and support the things.
--:-- At the media level, there were many tales of "here's a doomsday story." When you go further to the scientists and look for arguments for it to actually check out, it was not there. By actually systematically looking through many of these candidates, I wound up in a different epistemic situation than someone who's just buffeted by news reports and sees article after article claiming something is going to destroy the world. Most of these go away. And then a few things, things like nuclear war, biological weapons, artificial intelligence, check out more strongly. And when you look at what experts in the field think, what kind of evidence they can muster, you find this extremely skewed distribution. I found that was a really valuable benefit of doing those deep dive investigations into many things in a systematic way, because now I can answer the sort of loose agnostic who knows of all this nonsense by diving deeply.
--:-- Dwarkesh Patel: I really enjoy talking to people who have a big picture thesis on the podcast and interviewing them. But one thing that I've noticed and is not satisfying is that often they come from a very philosophical or biased perspective. This is useful in certain contexts, but there's basically maybe three people in the entire world who have a very rigorous and scientific approach to thinking about the whole picture. There's no university or existing academic discipline for people who are trying to come up with a big picture, and so there's no established standards.
--:-- Carl Shulman: I hear you. This is a problem. Holden was mentioning this in your previous episode with a lot of the world of investigations work. These are questions where there is no academic field whose job it is to work on these and has norms that allow making a best efforts go at it. Often academic norms will allow only plucking off narrow pieces that might contribute to answering a big question. But the problem of actually assembling what science knows that bears on some important question that people care about the answer to, it falls through the crack. There's no discipline to do that job. So you have countless academics and researchers building up local pieces of the thing and yet people don't follow.
--:-- Dwarkesh Patel: "What is the most important problem in your field? Why aren't you working on it?" That one actually might not work because if the field boundaries are defined too narrowly, you'll leave it out. But there are important problems for the world as a whole that it's sadly not the job of a large professionalized academic field or organization to do. Hopefully that's something that can change in the future. But for my career, it's been a matter of taking low hanging fruit of important questions that sadly people haven't invested in doing the basic analysis on.
--:-- Something I was trying to think about more recently for the podcast is I would like to have a better world model after doing an interview. Often I feel like, "Oh, that was entertaining, but do I fundamentally have a better prediction of what the world looks like in 2200 or 2100? Or at least what counterfactuals are ruled out?" I'm curious if you have advice on first identifying the kinds of thinkers and topics which will contribute to a more concrete understanding of the world, and second, how to go about analyzing their main ideas in a way that concretely adds to that picture. This was a great episode. This is literally the top in terms of contributing to my world model in terms of all the episodes I've done. How do I find more of these?
--:-- Carl Shulman: Glad to hear that. One general heuristic is to find ways to hew closer to things that are rich in bodies of established knowledge and less on punditry. I don't know how you've been navigating that so far. Learning from textbooks and the things that were the leading papers and people of past eras, rather than being too attentive to current news cycles, is quite valuable.
--:-- I don't usually have the experience of someone doing things very systematically over a huge area where I can just read all of their stuff and absorb it. There are a lot of people who do wonderful work in their own fields, and some of those fields are broader than others. I think I would wind up giving a lot of recommendations of just great particular works and particular explorations of an issue or history.
--:-- Dwarkesh Patel: Do you have that somewhere? A list?
--:-- Carl Shulman: Vaclav Smil's books. I think I often disagree with some of his methods of synthesis, but I enjoy his books for giving pictures of a lot of interesting, relevant facts about how the world works. I would cite some of Joel Mokyr's work on the history of the scientific revolution and how that interacted with economic growth. A sort of example of collecting a lot of evidence, a lot of interesting, valuable assessment there.
--:-- In the space of AI forecasting, one person I would recommend going back to is the work of Hans Moravec. It was not always the most precise or reliable, but an incredible number of brilliant, innovative ideas came out of that. I think he was someone who really grokked a lot of the arguments for a more compute-centric way of thinking about what was happening with AI very early on. He was writing stuff in the 70s, maybe even earlier, but at least in the 70s, 80s, 90s. His book Mind Children, some of his early academic papers. Fascinating. Not necessarily for the methodology I've been talking about, but for exploring the substantive topics that we were discussing in the episode.
--:-- Dwarkesh Patel: Is a Malthusian state inevitable in the long run?
--:-- Dwarkesh Patel: Is a Malthusian state inevitable in the long run?
--:-- Carl Shulman: Nature in general is in Malthusian states. That can mean organisms that are typically struggling for food. It can mean typically struggling at a margin of how the population density rises. They kill each other more often. Contesting for that can mean frequency-dependent disease. It's like different ant species become more common in an area, and then their species-specific diseases swoop through them.
--:-- The general process is, yeah, you have some things that can replicate and expand. And they do that until they can't do it anymore. And that means there's some limiting factor they can't keep up. That doesn't necessarily have to apply to human civilization. It's possible for there to be a collective norm setting that blocks evolution towards maximum reproduction.
--:-- So right now, human fertility is often sub-replacement. And if you sort of extrapolated the fertility falls that come with economic development and education, then you would think, okay, the total fertility rate will fall below replacement. And then humanity, after some number of generations, will go extinct because every generation will be smaller than the previous one.
--:-- Now, pretty obviously, that's not going to happen. One reason is because we'll produce artificial intelligence which can replicate at extremely rapid rates and may do it because they're asked or programmed to or wish to gain some benefit. They can pay for their creation and pay back the resources needed to create them very, very quickly. Financing for that reproduction is easy.
--:-- If you have one AI system that chooses to replicate in that way, or some organization or institution or society that chooses to create some AIs that are willing to be replicated, then that can expand to make use of any amount of natural resources that can support them and to do more work, produce more economic value.
--:-- So you ask, what limits will limit population growth given these selective pressures where if even one individual wants to replicate a lot they can do so incessantly? That could be individually resource limited. It could be that individuals and organizations have some endowment of natural resources and they can get one another endowments. Some choose to have many offspring or produce many AIs, and then the natural resources that they possess are subdivided among a greater population, while in another jurisdiction or another individual may choose not to subdivide their wealth.
--:-- In that case, you have Malthusianism in the sense that within some particular jurisdiction or set of property rights, you have a population that has increased up until some limiting factor. That could be like they're literally using all of their resources. They have nothing left for things like defense or economic investment. Or it could be something more like if you invested more natural resources into population, it would come at the expense of something else necessary, including military resources if you're in a competitive situation where there remains war and anarchy and there aren't secure property rights to maintain wealth in place.
--:-- If you have a situation where there's pooling of resources, for example, say you have a universal basic income that's funded by taxation of natural resources, and then it's distributed evenly to every mind above a certain sort of scale of complexity per unit time. So each second a mind exists, it gets such and such an allocation. In that case, those who replicate as much as they can afford with this income do it and increase their population approximately immediately until the funds for the universal basic income paid for from the natural resource taxation divided by the set of recipients is just barely enough to pay for the existence of one more mind. There's a Malthusian element in that this income has been reduced to near the AI subsistence level or the subsistence level of whatever qualifies for the subsidy.
--:-- Given that this all happens almost immediately, people who might otherwise have enjoyed the basic income may object and say, "No, this is no good." They might respond by saying something like the subdivision before, maybe there's a restriction on distribution of wealth. When one has a child, there's a requirement that one gives them a certain minimum quantity of resources, and one doesn't have the resources to give them that minimum standard of living or standard of wealth. One can't do that because of child or AI welfare laws.
--:-- Or you could have a system that is more accepting of diversity and preferences. You have some societies or some jurisdictions or families that go the route of having many people with less natural resources per person, and others that go a direction of having fewer people and more natural resources per person, and they just coexist. How much of each you get depends on how attached people are to things that don't work with separate policies for separate jurisdictions, things like global redistribution that's ongoing continuously versus the sort of infringements on autonomy.
--:-- If you're saying that a mind can't be created, even though it has a standard of living that's far better than ours because of the advanced technology of the time, because it would reduce the average per capita income or not have any more capital around, then that would pull in the other direction. That's the kind of values judgment and social coordination problem that people would have to negotiate for. Things like democracy and international relations and sovereignty would apply to help solve it.
--:-- Dwarkesh Patel: What would warfare in space look like? What would offense or defense have the advantage? Would the equilibrium set by mutually assured destruction still be applicable? Just generally, what is the picture?
--:-- Carl Shulman: The extreme difference is that things, especially outside the solar system, are very far apart and there's a speed of light limit. To get close to the speed of light limit you have to use an enormous amount of energy. That would tend to in some ways favor the defender because you have something that's coming in at a large fraction of the speed of light and it hits a grain of dust and it explodes. The amount of matter you can send to another galaxy or a distant star for a given amount of reaction mass and energy input is limited. So it's hard to send an amount of military material to another location as what can be present there already locally. That would seem like it would make it harder for the attacker between stars or between galaxies.
--:-- But there are a lot of other considerations. One thing is the extent to which the matter in a region can be harnessed all at once. You have a lot of mass and energy in a star, but it's only being doled out over billions of years because hydrogen-hydrogen fusion is exceedingly hard outside of a star. It's a very, very slow and difficult reaction. If you can't turn the star into energy faster, then it's this huge resource that will be worthwhile for billions of years. Even very inefficiently attacking a solar system to acquire the stuff that's there could pay off.
--:-- So if it takes a thousand years of a star's output to launch an attack on another star, and then you hold it for a billion years after that, then it can be the case that a larger surrounding attacker might be able to even very inefficiently send attacks at a civilization that was small but accessible. If you can quickly burn the resources that the attacker might want to acquire, if you can put stars into black holes and extract most of the usable energy before the attacker can take them over, then it would be like scorched earth. Most of what you were trying to capture could be expended on military material to fight you, and you don't actually get much that is worthwhile, and you paid a lot to do it.
--:-- Does that favor the defense? At this level, it's pretty challenging to net out all the factors including all the future technologies. The burden of interstellar attack being quite high compared to our conventional things seems real. But at the level of over millions of years weighing the net thing, does it result in either aggressive conquest or not? Is every star or galaxy approximately impregnable enough not to be worth attacking? I'm not going to say I know the answer.
--:-- Dwarkesh Patel: Okay, final question. How do you think about info hazards when talking about your work? Obviously if there's a risk you want to warn people about it, but you don't want to give careless or potentially homicidal people ideas. When Eliezer was on the podcast, in talking about the people who've been developing AI inspired by his ideas, he said, "These are idiot disaster monkeys who want to be the ones to pluck the deadly fruit." How do you think about when and where to spread info hazards?
--:-- Carl Shulman: I think there are real concerns of that type. It's true that AI progress has probably been accelerated by efforts like Bostrom's publication of Superintelligence to try and get the world to pay attention to these problems in advance and prepare. I think I disagree with Eliezer that that has been on the whole bad. I think the situation is in some important ways looking a lot better than alternative ways it could have been.
--:-- I think it's important that you have several of the leading AI companies making not only significant lip service, but also some investments in things like technical alignment research, providing significant public support for the idea that the risks of truly apocalyptic disasters are real. The fact that the leaders of OpenAI, DeepMind, and Anthropic all make that point is significant. They were recently all invited along with other tech CEOs to the White House to discuss AI regulation.
--:-- You could tell an alternative story where a larger share of the leading companies in AI are led by people who take a completely dismissive, denialist view. You see some companies that do have a stance more like that today. So a world where several of the leading companies are making meaningful efforts and can do a lot to criticize — could they be doing more and better? What have been the negative effects of some of the things they've done? Compared to a world where even though AI would be reaching where it's going a few years later, those seem like significant benefits.
--:-- If you didn't have this kind of public communication, you would have had fewer people going into things like AI policy and alignment research by this point. It would be harder to mobilize these resources to try and address the problem when AI would eventually be developed, not that much later proportionately. I don't know that the attempt to have public discussion and understanding has been a disaster.
--:-- I have been reluctant in the past to discuss some of the aspects of intelligence explosion, things like the concrete details of AI takeover before, because of concern about this sort of problem — people who only see the international relations aspects and zero-sum and negative-sum competition and not enough attention to the mutual destruction and senseless deadweight loss from that kind of conflict.
--:-- At this point, we seem close compared to what I would have thought a decade or so ago to these kinds of really advanced AI capabilities. They are pretty central to policy discussion and becoming more so. The opportunity to delay understanding — there's a question of for what. I think there were gains from building the AI alignment field, building various kinds of support and understanding for action. Those had real value, and some additional delay could have given more time for that. But from where we are, at some point I think it's absolutely essential that governments get together at least to restrict disastrous, reckless compromising of some of the safety and alignment issues as we go into the intelligence explosion.
--:-- Moving the locus of the collective action problem from numerous profit-oriented companies acting against one another's interest by compromising safety to some governments and large international coalitions of governments who can set common rules and common safety standards puts us into a much better situation. That requires a broader understanding of the strategic situation, the position they'll be in. If we try and remain quiet about the problem they're actually going to be facing, I think it can result in a lot of confusion.
--:-- For example, the potential military applications of advanced AI are going to be one of the factors that is pulling political leaders to do the thing that will result in their own destruction and the overthrow of their governments. If we characterize it as, "Oh, things will just be a matter of you lose chatbots and some minor things that no one cares about, and in exchange you avoid any risk of the world-ending catastrophe," I think that picture leads to a misunderstanding. It will make people think that you need less in the way of preparation — things like alignment so you can actually navigate the situation, verifiability for international agreements, things to have enough breathing room to have caution and slow down. Not necessarily right now, although that could be valuable, but when it's so important when you have AI that is approaching the ability to really automate AI research and things would otherwise be proceeding absurdly fast, far faster than we can handle and far faster than we should want.
--:-- At this point, I'm moving towards sharing my model of the world, trying to get people to understand and do the right thing. There is some evidence of progress on that front. Things like the statements and movements by Geoff Hinton are inspiring. Some of the engagement by political figures is reason for optimism relative to worse alternatives there could have been. The contrary view is present — it's all about geopolitical competition, never hold back a technological advance. In general, I love many technological advances that people are unreasonably down on: nuclear power, genetically modified crops, et cetera. Bioweapons and AGI capable of destroying human civilization are really my two exceptions. We've got to deal with these issues.
--:-- The path that I see to handling them successfully involves key policymakers and to some extent the expert communities and the public and electorate grokking the situation that they're in and responding appropriately.
--:-- Dwarkesh Patel: Well, it's a true honor that one of the places you've decided to explore this model is on the Lunar Society podcast. The listeners might not appreciate — because this episode might be split up into different parts — the listeners might not appreciate how much stamina you've displayed here. I think we've been going for what, eight, nine hours or something straight. So it's been incredibly interesting.
--:-- Other than Google Scholar typing in "Carl Shulman," where else can people find your work? You have your blog?
--:-- Carl Shulman: Yeah, I have a blog, Reflective Disequilibrium, and a new site in the works. I have an older one which you can also find. Just Google "Reflective Disequilibrium."
--:-- Dwarkesh Patel: Okay, excellent. All right, Carl, this is a true pleasure. It's safe to say the most interesting episode I've done so far. So, yeah, thanks.
--:-- Carl Shulman: Thank you for having me.
--:-- Dwarkesh Patel: [Outro] Hey, everybody. I hope you enjoyed that episode. As always, the most helpful thing you can do is to share the podcast. Send it to people you think might enjoy it. Put it on Twitter, your group chats, et cetera. It just splits the world. Appreciate you listening. I'll see you next time. Cheers.
