Main Topic: The Future of Defense Intelligence in the Age of Artificial Intelligence
Key Participants:
Key Points:
Takeaways:
Implications:
Future Directions:
Artificial intelligence has taken the world by storm. But despite the hype around personalized avatars or podcast language translation, artificial intelligence is not only impacting the creative spheres; in fact it’s hard to find an industry that isn’t being touched by this technology – and defense of our country is far from excluded.
In this episode, originally recorded in the heart of Washington DC this January during a16z’s American Dynamism Summit, a16z General Partner Martin Casado and a16z enterprise editor, Derrick Harris are joined by the first-ever CTO of the CIA, Nand Mulchandani.
In this wide-ranging conversation, they discuss the evolving relationship between analysts and AI, how governments can keep up with this exponential technology, and finally, how it’s impacting both offense and defense.
This episode is essential listening for anyone interested in the intersection of technology, national security, and policy-making in the age of artificial intelligence. Stay tuned for more exclusive conversations from a16z's second annual American Dynamism Summit in Washington DC.
Topics Covered:
00:00 - Intelligence in the Age of AI
02:28 - Rethinking Jobs and AI's Asymmetric Power
05:00 - The History of AI in the Intelligence Community
07:00 - Operational Utilization of AI
10:40 - Analytic Capabilities and Uncertainty
12:56 - AI's 'Hallucination' Concerns
16:37 - Analyst Skill Sets and AI Tools
26:29 - Supply Chain and Open Source
31:35 - Public-Private Partnerships
41:33 - Government as a Customer and Partner in Tech
42:43 - Policy, Technology, and Regulation
Resources
Learn more about AD Summit 2024: www.a16z.com/adsummit
Watch all of the stage talks at AD Summit 2024: https://www.youtube.com/playlist?list=PLM4u6XbiXf5pAKmk1AeZ9964KGScf4lHM
Read the CIA’s announcement around the new CTO role: https://www.cia.gov/stories/story/cia-names-first-chief-technology-officer/
Find Nand Mulchandani on Twitter: https://twitter.com/nandmulchandani
Find Martin on Twitter: https://twitter.com/martin_casado
Find Derrick on Twitter: https://twitter.com/derrickharris
Stay Updated:
Find a16z on Twitter: https://twitter.com/a16z
Find a16z on LinkedIn: https://www.linkedin.com/company/a16z
Subscribe on your favorite podcast app: https://a16z.simplecast.com/
Follow our host: https://twitter.com/stephsmithio
Please note that the content here is for informational purposes only; should NOT be taken as legal, business, tax, or investment advice or be used to evaluate any investment or security; and is not directed at any investors or potential investors in any a16z fund. a16z and its affiliates may maintain investments in the companies discussed. For more details please see a16z.com/disclosures.
Stay Updated:
Find a16z on X
Find a16z on LinkedIn
Listen to the a16z Podcast on Spotify
Listen to the a16z Podcast on Apple Podcasts
Follow our host: https://twitter.com/eriktorenberg
Please note that the content here is for informational purposes only; should NOT be taken as legal, business, tax, or investment advice or be used to evaluate any investment or security; and is not directed at any investors or potential investors in any a16z fund. a16z and its affiliates may maintain investments in the companies discussed. For more details please see a16z.com/disclosures.
Hosted by Simplecast, an AdsWizz company. See pcm.adswizz.com for information about our collection and use of personal data for advertising.
Narrator
We're a human intelligence operation. We can now make anybody be anybody and sound like anybody and look like anybody. Stop thinking about automating getting a 10%, 20%, 30% on your job. Tell me in 5 to 10 years how you're going to reimagine your job. This is no asymmetric superpower, which by the way is very, very different than the internet. It's because policymakers now have a knob where we now have to decide explicitly this or that. These are the largest compute projects that mankind has ever done before. Like, we've never done anything close to this. Life finds a way, which if there's demand for it, people will supply it.
Artificial intelligence has taken this world by storm. I mean, just think about it. Here in 2024, anyone with an internet connection and a few minutes to spare can literally spin up a Disney avatar of themselves, translate a foreign podcast into their native language, and even get help writing their vows. But artificial intelligence is not just impacting the creative spheres. In fact, you'll be hard-pressed to identify an industry that's not touched by this technology. And the defense of our country is no exception.
In today's episode, originally recorded in the heart of Washington, D.C., back in January during A16Z's American Dynamism Summit, A16Z general partner Martin Casado and A16Z Enterprise editor Derrick Harris are joined by the first-ever CTO of the CIA. Yes, that is the Central Intelligence Agency. We're joined by CTO Nand Mulchandani to discuss the future of defense intelligence.
In this wide-ranging conversation, they discuss the evolving relationship between AI and analysts, how governments can keep up with this exponential technology, and finally, how it's impacting not just offense, but also defense. I hope you enjoy this conversation.
As a reminder, the content here is for informational purposes only, should not be taken as legal, business, tax, or investment advice, or be used to evaluate any investment or security, and is not directed at any investors or potential investors in any A16Z fund. Please note that A16Z and its affiliates may also maintain investments in the companies discussed in this podcast. For more details, including a link to our investments, please see A16Z.com slash disclosures.
Derrick Harris
Martina, I'll assume most people watching or listening to this are fairly familiar with you and your role at A16Z. But because the CIA CTO role is relatively new, can you give us a quick background on that role and what your objective is?
Nand Mulchandani
Yeah, so there are really two stories here. One is the agency needing a CTO and what created that, and then my own journey to it. It all starts with Director Burns taking over the agency at the beginning of the administration. He, just like any great business leader, sat down and did a business review: what business are we in? What are the big threats? And the other pieces. The conclusion was actually fairly interesting and not a surprise: we had to pivot from CT, which had been the big focus for the agency for a couple of decades, to great power competition. And then the interesting, unusual thing was this big amorphous thing called technology. There was huge interest from policymakers in technology that we needed to start looking into and build policymaking around.
I think it might be helpful to the listeners to understand what CIA actually does, because I had to learn a lot about the agency when I came on board. Amy Ziegart calls it "spy-tainment" versus what we actually do. In this pivot to rethink the agency's focus on technology, three things happened. One, we created the China Mission Center, which is how we actually focus on threats and opportunities. Two, T2MC, which is the Transnational Technology Mission Center. And three, this weird thing called a CTO.
We're a 76-year-old spy agency, so we've been doing technology for a long time. But technology has been somewhat latent. With this focus on tech, we needed to focus on three different things that were different from what we were doing before. Number one: we are, as an agency, fairly vertically focused. We have five directorates that focus on five things that come together in the mission centers. The idea of the CTO is really to go horizontal versus vertical. The second thing: we are very focused on the here and now – there's a crisis and we jump on it. The CTO function gives us the luxury of looking a little bit out rather than focusing inward. The third thing is a little more external versus internal: engaging with the outside world. Those are the big dimensions.
Derrick Harris
Martin, people know you as an investor and founder, but maybe it's worth noting you also worked in the intelligence community for a little bit. So you come to this with some background.
Martin Casado
Yeah, that's right. About 20 years ago.
Derrick Harris
Before we zoom back out to policy and broader discussion, let's start with talking about AI, which is disrupting everything at the moment. I'd love to get both of your perspectives. Martina, maybe we can start with you on how you think about AI as it relates to the intelligence community – where we're at and probably where we're headed.
Nand Mulchandani
AI has been around for a very long time. Even when I was part of the intelligence community 20 years ago, we talked a lot about if you have all this information, how do you detect signal? A lot of that was very significant big data processing. A lot of the more advanced notions actually came out of the intelligence community. The intelligence community deals with things like entropy and covert comms, and these ideas are fundamentally tied with AI. There's a longstanding history.
What's interesting is how this new generative AI world impacts the intelligence agency or intelligence in general. One idea is: we can now make anybody be anybody and sound like anybody and look like anybody. That could be a huge problem because deepfakes could be a problem. But my conclusion is that if a computer generates something, the ability to fingerprint that isn't that difficult. It's actually not that hard. So I think it becomes much easier to detect if people are using AI tooling. But that also means we can't use them as tools. We've got to go much more to the fundamentals.
In this weird irony, we've got this set of new tools that allows anybody to sound like anybody or be anybody, and that's going to be heavily used around the world. But for those who are sufficiently sophisticated, it's going to be quite possible to detect them. In some ways, this creates a nice cover in chaos for what our agencies are very good at: a much more human-focused, less core technical approach to these intelligence problems.
Derrick Harris
It's probably fair to say AI has been around for a long time – machine learning, deep learning back a decade ago. Generative AI and LLMs are definitely newer. Do you think about the defensive and offensive side like Martine referenced?
Nand Mulchandani
Yes, and I'll compliment what Martine said. When you look at the two big functions we have as an agency, we've got the operational side and the analytic side. Those are broadly the two big functions. What Martine absolutely nailed is the operational side – spy versus spy, cat and mouse, all the usual stuff. The democratization of this stuff and the availability is going to drive each of our competitors to drive this up. We're going to be aware of that, and we're going to drive our own stuff up. We don't know where this is going to go, but it's definitely not pulling back. The stops are off.
However, we've always got to keep in mind that we're a human intelligence operation. We're a foreign intelligence operation. We're all source. We have a particular focus within the 18 intelligence agencies to focus on a particular thing. Everything Martine said about using, wielding, or scaling this AI in operations is within the context of taking our case officers and operations teams and making them successful. Any good team sport, we play a particular role on the field. Applying this tech to scaling and making our folks more effective and win is the game.
On the analytics side, it's a complementary but different problem. This is where the big data comes in. What's revolutionary about this is the promise of discovering patterns in large amounts of data that humans typically can't see. What's different now from the older days of big data analytics – I call this the pull model versus the push model. The analyst had to come up with creative stuff first and say, "I'm going to put a query in to find the information." The problem is the conceptual boundary of an analyst to hit the query that hits the data – it was like spear phishing. You're trying to get one analytic idea out of this massive data.
What's beautiful about this tech is that it can push stuff. You can almost invert the process where a lot of stuff gets pushed to you because it starts to understand what you're looking for and starts tailoring the stuff deeper and deeper over time. That has an evil twin problem that I've been thinking about: what I call the rabbit holing problem. The very thing that makes these products so effective – learning about you and being built to please you – can also take the things you as an analyst are weakest at and start rabbit-holing you down a path that amplifies your biases. We've got to be very careful about where we apply it, when, and how.
Derrick Harris
How close are we actually? Going back to the early days of big data, how personalized something will be or how easy it will be to sort of stuff? Are we still a ways away from automating a CIA analyst?
Martin Casado
I think we're now understanding the power and limits of this technology. There's an important topic: will this change the existing equilibrium in intelligence between offense and defense? My deep belief is exactly what Nan says: it doesn't change the equilibrium. There's going to be more tooling for offense and more tooling for defense. This is no asymmetric superpower, which is very different than the internet. That's why this is such a bad analog.
The internet was asymmetric – the more capabilities you had with the internet, the more vulnerable you were. When we were working on the terrorist threat, we were the most vulnerable nation in the world, and many of the people we focused on didn't even have laptops. AI is something that anybody can use. It doesn't change some fundamental equilibrium in an asymmetric way. So the first answer to your question is: does it allow for personalization or anything specific? I think it doesn't create any imbalance between two opposing sides.
The second answer is that we're pretty clear about the individual limits of LLMs right now. They're very good for a few use cases. They're very good if you ask one question and they give you one response. If that question is in the corpus of training data, it'll give you a good response. The problem is if the question is out of the corpus of training data – you don't really know if the response is good or not. That's fine if you're just asking one question because most humans can check that. So it's very good as a co-pilot for an analyst to help with their job.
The thing we've not seen any evidence of is agentic behavior. By agentic behavior, I mean you ask one question, then you step away and go get a coffee, and it does its own thing. The reason is that if it generates anything out of distribution – meaning it's not commonly represented in the training set – that error is going to accrue, and it tends to accrue exponentially, provably. So right now we view these as new tools that you use side by side, but they don't become their own separate entity.
Nand Mulchandani
That's an absolutely brilliant point – the idea of the accrual of probabilities times probabilities, which gets to the hallucination stuff. When it comes to my 11-year-old drawing unicorns, that's a feature, not a bug. Hallucinate away – that's amazing. When it comes to analytic capabilities and operations, we cannot have this level of uncertainty and not knowing and lack of explainability.
The interesting thing is that we're in such early stages of this game. Everything we're talking about is like we were in 1995. This was two years ago. We were just talking before this, all three of us old folks sitting around the porch. This could be a porch with rocking chairs, us talking about the early days of the internet. We could not have possibly imagined the stuff that's happened over the past years.
At the agency, we're saying there are a whole bunch of basic use cases that this stuff can be applied to. We're all in on it. I never want to give the impression that we're slow-rolling it. But to Martine's point, the applications on a per-use basis have to be thought through. This is not peanut butter that you can spread everywhere and get goodness with no thinking.
For instance, we're public with the fact that we have LLMs in production at the agency. We have it in production in the open source team. For those easy use cases – business automation, other pieces – we're experimenting, trying, doing stuff. The co-pilot piece: people typically jump immediately to the hardest of hard problems and say, "We're going to replace this." What we're challenging everyone inside the agency to do is: it's one thing to look at the low-hanging fruit and get that automated. The most interesting thing is we're challenging folks to stop thinking about automating getting a 10%, 20%, 30% on your job. Tell me in five to 10 years, how are you going to reimagine your job?
Here's where things get tricky. Many of the people challenged to reimagine their job are looking at this technology and learning it for the first time, understanding its power. All of us are still very far away from knowing where this is going to go. It's really hard for them to imagine how this prototype thing that's still playing around is going to impact or rethink their job. We're pushing and experimenting and encouraging everyone to try stuff, learn stuff, get up the experience curve. But we're not going to settle on an answer because that's not going to appear magically.
Derrick Harris
I read something recently that was basically a CIA analyst imagining their job five or 10 years out. It was very much what you described: enter a prompt, go get a cup of coffee, and by the end of the day, we've arrested someone in France for something. That's how people were. It's like when we took the kids to Disney World and went on the Epcot ride – the ride of the future from the 1950s or 60s, imagining what the world would be like in 2024. It's a handheld phone with a video monitor. So it's really hard to see where this stuff is going.
Noting that it's early, realistically, is this a thing where analysts just have different skill sets going forward? Different tools at their disposal?
Martin Casado
I think we can say something relatively specific on this, and then there's a bunch of stuff we don't know. Here's the thing that's relatively specific. The way large language models work is they get a whole bunch of data, and they basically have a distribution of how common the data was represented. That's what they do – kernel smoothing over positional embeddings, which is just averaging a bunch of words. For any time you ask it a question, it gives you the most common outcome.
What does this mean? For mean things you want to do, for the average thing, it would be very good at getting you an answer. For any standard rote thing, it'll give you an answer. The problem is if you want to do something in detail or something new or an exception, it doesn't know how to do that. There's no mechanism within it that will do that. So much of intelligence work, I would argue, is actually in the tail. These are the problems the intelligence agency is particularly good at.
I think we can believe there's a world where every analyst will have strapped on an LLM that will help them with the routine stuff. But so much of the job is tail reasoning – reasoning in the tail – that this is not going to remove the humans. This is just the intelligence community, but I think this is most work, particularly acute in intelligence.
Nand Mulchandani
I break it up into three pieces. First, does this technology replace the analyst? Second, the co-pilot model that Martin talked about – I do my work and have a little wing person that helps with routine stuff or scaling. But it's exactly his point: it's not the creation of new information from old information. Human beings uniquely create new things, new information. It's unclear whether these systems actually produce new information or new thought. They're just finding it or routinizing it.
The third is what I call the crazy drunk friend problem – the hallucinating, which has a role in some disciplines like making new art or poetry. In the analyst function, it's about finding that point on the distribution. If the average policymaker could think through the average use case, then what's the role of the analyst? The role of the analyst is to have this holistic piece of thinking through probabilities.
Martin Casado
That was a great explanation, and I think it's exactly right. Imagine you're an analyst paired up with someone who's been in the agency for 40 years. This person doesn't know new technology, just knows what everybody's done a whole bunch. Is that the entire perspective? Of course not – you have to evolve, but it's a very important perspective. LLMs are very good at being that old person. They're very good at saying, "This is how we've done it in the past. Here's a recommendation." But that's why we have people to make a decision: do I want to do something new, something in the tail, or do I want to listen to this person who's been around for a very long time? It's a very concrete mental bookend. The majority of the value – the new stuff – will still remain with the person.
Nand Mulchandani
One interesting thing for me, having spent 25 years in the Valley doing startups and now being on the government side, is that a lot of the tech discussion is about possibilities and creation – innovation, invention, getting great people to do awesome stuff. But you have to flip over to the buy side of technology. This is my second run – two and a half years at the Pentagon and now at the CIA – being on the buy side of technology. I just took a red-eye in from California. Over there, it's all about possibilities. Here, it's about how we take the job and function we have to operate in with all the constraints, which are not artificially imposed.
The CTO office at the agency right next door is where the PDB gets made for the president. By the time something hits that threshold of getting into the presidential daily brief, you can imagine the level of scrutiny and analysis our analysts put into this work. The excitement and hype about this technology versus us absorbing it and making it battle-ready is a long distance. Hopefully, we're portraying that side of the equation – it doesn't happen rapidly. It takes a long time for folks in their own disciplines and careers to understand the actual impact. The absorption of this technology does take a long time for it to disrupt a particular individual or discipline.
Martin Casado
I experienced something similar when I was in the intelligence community, but modestly different. I was in ops, and we would have missions with very specific requirements. The piece of technology that came from the private sector just wasn't suited. Little-known story: my PhD work was in software-defined networking. That was before it was done in hardware – it was too tough to program. Honestly, the insight for that work all came from my time working with the intelligence agencies. We had to build a network that had to deal with certain things. I came from the computing side and thought, "These things aren't programmable. To do what we need to do here, I have to program them because Cisco just doesn't know what we need to do." So there's a flip side. Everything you're saying, I totally agree – it takes a long time to be adopted. But your requirements and needs are a bit different than what the market is.
Nand Mulchandani
I've been a huge software-defined networking fan for a very long time. I stole it for a paper I co-wrote with General Shanahan called "Software-Defined Warfare." I love it. This idea of how you take something on a hardware curve and push it onto a software curve – how do you do this reprogramming? The question we've been asking inside is: what does software-defined intelligence look like? What's the next level where it's maybe more push versus pull, going across that distribution curve?
One other piece right in the middle of this is the policymaking debate. Even in this analytic function, the work each of us does has encoded in it the policies and outlines of what we have to do as part of a job function. I call this "code as law." When you look at the applications we use at the agency, we encode all of those rules and regulations inside of them. I call this a thresholding problem. Inside a line of code in our application, there is something that says, "If probability of X happens beyond this, then do this versus that." We have millions of lines of code with those if statements.
What that means is we've been implicitly taking human decisions that a programmer or policymaker made and encoding them into code. With these new AI-based systems – both previous supervised and unsupervised learning and now newer algorithms – these are still probabilistic algorithms. Except now the probabilities stare you in the face in a way previous systems didn't push. Previous application systems never said, "Do you want the 49% or the 69% answer? Now you decide whether 69 is high enough." It would encode an arbitrary number 50 and anything above versus below.
Why this is becoming such a debate is because the probabilities are now surfaced to the user's face. If they aren't, we have to train people to start thinking about how to make a decision on a probability when the system punches out a number. That's the big difference between before and now – we're having to retrain everybody, and that's why it's become a policy issue. Policymakers now have a knob where we have to decide explicitly this or that. We're in the same world, but more in an explicit world.
Derrick Harris
Before we get into policy specifically, is there a sense where the advent of open source and the general acceptance of open source now, especially in AI and other emerging tech, eases the adoption to some capacity? In Intel, we have the open source intelligence question as well. There's commercial technology not up to par, but you need to be able to remake it in your own image and get something functional for you. I'd love to hear your current perspective, Nan.
Nand Mulchandani
The thing we're challenged with on this landscape right now is that each company is offering a particular LLM. To me, it's turning into different types of wine or different varietals. I'm not a wine snob, but I'm imagining it has something to do with lineage, exquisite data. "This one was grown in the hills of Normandy. This one comes from this part of Italy."
Martin Casado
You're saying it's all nonsense.
Nand Mulchandani
Yeah, you're saying it's all nonsense. Right, so you're getting these elephants with different lineages and different vintages and different data. Each of these systems is going to behave very differently over time. The question is: everybody's going to train on the whole internet, so everything's just going to look the same? There's a second question we're having to deal with, and this is where the open source question comes in: the ability to start training with your own data.
The question becomes: do I take a base algorithm or system that somebody has built that has ingested the entire internet – both good and garbage – and use that as a base platform? It may have a certain set of biases it brought along with the garbage and cesspool that the internet is. All of a sudden, I'm using that as my base. Or do I want my pristine, handcrafted, bespoke thing?
That's where the availability of these algorithms becomes really interesting. However, it has the opposite problem: it doesn't have the imprimatur and the stamp of a large company that has the experience building large software systems, training, verification, and other pieces. So we then have to understand the stuff ourselves and do all that work. I think that's a trade-off we're dealing with.
Martin Casado
Yeah, yeah, yeah. That makes a ton of sense. Here's a bit of a historical perspective on this, similar to what I touched on before. In my experience, the government and the intelligence agencies have to solve problems that market forces don't really solve for. To do that, there has to be some sort of flexibility and programmability, and open source has always been a key component. Quite famously, SELinux came out of the NSA. They used Linux because they required that flexibility. There have been a number of changes to algorithms like crypto. I don't remember the details, but the NSA was like, "Quadratic linear programming will break this, so go ahead and make this kind of change to the algorithm." These are types of things that have been coming out of intelligence for quite a while.
I remember 20 years ago when I was deep in the weeds working on one of these problems. There was an old timer there, a tech old timer, and he said to me, "It's so great you have this open source because we can work with it." He said, "I remember a time when all we would get was supercomputers. They would come out of IBM, or SDI, or KSR, or whatever. We'd buy one of everything. But we only got what the vendors created." Something I do think about—and I'd love to hear your perspective, Nan—is that it's one thing to open source weights and biases. That's one thing. But these are the largest compute projects that mankind has ever done. We've never done anything close to this. So even if the weights and biases are open source, I don't know how much you can modify it. It almost feels like we're going back to this old mainframe day where it's great to have it and you can operationalize it, but you're not going to have the same level of flexibility as with traditional software. At least, that would be my guess.
Nand Mulchandani
No, absolutely. Anyone's ability to do something with that information is limited because it's not just the numbers. You have to have the expertise and the compute. You need all the other pieces there.
Martin Casado
How many of these training runs are there? Hundreds of millions. That's my point.
Nand Mulchandani
Great. I have that information. I just don't have the compute cycles to be able to do anything with it, or modify it, or change it. So you're totally right. It will help with verification, training, testing, all that stuff. But you're totally right about retraining. It does feel like we're going back into this kind of almost supercomputer era. You're completely right. By the way, the government is the best at procurement and using it. There's a long history of that, but it's a very different ballgame now.
The other piece with open source that we have to deal with a lot is the supply chain issues.
Martin Casado
Of course, yeah.
Nand Mulchandani
Right. This is a whole new level of paranoia that you need to have working in a spy agency. It's one thing for your program to not run or for some e-commerce customer to have a bug. These issues are still really, really important for us. So what does that look like? If you make the supercomputer analogy, what does a new public-private, D.C.-Silicon Valley partnership look like in terms of actually implementing and operationalizing AI models and AI systems?
Martin Casado
I'll take a shot at that because it's something that I think the US government is thinking about writ large. When you go talk to universities right now, one point they make is they don't have the compute power to rival Microsoft and OpenAI and these companies, which is mind-boggling, right? Because to your point, the supercomputer systems and everything. When I was at Cornell, we had a supercomputing site on campus with money the government had put in place to have these supercomputing centers. Illinois had one, Cornell, et cetera. We don't have that equivalent now. So the National Science Foundation, I think some of the money the government has allocated, is presumably going towards building these large-scale systems. Or the opposite.
Nand Mulchandani
We're not a policymaking organization. We're actively trying to keep people from building them.
Martin Casado
Well, we should. I think there needs to be a model where this comes away from just corporate organizations doing this. You're part of government, so it's a different thing. But the question is whether this needs to get opened up in a bigger way.
Nand Mulchandani
The thing that brought me into the intelligence community is I was doing computational physics on supercomputers at a national lab in the weapons program when 9/11 happened. They said, "You have all the clearances, you need to move to this new area." And they moved me to the intelligence community. I learned a bunch of stuff that way. I will say the work we did on those supercomputers, the government was the best. It created entire new disciplines of scientists and career professionals in universities. It leaned totally into this. And that's why we maintained a leadership position in the world through that, in compute. And we still do. I do think it's a risk that we don't take the same attitude now.
Martin Casado
Listen, I'm a VC here and I'm saying I don't think the private markets solve everything. I do believe in public-private partnerships. I do believe in institutions. I do believe the government has a big role to play here. But I think that role is investing heavily in people, tech, and careers, and reaching out. My fear is that they're doing the opposite. Back in the 90s, it was exactly what you said. I went to Northern Arizona University, a small mountain school, and we had ASCII programs where they would come out and invest in us. I feel that now, even though we have this new technology that is very powerful, and we are the leader, and it came from the United States, we're kind of pulling back from it. I do think this is a moment where the U.S. has to take pause and understand: are we undergoing a doctrine change where when new technologies come, we run away from it instead of towards it? I think it's a real quandary.
Nand Mulchandani
Yeah, it does seem like that. Maybe there was a shift at some point. Maybe it was the Internet. I'm not sure where. In the supercomputer days, yes, you bought a system from IBM or XGI. But that was a hardware system running very specialized software. Today, everything comes out of these huge companies that have access to all the data and all the computing power. I don't know how that affected the power shift, but do you sense a way to bridge that gap?
Martin Casado
Yeah. Here's the other side of the argument. The very dynamics that have led us to this point of creating these algorithms and systems and breakthroughs—there are also hundreds of companies that you and other VCs are funding, hacking away at the problem to make these things available. There's huge amounts of work going on in AI-specific chipsets, both on the training side and the inferencing side. A whole bunch of algorithmic changes are going to happen that refactor these algorithms to do a better job in terms of scaling and being able to shrink them without a dramatic loss of performance. We're in the early innings of this game.
We don't know what the next five years will bring. But for sure, you've got thousands of really smart people hacking away at the problem. I think they will come to some medium where, yes, hopefully the government funds or academia ends up with these large compute places to rival commercial. And at the same time, the availability of hardware commoditization will get to a point where we can run all kinds of interesting algorithms at scale with really cheap, readily available hardware. That's the techno aspect of it. As I say, life finds a way. If there's demand for it, people will supply it.
Nand Mulchandani
Yeah. The question is, will the private market solve the problems needed for things like global defense or national security? Historically, the government has played a role in innovation and training. Think about nuclear engineers. A lot of that came up from government programs. This is a slightly different topic: what are we doing in government to do a better job of working with industry? A large portion of this job I have is this idea of American dynamism. It's Silicon Valley leaning in and the government having to lean in together for us to meet in the middle, to be a better supplier and a great customer. In my role at the agency, one of the big areas of focus is how do we become a dramatically different customer? I spent two and a half years at the Pentagon, which was its own gigantic problem in terms of where software-defined warfare ideas came from. In all honesty, we don't do a great job in certain things where we could be world-class. We are working really hard to change that.
For instance, there is no ready app store for spy software. There are absolutely certain things we need to build and write inside the agency that are very specific. It's also our competitive advantage. We're not going to buy stuff that's readily available for everyone. We have our secret sauce. We build it. However, what we don't do sometimes is analyze that. We take it too far. There's stuff that's readily available outside from commercial land that we don't think about buying, deploying, and implementing at scale. In the past year and a half, I've personally spent a lot of time focusing on what I call "commercial first."
We need to rethink our strategy. If something is available on the outside, how do we bring it in? But we have procurement processes, ATO processes, and security processes that don't lend themselves well to rapid acquisition. So we're trying to hack away at those. On top of that, the security needs to run stuff on the high side is very expensive. For commercial vendors to go through that process is expensive and an investment. We have to create incentive structures to bring them in. It's not simply that we can will it into existence. It's a systemic problem we're trying to attack. There have been certain breakthroughs inside the agency over the past year and a half.
I probably can't tell you what those are, but we've made huge progress in rethinking things.
As an agency, there are a number of cultural shifts we're going through internally. First is the human versus tech. As a human intelligence organization, it's very interesting because of the changes in tech outside that our case officers operate in. We publicly talk about what we call "ubiquitous technical surveillance"—UTS—video cameras, biometrics, et cetera. As a spy agency, we hate tech when it's applied against us, but we also wield it. The other interesting cultural thing at the agency is the power of the individual. We train individuals to go do heroic efforts. That's our job. That's the agency's job: to go into foreign countries and spy. However, the tech aspect is very concrete. The big change in tech over the past couple of decades is scale. This idea of the individual versus scale is a cultural thing we're trying to rationalize. It's applying enterprise large-scale tech to an organization that teaches individuals to have agency to go do things. That's a very interesting one.
Then there's short-term versus long-term. We're an agency that's ready to go at a moment's notice, which we do incredibly well. But to do large-scale, enterprise-wide technology transformations takes time. Then there's open versus clandestine. As an agency, our folks are not trained to be out there in public. Director Burns has made engagement with the outside world a big priority—engagement with the technology industry, which happens out in the open. The idea of carrying a business card, being present, being on podcasts—these are all culturally new things for the agency. We ourselves are going through this huge transformation, dealing with tech. How do we change our thinking in this new world? And then AI on top of it is another layer of complexity in terms of changing how we operate. It's a very interesting, exciting, but also somewhat confusing and transformational time for the agency.
Derrick Harris
Do you think that given American dynamism is having a moment right now—D.C. and Silicon Valley are going to be talking more than ever, certainly not since the early days—people are talking like you guys both have Intel and software experience.
Martin Casado
We're at the exact intersection.
Derrick Harris
Do you think that starts to help ease some of the friction in terms of making procurement processes faster, making adoption a little easier, making it easier to hack on stuff?
Martin Casado
Maybe I'll say something that would be great if you took us home. The issues you just talked about have been around for a very long time. I don't think that's the harder bit. Of course, we can make procurement better. Of course, we can make communication better. Of course, we can have better public-private partnerships. I remember talking about this in the 2000s and the 90s. Here's what I think is the most important, and you alluded to it before. The internet caught the U.S. flat-footed a bit. There was this notion of asymmetry. It ended up having exponential effects because there was so much connectivity.
When it came out, it took us a while to come to grips with it. My biggest fear is that with AI, people are fighting the last war, and it's to our detriment. A lot of things people are concerned about with AI are really internet things, and we've gotten on top of them. You can't take that mindset. You can't take the mindset that investing in this stuff is bad because it's asymmetric. You can't take the mindset that this is inherently exponential. You can't take the mindset that this is core critical infrastructure because it's very, very different. This is a new type of technology. It's as useful for us for doing good as it is for changing the threat environment—which I actually don't think it changes the threat environment a lot.
I think both the government and us in industry need to come together and acknowledge this is a new technology that's beneficial. We're better off learning about it than running away from it. We can't take these old lessons from the internet and apply them, because then we're going to miss the train. For me, this is the higher bit and the most important thing. A lot of the things you've talked about are important, but they will follow in due course.
Nand Mulchandani
Wow. I'm not sure I'm going to top that. I do have to be careful. The big disclaimer I have to put on this is that the CIA is not a policymaking shop. Our job is to support policymakers with very objective, by-the-book analytic support on the questions they're asking. What Martin just outlined is exactly the policymaking debate and discussion going on: how do you regulate versus incentivize? What happened with things like privacy and security impacted consumers, and therefore it impacted lawmakers. So we got pulled into leaning in. Now, whatever is happening in that area, the issue around 5G was a big national security concern. Then the CHIPS Act for semiconductors. Look at what happened in crypto. And now AI. I'm listing them because at CIA we track five, ten, fifteen emerging tech areas. We have world-leading domain experts that follow each one because there's demand from downtown on those questions.
I listed four, and the fifth is privacy, security, and regulation around social media. These five areas of technology now have a spotlight on them. To Martin's point, where it lands and ends up is purely a policymaker domain. Our job is tricky in many of these situations because these are, by definition, emerging tech areas. 5G and semiconductors are scaled industries, but the rest are emergent industries. In emergent industries, having been in startups, we don't know where this stuff is going to land. It becomes really hard to understand who to talk to, who to believe. Do we forecast? Do we not forecast?
Where do we think it's going to land? There are no solid answers to any of these questions because every six months it's going to be something new. How do you build policymaking on top of emerging tech areas? That is an art. It's up to lawmakers and policymakers to figure this out. That's where you end up with things like executive orders rather than laws. It's very interesting how the policymaking apparatus works. You end up with 100 pages of executive order stuff that outlines general ideas and questions. I think there's going to be a leaning in and convergence between industry and regulators, because this stuff is moving.
The tech is moving. Policymakers are learning more. They ask more questions. The tech industry moves this way. It's an iterative, long-term process, but it's up to the players—including the folks with the money and the investors—to have a seat at the table. The good news is the agency is playing the role of "we're friends with everyone." We talk to everybody. That's our job. Gather a lot of intelligence. Analyze it, hopefully with AI. And then help our policymakers create this stuff.
Derrick Harris
We very much appreciate what you do. Thank you for that.
Nand Mulchandani
Thank you.
Narrator
Now, if you have made it this far, don't forget that you can get an inside look into a16z's American Dynamism Summit at a16z.com/ADSummit. There, you can catch several exclusive stage talks featuring policymakers like Deputy Secretary of Defense Kathleen Hicks or Governor Wes Moore of Maryland, plus founders from companies like Anduril and Coinbase, and funders like Mark Cuban, all building toward American dynamism. Again, you can find all of the above at a16z.com/ADSummit. We'll include a link in the show notes.