← Big Technology Podcast
Loading summary
Transcript111 lines
- [00:00]
A
How can governments use AI to become more efficient? We'll dive into it in a fascinating conversation with the CTO of Booz Allen and a former Amazon executive right after this. Welcome to Big Technology Podcast, a show for cool headed nuanced conversation of the tech world and beyond. Today we have a conversation that I've been looking forward to for quite some time. We're going to talk a lot about how AI can be used to make the government more efficient and effective. And not only that, not only the how it can be, but how it is being used today. Because today we're joined by the CTO of Booz Allen, Bill Vass. He is the man that is on the ground working on this and he's going to tell us what's going on inside the United States government, what the state of Doge is, and then everything else from robotics to quantum. It's going to be great. Bill, so great to see you again. Welcome to the show.
- [00:50]
B
Yeah, thanks for having me on. I'm excited to talk a little bit about what we're doing.
- [00:54]
A
Me too. So we're going to cover AI, we're going to cover DOGE at the very beginning here. But first, for those who don't know Booz Allen, I'd love for you to tell us exactly what it does in about 60 seconds. My understanding is it's a government technology contractor and about 95% of Booz Allen's work, or even more, is connected to government work.
- [01:15]
B
Yeah. So Booz Allen used to be a business consulting company and they sold that off in 2008 and that they have 22,000 engineers, about 3,000 AI gen AI experts, and about 8,600 cyber experts. And primarily we do hardware and software primarily for the government. We have some commercial business as well, and that's starting to grow also. But basically just a bunch of software developers that do everything from building hardware qubits for the government to running the GPS satellites and intelligence satellites, to 3D printing organs, experimenting with that for organ transplants with 3D printing. So it's a pretty broad range of tech, pretty exciting actually.
- [01:57]
A
And talk a little bit about how we have so much redundancy in government. I mean, to me, I'm not in government. I spent a little bit of time working at New York City government or New York City's Economic Development Corporation, which is a quasi governmental agency. I don't want to bore you with the details, but I'm stunned and sort of upset as a taxpayer that there could be this many. I mean, what did you say 255 different systems in the Pentagon.
- [02:24]
B
That was back when I was in the 90s when I was at the Pentagon.
- [02:28]
A
So who knows what it is today?
- [02:30]
B
Yeah, it's hard to know. It could be less. There's been also a lot of consolidation that occurs that across systems.
- [02:38]
A
I somehow don't believe that it's less given the sprawl of this.
- [02:42]
B
But I don't know. That's one of those things I'd have to go look at to give you an accurate number on what. Right, but so let's use 255 then. And some of it is that you have just all these parallel stove pipe organizations. Right. That are operating independently. Right. The government's broken into a lot of different agencies that operate independently from each other. And I think it's very hard for them to coordinate. It's interesting, Jeff Bezos at Amazon used to have this saying that 2 is better than 0. So we would have redundant systems at Amazon but then work to consolidate. Consolidate them over time. Some of it is politics. You've got different agency heads and other things like that over time and different divisions that want to do it themselves and want to do it their way and they think it's better than the other agencies way. When I used to work in the IC1 agency would sometimes do the opposite of the other agency just to avoid overlapping. And that used to piss me off as a taxpayer but there was not too much I could do about that back then. But I think there's just a lot of places like that where that kind of stuff shouldn't be tolerated. And I think that the push to consolidate is a good thing.
- [03:58]
A
Okay. And so I just want to get your on the ground knowledge here. So again speaking of Doge, a lot of people have talked about the layoffs but is this actually happening now? Like is this agency? I suppose it's not really an agency. It's kind of like a. A side agency I think is the best way to put it because it was the US Digital Service. Now it's Doge. Is it working to actually centralize technology today and is Booz Allen working to help that division on doing this?
- [04:31]
B
Absolutely, absolutely.
- [04:32]
A
I think material security is transforming how companies protect their most critical cloud assets like Google Workspace and Microsoft 365 with modern purpose built security that actually works the way people do the biggest cloud threats walk through three doors. Email, identity and data. Material was built from the ground up from day one for Google Workspace and Microsoft 365 not as a retrofit so it closes all three. It provides continuous protection before, during and and after an attack. You detect problems early, contain them fast and recover without chaos. Lean security teams scale through intelligent automation rather than adding headcount material blocks evasive phishing and impersonation. It protects sensitive content using built in rules for personal, health and financial information or rules you define. It also spots risky applications, unsafe settings and sketchy verification or password reset attempts, all without slowing people down. So get the overview@material.security. that's Material dot Security.
- [05:39]
B
The other thing they're pushing on which we like is moving to outcome based firm fixed price contracts from Cost plus and Time and Material. I always hated Cost plus and can.
- [05:50]
A
You, can you define what this, what this is?
- [05:53]
B
So outcome based firm fixed prices is, you know, to put like you're going to have a house built, you have a price that you pay for the house up front. Time and material is you have a house built and you just pay as you go based on the changes and all those other things. Both have advantages and both have disadvantages. I think early days in the government, many things were from fixed price and outcome based. In other words, you want an outcome at the end. I want to land a person on the moon or I want to, whatever it happens to be, it could be an outcome based type contract. Sometimes time and material makes a lot of sense. When you're asking the government to do something extraordinary they've never done before or no one's ever done, like 3D printing, an organ transplant. Right. We don't know that we can do that. No one's going to sign up for an outcome based contract like that. But migrating from on Prem to the cloud should be an outcome based contract that we know how to do that. So when you know how to do something, outcome based makes a lot of sense. And firm fixed price makes a lot of sense when it's something that the government's really pushing the edge of technology on, that's when you sort of have more of a Time and Material kind of contract in place. And I think what's happened is there's just too many time and material contracts over time. And the shift back to outcome based that Doge is pushing is a really good thing in my opinion. It's good for the taxpayers, it's good for delivery. Alternately, though, a lot of people in the government may not like it as much because they don't have as much flexibility. Right. They define it and they get what they've asked for and they like to make lots of you know, changes in pivotal cues. Again, back to this analogy of building a house. Oh, I don't like. We painted the dining room green. I didn't realize the green would look that bad. I want to paint it white, you know, and the, you know, time material you're paying for that with a firm fixed price contract, you couldn't change that. It would be like you got to live with a green living room. Right.
- [08:00]
A
That's good. It raises the stakes for people that are making the decisions in government and frankly, they should be raised. I've heard the term good enough for government work. I don't know if you've heard that as well.
- [08:11]
B
I have heard that. I don't agree with that.
- [08:14]
A
Makes me so upset because like you put it, this is something that does land on the taxpayers doorstep at the end of the day.
- [08:23]
B
I think though, having been in the government, I think that a lot of people don't understand. There are a lot of people who are incredibly technical, incredibly good, incredible, incredibly committed, working within the system and are delivering amazing things for our country and our war fighters. I mean, you know, the, I mean look at all the things that came out of the government. Integrated circuits, the Internet, gps, you know, it goes on and on and on. Those came from government programs. All of Silicon Valley is built on top of it, right?
- [08:52]
A
Definitely.
- [08:53]
B
And so I think that kind of core research is still important. I think it's still a place where the, the government can innovate and continue to deliver there.
- [09:01]
A
Okay, so let's talk now about technology centralization. The contract thing I think is important. Thank you for bringing that up. But I think that we should talk a little bit about the technology centralization efforts that are going on within the US Government. By the way, this is a model, I think like we're going to talk about us today because that's where Bill is working or working as a partner of. But I think a lot of this is going to be applicable to all governments and especially the AI components. So this is going to build right into that. But talk a little bit about technology centralization and whether Booz Allen or whether from your vantage point we're seeing the government actually work to consolidate those, you know, let's say many multiple systems that seem to be the same thing for, you know, different agencies.
- [09:46]
B
Yeah, I do see that direction and I think that's the big push. So for example, there's a bunch of different organizations that manage satellites. There's a bunch of different organizations that manage financial data. There's a bunch of different organizations that manage healthcare data. And in some cases you can consolidate them, in some cases you can't. And so I think it's just a matter of judgment where you can and can't consolidate them. For example, there's a lot of healthcare data in va, there's a lot of healthcare data in HHS and other things like that. And there's some consolidation and overlap that can be done. But there's also a very bunch of unique things in taking care of our veterans. There's things that veterans are exposed to and have to go through that you and I don't. And so they need a certain amount of uniqueness there, for example. So it varies. I think you just have to use your judgment on that. Right. I think there's a lot of places where citizen services could be much better through consolidation, making it easier to do your taxes, easier to make payments, easier to get payments from the government, those kinds of things. And I think you're gonna see a lot of that.
- [11:05]
A
Okay, I mean, did you see there was that story of, I think it was veterans records being held in a cave. How does that happen?
- [11:15]
B
So that's not really accurate. Exactly. Right. There is the need for long term storage at NARA and other places like of data that is underground and that data is also stored purposefully on non electronic formats. And the reason for that is we have legal requirements, the government to keep that data forever. Now you could change those legal requirements. There's reasons for those legal requirements. We have legal requirements to keep that data forever. And if you stored it on some type of technology, you'd be constantly having to upgrade that technology. You would have started storing it on 1600 DPI tapes and you would have had to migrate that to 2800 and then you would have had to migrate that to 6250. Then you migrate that to 37k, then you would have migrate that to disks and on and on and on and on. So there's a certain amount of logic to that. It is stored in OCR characters, so it can be automated at any time. So there are, there, there are certain things that are true. Other things are, in my opinion, are misrepresented.
- [12:28]
A
Okay. And at the beginning of our conversation you mentioned that the Doge team took a look at the Booz Allen technology and deemed it to be good so far. And I want to tell you or talk to you a little bit about the perception of government technology. And I mean and then sort of get your perspective on what the truth is. And again, this will lead into AI and I do keep teasing it, but I know we're going to get to it. But I think this is an important foundational question before we start that part of the conversation. All right, so I think the perception of government technology is that it's terrible that there's a certain amount, and this is not a comment on Booz Allen, but there is a perception that there's a certain amount of companies that figure out what to do to get through government procurement processes. And, and they are the ones that end up serving a lot of these government agencies. And while everybody else is on the current technology and using ChatGPT, we get the sense that the government is running on Windows 95 and like the nuclear processes are like basically running on Ms. Dos. Now I'm exaggerating a little bit, but I'll just give you one example. I did this internship on Capitol Hill and anyone who did it in the time that I was doing it, you had to use the system called iq, which was basically their CRM. This was like maybe a decade behind the state of the art technology. Now, of course it's a lot of work to modernize government tech, but how close is this perception of the government working on outdated technology to reality and what can be done to change it if it's true?
- [14:06]
B
So I think the government's a big organization and what you're doing, what you said is going to be true in certain areas depending upon how it's funded and how it's planned. Right. I mean, I assume you drive a car and you go places and you use GPS every day. Do you think your GPS is out of date?
- [14:25]
A
No, GPS is working great.
- [14:27]
B
Yeah. That's a government technology, right?
- [14:29]
A
That is, but that's a government. This is an important distinction though. That is a government developed technology that companies like Google have with a, I guess a profit motive developed and put into Google Maps. And that is the technology I use. But I'm talking about, because we're again talking about how a government operates and the operating systems for the government, those logistics systems. This is what the perception is.
- [14:53]
B
So like I said, I think you have things like GBS and the intelligence satellites and the Mars rover that you would say are working incredibly well. Right. The Mars rover has done amazing things on.
- [15:05]
A
We like the Mars Rover over here for sure.
- [15:07]
B
Yeah. You know, I think there, I think categorizing it as all government technology is bad is absolutely wrong. A lot of it is quite good. A lot of it is quite impressive. There are times when the government, the taxpayers decide and the administration decides to underfund things. And when they underfund things, then you have stale technology over time. There's also. And this happens in private industry as well. Geez, I can't tell you, having migrated so many companies to the cloud, how many ancient Windows 95 systems I've seen in private industry in the OT and IoT environments.
- [15:43]
A
Scarecrow, isn't it amazing how many people are still using Windows 95? I mean, the system really had legs.
- [15:50]
B
Well, because it just. They didn't change it. Right. They didn't have time to change it. And it's not. It was a decision in that corporation not to fund that. Right. And you, you hear about it all the time. So I don't think that this is unique to the government. I think it's a normal thing that you see. I would not say that the government is necessarily behind in a lot of other areas. I mean, we do a tremendous amount of GENI with the government. We've been doing it for two years. We started doing it before it was vogue in Silicon Valley, Right before chat GDP was so popular. We were using it in quite a number of places. We started using AI7, eight years ago, 10 years ago. We started like when I was. In 1978, when I was doing government contracting for autonomous vehicles in the ocean for an ocean engineering company. We were writing a neural network using AI in 1978. Right. So I think it's a mischaracterization. I think what you see is there's areas where we haven't invested intentionally. There's been decisions made there that they'd rather spend money on other things that do, you know, age over time and are not the best technology. Right. And there are areas where we've really focused in the military and the intelligence agencies, things that are life critical, where we have spent the necessary money and spent the necessary investment in technology and the latest architectures coming out of DARPA and the latest things that you see. So I think I see more cutting edge technology in the government often than I see in Silicon Valley and having been in Silicon Valley a lot as well. And I also see things where the government's partnered up with Silicon Valley to deliver things. I don't think there's any. You see Palantir all over the government. You see things happening with Andrew, you see shield, AI and Scout, and it goes on and on and on. So I don't think there's any lack of the government's interest in adopting the latest technology and being the most competitive. But at the same token, you could get to, I don't know, a building entry badge swiping system that's still running Windows xp. Right. Or whatever. And that is true in private industry too. I've seen it in private industry also. So I don't think that's unique to the government. I think that's just, you know, a matter of priority.
- [18:27]
A
Yeah, it's definitely tough to see this issue really resonate on the campaign trail. It's sort of like we're going to fix health care and everybody cheers and we're going to help small businesses operate without the red tape and everyone cheers. And it's like we're going to make sure the Department of Energy has a badge swiping system that doesn't run on Windows XP anymore.
- [18:45]
B
And. Yeah.
- [18:47]
A
And it goes wild.
- [18:48]
B
Yeah, I mean it's, you know, it's, it's, you see these kinds of things, like a simple thing like let's have a common health care record that all the insurance companies can use would save so much money. Right. A common format, common way to store data, a common healthcare exchange. We've been trying to do that for years. But what you end up with is all these different companies and all these different software providers and all these different congressional folks impact the technology significantly in positive and negative ways. I mean, when I first got to the Pentagon, I'll never Forget this in 19, this was about 1994 ish, 95 ish. Right.
- [19:34]
A
In time for the best edition of Windows.
- [19:37]
B
Yeah, yeah. My boss, who is the CIO for the all of Dodd was complaining that our security facilities still had those old fashioned piezoelectric buttons to put your combination in as opposed to a biometric and other things like that. So unfortunately for him, he mentioned that during his confirmation hearing and Senator Byrd, the company that made those was in Senator Byrd's district and they held up his confirmation because he threatened to upgrade the piezoelectric button.
- [20:15]
A
That's infuriating.
- [20:16]
B
Yeah, but that's how these things happen. It's not that Art Money didn't want to have a full biometric system, which we eventually did. It isn't that he didn't want to have all these other things. It's, you know, you run into these areas where you've got people protecting their technology. I mean, look at it this way. The way I view this in corporate it's very true and in government is very true. Whenever you see a bad technology decision, it's always politics.
- [20:43]
A
Yes. Okay, so how does AI then fix this? You mentioned that you've been Using. So again, Booz Allen is a government contractor. 95% plus business that Booz Allen does is basically building things for the government. So how has generative AI come into play here? And what, I mean, yeah, what sort of things have you found with chatbots in particular or any large language model? How does that end up make. How does that end up enabling the government to work more efficiently and more effectively?
- [21:16]
B
Yeah. So let's start with something we just did. So we just put LLAMA on the International Space Station on the edge on satellites. Right. So that enables the astronauts who are working on the International Space Station to have LLAMA to chat with in space with no latency to determine when things go wrong, how to fix them better. All of the manuals are ragged into that for the International Space Station or augmented into LLAMA running on the International Space Station to allow them to more quickly diagnose problems and help them diagnose problems. We have large language models going on to satellites to allow them to identify and tip and queue faster. We have large language models helping the VA do claims processing. What used to take many hours for a person researching on the claim process happens in a few seconds through the use of a large language model. We have large language models being engaged for autonomous systems. There's a big fight going on right now between what I'll call traditional AI and procedural based autonomy and large language model based autonomous autonomy. So scout AI, for example, a company we just invested in, is very focused on these large language model based autonomy. Right. And based on procedural input from humans learning. Most autonomy systems convert from a perceived environment into a 3D environment and then navigate through the 3D environment in the machine's brain, if you like. What they're doing is saying, well, we don't need to do that. We can go straight from the 2D image that comes from the cameras directly into navigation by learning from humans. That's a transformation in how autonomy will happen. There's large language models involved in how we're doing autonomy in general or coordinating across ISR information, intelligence, surveillance and reconnaissance environment. I mean, it's everywhere and it's in everything already. Right. So I would say that the government has been an early adopter of machine learning and an early adopter of a lot of these large language models in specific areas where it makes sense.
- [23:38]
A
So we just had a go ahead.
- [23:40]
B
It's not everywhere, right? It's not everywhere. Another thing that I'm seeing more and more, I mean, certainly we use large language models for code development. We use copilot and Claude and Q and Cursor and Klein for doing code development. Here at Booz Allen, I see the government using it more and more for code development, for their internal development as well. So I think those kinds of tools are happening also to accelerate development. I think it's. I wouldn't say there's other areas where it's not being used at all. And it should, though. I mean, this isn't going to happen everywhere overnight. I think it should be used. It should be used a lot more for doing fraud management and financial systems. It could be used a lot more in the irs. It could be used. I mean, I could go. I mean, there's a lot of other places it can be used to, you know, and, and large language models aren't a panacea. They're not perfect in everything. You need to have the proper guardrails in place. You need to have one model checking another model to make sure there isn't hallucination going on. You need to often have, for example, you know, it. It being the first round of things, and then a human checking it in the second round. So, so, so, for example, if you're doing with just regular AI. At Amazon, we did a lot of cancer identification from MRIs and CAT scans. The ML was about 98% accurate, which is tremendously good. It's 100% accurate, though. So you do still want a doctor to look at it. You have the ML filter ahead of time and then it goes to the doctor with recommendations. I think there's. And then as the doctor provides feedback, the model just gets better and better and better. You know, I mean, the reality is this is all just math, right? All ML is just math. It's, you know, vectors and it's tensors and it's, you know, it's all just math. It's not magic, it's just math. And so the more dense data you provided that's accurate, the more accurate the model's going to be be over time. The more you control the tuning parameters, the more direct it's going to occur. Right into what you're trying to get an outcome of.
- [26:11]
A
Right. So, Bill, we just had a couple of AI critics on the show a couple weeks ago. They wrote this book called the AI Con. They don't really trust that AI should be used for information retrieval. I suppose hallucinations are an issue. I suppose they think that the, the, this may.
- [26:30]
B
I don't know.
- [26:31]
A
I don't want to speak for them, but it's top of mind. I guess they would Suppose that instead of going out and doing the research yourself, having the AI go and do it for you will atrophy your brain. So I'm curious a couple. Let me ask you. So given that, let me ask you a couple of questions.
- [26:49]
B
Okay.
- [26:50]
A
Having astronauts use generative AI to decide what to do on a spaceship is pretty high stakes. So how could we be confident that they're not going to kill themselves in the process of using these chatbots? And then secondarily, do you worry that we're going to get government workers relying on these AI bots and then not able to think critically about the work they're doing?
- [27:16]
B
Do you use a calculator?
- [27:19]
A
Well, you use a calculator, so no. Bill, I've heard this before. I'm just again, channeling the critics. But I want to address this. This is the big question. Sam Altman would say that large language models are just like the calculator. But there has been research, including some research from Microsoft, that shows that the reliance on LLMs can decrease the ability to think critically. And in fact, you've brought up GPS a couple of times, and there have been some studies that say overreliance on GPS also limits the ability to think critically. So I do think that there's an argument to be made, and I'm still not sure where I fall on this argument, which is why I love speaking with experts like you, that a calculator and a large language model are two very different technologies when it comes to this question.
- [28:07]
B
Yeah, I don't think so. I think the astronauts are using the large language models to augment where they have to go through tons of manuals, and it references the manuals directly so they can see what the manuals say and they can still search the manuals directly. So I think there's. We use large language models at Amazon to help debug things in our fulfillment centers, and it was very successful in those areas. But you still have it references the manuals directly so you can avoid hallucination. You can see what it actually found and how it found it. Does it atrophy us? Geez, that's an interesting question that's hard for me to answer in some ways. I mean, we certainly like myself personally, I use GPS all the time and I use navigation systems all the time. And if you asked me to drive somewhere that I drive a few times with GPS without the gps, I'd have to really go look at a map and figure it out. I mean, is it like people ask me, well, did you take this street or that street. I'm like, I didn't pay attention to the names of the streets, but is that important? I can always go look at a map. I can always do those kinds of things, right. I think that just like any tool, you know, you can cut yourself with a knife in the kitchen. You know, you don't, you don't have to tear things apart with your hands, right? It's, you know, oh, geez, we've lost the talent of tearing things apart with our hands because we've invented knives. Right. I think that that's overblown. But I also think that people just need to remember that it just like any other tool, it's not perfect. It's going to have limitations and they need to understand those limitations. Right? And I mean, I'll just give you a perfect example myself personally, when I was first I was using chat GDP and I'm putting in a whole home theater. And I asked it, I want the best laser projector. And laser projectors used to be like 30 grand and now they're like 2 grand. And so they're getting to be, you know, affordable. And so I asked it for, gave it all these parameters and it came back with five laser projectors, several I'd already heard of. And then there were two that were perfect. And I must have spent 20 minutes on Google looking for them and realized that it had made them up. Okay, that's perfect. I asked for.
- [30:31]
A
Maybe that's a business idea.
- [30:33]
B
Yeah, but, exactly. But that's really though, the important thing to understand on how these tools work. They're just doing stat statistics, right? And understanding the two parameters and all these other things, they really are just doing a lot of math on what's the most likely answer that you're asked for. Right? But the same thing could be true for a Google search. I think people will say that Google has made people lazy too, because all the world's information at your fingertips. But that's a wonderful thing too. But just like on Google, you can go down a rat hole of all sorts of things that don't really exist. Same true of these models.
- [31:12]
A
So let me ask you just one last question about this. Then we're going to move on to robotics and quantum and some other cool experimental technology in the second half if we dream about what the. And by the way, I love this conversation because we never talk about public sector here and we really should. So again, appreciate you being here. If we think about the best case scenario, like we've outlined a number of problems with, and some Good things. But a number of problems with the way that the government operates this. If we get to a place where AI lives out its promise, what does the public sector, what does it look like? Like, what are the benefits that we see within the government? Does it enable the government to provide services better? Does it enable us to interact with citizens in a smarter way? Like if we dream about a best case scenario, what does that look like?
- [32:03]
B
I think that that's exactly where it would be is better citizen services. Faster, more efficient delivery of citizen services, a reduced overall cost, ideally. But remember, on the reduced overall cost piece, these models use a lot of GPUs. They are really expensive to train and they are really expensive to run inference on today. So that's another area that we really question sometimes the ROI of some of these things because of the cost of all of it. So that's another balancing factor. I think we don't have good data yet on the roi. And so that'll be the cost of operating the model and training the model and running the inference on the model versus the feedback. And I think some of that is we don't have good metrics to be able to track those things. And so we're working on those as well. That is something we're working on. But I would imagine a world that's got better citizen services that can deliver things faster and get things done faster and do validations faster. But you know, there's other sides to this too that where you, you shouldn't go overboard. At some point in time a citizen should expect to talk to a person.
- [33:08]
A
Yes.
- [33:09]
B
All right.
- [33:09]
A
That's going to be the case for the all companies that go to this. But I guess I would take a really smart large language model over a phone tree where you hit the number and it says goodbye. But anyway, these are personal gripes. Okay, you just made me think of one more thing. I'm going to ask this before we go to the break here which is this week we're talking at a week where President Trump is out in Saudi Arabia. This episode will air a couple weeks after. But the investments, I don't think they're time bound and that is that we see that Nvidia is going to do multi hundred thousand hundreds of thousands of GPU data center with the Saudis, Amazon, your former employees committing, your former employer is committing to invest 5 billion in Saudi Arabia. What they're going to do is, I think it seems like it might be the largest scale sovereign AI experiment we've ever seen. So I'm Kind of curious if you think that that is going to be a good testing ground for what, what governments can do with this technology and will u at booze. And do you think the world will be watching closely what Saudi does there?
- [34:19]
B
Yeah, we'll definitely be watching. I mean, I was actually in at aws. I was a big advocate for the Saudi region, and I was actually at the Saudi region launch at the LEAP conference just outside of Riyadh there. I think there's a tremendous amount of brain trust happening in Saudi Arabia and investment there in their movement to technology and their movement to, you know, diversify their oil investments into other areas. And, you know, both. Both clean energy and tourism and technology is really the areas that SMB is focused on. So I was excited to see all that. I thought it was moving in a positive direction, but certainly we'll be watching it. We'll be watching it, how it evolves, and, you know, hopefully, you know, at some point I'll be involved in it again. I really enjoyed the work that I did getting the region up and running in Saudi Arabia and the work I did in the UAE and others when I was out working at Amazon. And I think that's an area to watch. I think that's a good investment and the right thing to do to transform that region in a lot of ways.
- [35:36]
A
Okay, well, look, we're gonna go to break now and now and then talk about some of, like, the more sexy tech topics after this. We're gonna talk about robotics, autonomous, Quantum, and. And maybe a little Amazon with Bill when we come back right after this. And we're back here on Big Technology Podcast with Bill Vass. He's the Chief Technology Officer of Booz Allen, and it's been a fascinating conversation so far. All right, look, during the break, I said I got to ask. I kept pushing the brake off. So we're back from break, but I have one more question that I want to ask you sort of related to our last segment, and then we move on to autonomous and robotics. Amazon had very clearly or has very clearly defined leadership principles set by really one leader, Jeff Bezos, and that's been the way that the company operates. Are there. What would you say the leadership principles are for the US Government and do they shift time to time because of the fact that the CEO shifts every couple years?
- [36:40]
B
That's interesting. I think that you caught me off guard. It'd take me a while to come up with leadership principles for the government, but they certainly do shift, and it depends on the focus of the government. At different times in different areas. Right.
- [36:58]
A
How about today then?
- [37:00]
B
How about today? I think there is a focus on efficiency. The other thing that I like is there is a focus that we had at Amazon. We had a leadership principle. One of my favorites there's a number of them was bias for action. Right. That was one of my favorites. And so I think the government's got a lot more bias for action right now. And I think that's a positive thing. The other thing that was a great Amazon principle I liked was think big, because a lot of working on innovative things. And I think that people are willing to think big about what could be accomplished and throw off some of the shackles that have been there before and accomplish big things. Customer obsession is one of my favorites at Amazon. I don't think the government is as customer obsessed as it should be and they need to be thinking about that instead of saying and services. And I think that's an area that could be improved. Another area that I'm seeing is dive deep. And that's another thing I like at Amazon as well, because I like to dive deep into the technology and I do a lot of whiteboard sessions, things like that of diving into how the architecture is going to work and how all the different components are going to work together. I was just actually diving deep into a big AI project we're working on to do actually transform contracts from time, material and cost plus to firm fixed price, which we talked about a little bit earlier, using AI to do that. But I think those are things I'm seeing and those are positive things and those are things that I liked at Amazon and continue to like.
- [38:33]
A
Okay, so it seems like what you're saying is that some of the Amazon thinking is starting to make its way into the US government, which is interesting. Okay, so, you know, speaking of think big, that is a good one and leads us to some of like the bigger projects that you're involved with. And one of those is autonomous driving. And I think if I'm right about this, those are some of the projects that are both related to the government and not. And some of the clients you might have that are outside of the government. And so can you give us a sense, I mean you're, you're very big into training in synthetic environments and that leading to results in the real world and adding synthetic data. But there's also, if you, if you think about the reality of where self driving is today, there's Waymo, which I think is obviously it's expanding fast and it generalizes a Bunch of tech of its technology, but also, you know, takes some shortcuts. I think there are a lot of human operators out there that will sort of get those waymos out of tricky situations, if I'm not mistaken. And then there's Tesla, which is, which is, I would say advancing, but not quite there yet. We don't have autopilot now. So how far away are we? I mean this is sort of the essential question for autonomous driving conversations. How far away are we from seeing this stuff be mainstreamed?
- [39:59]
B
So I have two Teslas and I play with full self driving all the time. It's entertaining, but I wouldn't trust it entirely. Right. If you trust it, you're going to be in trouble. So it's not 100% there yet. It's a hard problem. It's interesting that you mentioned that the picture on the whiteboard behind me is for software defined vehicle and all the different components of vehicle running across hundreds of thousands of synthetic simulations. Simulations. And so we work really closely for example, with Nvidia on Omniverse. So Omniverse is a synthetic simulator or environmental simulator that has full physics and full fidelity and that's really amazing. A lot of the autonomous driving training that has been done in robotics training has been done using Unity and Unreal Real over time. And those are great environments as well. They look very much like video games when you run them, but people don't watch them. They're all running in the machine memory. And Omniverse is sort of the first to go that next level of not being constrained on something that might have to run on a console. So it's pretty amazing. Rev out there, I've been working with him for years on this.
- [41:13]
A
Yeah, we have an episode with Rev Le Bedian in the library.
- [41:17]
B
So yeah, he's great. He's great. Yeah. So, and then, you know, you're working in that environment. He would have Talked about the three computer problem where you've got the computer, that is the training computer, that's their H2 hundreds and things like that. That is looking at or learning from the synthetic environment where you're feeding in real and synthetic data into it. And then there's the three that after you create your inference model, it runs in car and that's the smaller computer, that's a third computer. And I talked about this a lot in the Velocity article that I wrote for Booz Allen is how this flywheel is accelerating autonomous driving and all these other things. I know there's a very long answer. Getting back to your Question of when we will have it. I think you'll start to see real autonomous driving over the next five years. You know, maybe I'll, I'll be burned by that prediction. There's still a lot of complexity in doing it. I, I worry sometimes. I love having my Tesla drive itself. My wife hates it, but I love it. It's. It's entertaining, but I do have to take over and I do have to pay attention. I'm probably paying attention more when my car is driving itself than when I'm driving my car myself, because I'm, I'm watching everything he does and I'm very proud of it when it does things well, you know, and, and sometimes I get scared with some of the things it does also. So I, you know, and, and the, the thing that's interesting, Tesla gives me the option, when I correct it and take over, you can hit the steering wheel button and, and explain to the person who's going to look at what your correction, what you did and why. And I do that all the time because I want feedback, I want it to get better. Right. I. And that kind of feedback. Remember, the Tesla has this advantage very much like the Echo devices at Amazon, where they're able to crowdsource training from the users. So basically they're learning and training their model based on all the millions of people driving Teslas every day. That's given them a big upfront lead in autonomy in a lot of ways, because they have that training set and they have the ability to generate synthetic data for the edge cases in that training set as well. And the more data you have with these models, the more parameters you have, the more accurate the model becomes, which we discussed earlier. Right. If you don't have enough density in your parameters, you're not going to have a good model. There's areas where I think the models still have a long way to go. Like you probably look at somewhat at a stop sign, which way their wheel is faced in their car, like to know where they're going to go, even if they're not signaling. Right. I think that's a nuance that's going to be very hard to train a model to do at this time. Right. But eventually it'll have to learn to do that. The resolution will have to be good enough on the sensors to see that when you stop at a stop sign and you've all stopped at the same time, one person waves the other one on. The models couldn't understand those kinds of things today, but they're going to have to be trained to, to do that. We have a lot of traffic circles here in Washington, D.C. and not many people can drive in them. Well, and either can autonomous vehicles. There's right now an oblique angle with my Tesla, the stoplight going the other direction on an oblique angle. If it can see it, it thinks it's green on my stoplight. That's a bad thing.
- [44:50]
A
You don't want that.
- [44:51]
B
Yeah. So I think those are all of those edge cases will get solved over time and the models will continue to get better. So I'm optimistic that there will be a day when I can go to sleep in my backseat and the car can drive itself, but it's not tomorrow.
- [45:10]
A
And it's a similar system that actually is being used to train robots. Just like the Omniverse system with Nvidia trains, cars and simulated environments. I imagine the same system is being used. They have their own foundational model now to help robots, humanoid robots, navigate the real world. And it's interesting. I mean, I'm sure you saw there was this half we've talked about on the show. It's kind of hilarious. There was this half, but also interesting, there was this half marathon in China.
- [45:40]
B
Oh, yeah.
- [45:40]
A
Humanoid robots. And like, you know, most of them ended up falling on their face or one of them with some fans on its arms, I believe propellers took a hard 90 degree turn and you see its trainer with a rope attached to it, like flying out of the, out of the scene. And the robot crashes into the boards and falls apart. But one of them did finish and had to change batteries three times, but finished the half marathon in a respectable time.
- [46:09]
B
Yeah.
- [46:09]
A
And so the. I think there is a. Again, speaking. Throwing the conventional wisdom out there for you to comment on. There's a conventional wisdom that the US is behind China on this.
- [46:22]
B
And. Yeah.
- [46:24]
A
Yeah. So, but yeah, I'm curious, like, I'd love to hear you. I'll just say this and you can decide to bat it down or whatever. Is the US paying attention to what's going on there? And is. Does the government then take a role and saying we need to help accelerate this, or is it completely left to private industry? Because in China, we know the government is pushing it.
- [46:44]
B
Yeah. So I don't think China's ahead, but I don't think they're behind. And I think that's an important, important thing. One of the reasons I left at AWS and I loved being at AWS, I worked on 63 of the services there and built a lot of them myself, worked on quantum computing and robotics and a whole bunch of things is I was worried a little bit about government adoption of AI and investment in technology to keep up with the Chinese. And so Booz Allen, because we are so involved in the highest technology in the government, was a great way, I felt, to more directly influence and improve that technology. And that's why I joined Booz Allen is to pivot to really focusing on that. Because I was worried about. I'm worried about us falling behind the Chinese and the government. It's a combination of government and private industry that's going to do it. You're right. The government in China very much invests in technology. They're very smart and long term thinking about how they invest. And there's a blurred line between government and private industry in China. And I think some of the stuff we're doing now in pivoting to a big focus on AI and a big focus on what we call the pacing threat, which is, you know, making sure our technology is ahead of China in the event that there was some type of conflict. We want to avoid the conflict by making sure our technology is superior. And so that's what we want to do. And that's where the focus in the DoD on lethality, lethality system, that's the focus on advanced technology and pushing DARPA harder, That's the push the focus on this. Public and private investments in AI and public and private investments in space and public and private investments in silicon development and quantum computing are going to be very, very important as they've been in the past. Right. So I think the government needs to move faster. And it's good to see a lot of these things happening. And that's part of why I joined, was to make sure the government is moving faster, to take everything I learned at Amazon and at sun and at Liquid Robotics, where I did the autonomous systems, and bring all the best of private industry to bear in the government.
- [49:19]
A
Well, appreciate you doing it. Let's close here with Quantum. We rarely talk about Quantum on this show, not because it's not interesting, just because it seems so far off. In fact, there was this moment where obviously the stocks don't tell the entire story, but Quantum stocks were riding up. And then Jensen Huang was like, don't expect Quantum to show up anytime in the next decade and just sort of sawed off half the value of almost all these stocks. But you're touching Quantum stuff as well. What is the realistic picture of this? Where the state of Quantum is today?
- [49:58]
B
Yeah. So we've been. I started the Quantum initiative at AWS when I was there, and we've got a lot of great people working on that. I was, you know, involved in, in getting DoD to invest more in quantum in the, in the early 90s and some of the core research in there, especially around ion traps and electromagnetic cryogenic machines at the time. So the good news about quantum is that the machines actually work and you can get output. The bad news is that they're way too noisy to get valuable outputs yet. And so it's really the error correction that we're focused on right now. And so with your iPhone or your laptop, you've got error correction code on it. A very small amount of the compute because you have alpha particles flipping the memory on your machines we're working on right now, and they're correcting that in the error correction code code. So maybe 1 or 2% of your CPU usage or your compute usage is for error correction. On a quantum computer, it's the opposite. You have a massive amount of work you have to do to do error correction because the atomic particles are so affected by the environment. And so the big challenge is getting that error correction to work. Now, again, the positive news, we're at a point where we understand the engineering necessary to make the error correction get fixed. Right. And what it will take to get to hundreds of error corrected qubits. The goal would be to get to a thousand error corrected qubits. Right. But just put that in perspective, that's going to be around 7 million physical qubits to do that. That's a big number. And so the first machines that you're going to see coming, I don't think you will realize this yet, are going to be about the size of a football field.
- [51:54]
A
Wow.
- [51:55]
B
That'll be the size of the machine. And that's because you have to have millions and millions of qubits to get just a few fully functional error corrected qubits. You have to have them constantly error correcting each other. Quantum computers differentiate from digital computers, or classical computers that we call them now, in that they have this two unique things that are unique to quantum physics that are hard for people to understand. One is superposition and the other one is entanglement. And if anyone tells you they actually understand how those things happen, they're lying to you.
- [52:38]
A
Because I was about to say, I cannot tell you how that works.
- [52:40]
B
Yeah. But an analogy I'll use is I'm a car guy and, and when I hit the accelerator in the car, I know if I'm again in a gas car exactly how the cam works and the crankshaft and the spark plugs and the valves are an electric car. I understand exactly how the motor and, you know, the, the, the inverter and, and all those things work and the batteries are working together to do that. When my wife drives the car, she doesn't understand any of those things, but she can drive as well as I can. Right. She doesn't care to understand it. You have the skinny pedal, the fat pedal and turning the wheel. Right. You can drive a car without understanding other things. We can drive entanglement and superposition extremely well without actually understanding how they work, what causes them. And the way you program a quantum computer is by using superposition to control the qubits and microwaves for electromagnetic machines or lasers for the other machines, which are neutral atoms charged or ions atoms and photons primarily. And we can set it, we can operate it, we can measure it and we can entangle it and we can run formulas on it and get outputs today.
- [53:58]
A
And so what is, what does this enable? Like when this is, let's say you have that football field size quantum computer computer, what does that enable?
- [54:05]
B
So the biggest thing that it will enable first, because effectively you can think of it as building molecules in memory and using those molecules is going to be material sciences and chemistry first. So in fact, one of the targets for Amazon's working backwards document for our quantum computers, a thousand error cryptic qubits could do a Hamiltonian on ammonia. Ammonia is the most produced. We've been producing ammonia since 19 for almost over 100 years and it's probably the most produced chemical. It's in fertilizer, it's in petrochemicals, it's in plastics, it's in just about everything. And it's very expensive and energy intense to produce. We know by watching bacterial interactions that it can be produced at low energy state. We just don't know how. So in the past, like a high temperature superconductor, superconductors in general have been discovered accidentally in the labs and then leveraged in the future with a Hamiltonian simulation. You can say, here's the outcome I want. Give me the chemical formula that we'll give it so you can reverse engineer an outcome in chemistry on today's classical computers for ammonia. If you took all the iPhones and all the laptops and all the Android phones and all the cloud computers on Earth and put that, that simulation into it, it would run for longer than the history of the universe. Wow. So in other words, you can't do it With a thousand error rector qubits, it would take about three minutes on a quantum computer. Right. So it's tremendously life changing, if you like. It will change our lives in a big way as these material sciences come into fruition and we start using them.
- [55:51]
A
How far away are we from that bill?
- [55:54]
B
I think 2032.
- [55:58]
A
So less than a decade.
- [55:59]
B
Yeah, not that far. Not that far for the first ones, I think you'll see in 2027, 2028, the first hundred error corrected qubits on fast machines, I think you'll see that before on slower machines and on the neutral atom machines. I'll probably see that they'll be too slow to solve some of these problems, but they'll beginning to solve some of these problems. So material sciences will be the first thing you'll see. There's certainly worry in the government and banks and others about having quantum computers break cryptography. So we do, we are deploying today, both at Amazon and at Booz Allen and others, a quantum safe cryptography. Because quantum computers don't do everything well. You're not going to run a website on a quantum computer, right. It's not going to replace your computer. It's going to be like a math co processor if you like. Right. That's the way they'll be used. And so there are algorithms. Quantum computers as far as we understand, will not be able to solve well. And so we do classical encryption plus another layer of quantum safe encryption today. And the reason to start doing it now is in around 2040, we think there'll be enough qubits to start to break encryption and secrets last longer than that. So we need to start, oh my goodness, you need to start encrypting. So most of the banks are already using quantum safe cryptography. A lot of retail starting to use it, the government's starting to deploy it. But I think you really should have urgency on deploying and turning on quantum safe cryptography. That's something Booz Allen can help you with and others can help you with as well. You're worried about that because people can record the transport of your information and then break it later. And so that's, that's a big deal. And I think this is important for our country too. The country that has this first will have a tremendous lead over all the other countries in material sciences at first, but later in cryptographic sciences. And then ultimately a quantum computer will be able to solve the traveling salesman problem and things like that, which is very interesting to people like Amazon who ship packages around. So Optimizing the shipping of packages would save billions of dollars for Amazon. And so that's one of the reasons they're investing in quantum computers as well. It's not just to be ahead for the cloud, it's also for their internal use. And so I'm very bullish on where this will go. I think the we're at a point now where it's more engineering than science, which is a good point. Point to be. You know, when I started working with these machines, it was more science than engineering. And there's still a lot of hard problems to solve. There's scalability problems. How are you going to scale all this? One of our big investments at Booz Allen is a company called Seek, which I'm very bullish on. So the nice thing about them is they build like the equivalent on classical computers like the Asics and all of the BIOS that would sit around the cpu. That's what they build. They don't focus on the CPU or the qubits, they focus on everything around it. So they kind of will win. No matter which of the four different types of quantum computers win, they'll be able to provide the control systems and other things like that very efficiently. In fact, I'll be heading to New York in a few days to go do a deeper dive on their lab and things like that. So, yeah, it's an exciting area. It's not for the faint of heart. It is complicated. There are many still challenges to overcome, especially scaling machines to, to be data center size or football field size machines. For these first machines, having them be stable enough to run long enough to complete a calculation once you get them working, and error correction, error correction and error correction. I mean that's really the name of the game right now.
- [59:37]
A
Okay, you've convinced me that we have to cover this more on the show.
- [59:40]
B
Yeah, thank you. I spent a whole show going over.
- [59:44]
A
Maybe we should. Maybe we should. I'm sure we're going to get some feedback on this part. Okay, last question for you, then we're going to wrap. You were the president and COO of Sun Microsystems Federal.
- [59:55]
B
Yep.
- [59:56]
A
From 2006 to 2011.
- [59:58]
B
Yeah.
- [59:59]
A
So let me just put it. That's the, that is the federal version of Sun.
- [60:04]
B
Yeah. State, local, federal, all of that. Yeah.
- [60:08]
A
All right. So at Meta's headquarters, I'm sure you know this, they kept the old sun sign sort of as a indication that to themselves that you could be at the top of the world one day in tech. And things move so fast that next Thing you know, somebody else is using your building and your name is going to be painted over.
- [60:28]
B
Yeah.
- [60:29]
A
Having worked in the tech industry for quite some time, Bill, what is your sort of lesson about how fast this technology moves? I mean, it's interesting that you went from sun to a company whose, whose motto is always day one.
- [60:43]
B
Yeah, I know.
- [60:45]
A
So talk a little bit about what it takes to survive and sort of the lesson that we can learn from Sun.
- [60:53]
B
So the only constant is change in this industry. That's one motto that I have. Another one is don't let the best be the enemy. The better. You know, you can always be working that. Another one would be, you know, you must be your own best cannibal. That's an Andy Grove statement. Right. So whatever you do, that's great technology, celebrate it, get it working and then replace it. If you don't replace it, your competition as well. I think Sun's challenge, and I loved working for Scott McNeely. He's an amazing leader and it was fantastic working with him. And Andy Bechtelshine and Bill Joy and James Gosling, I mean son, invented a tremendous amount of technology. I would say always impressed. They invented routing and ip, they invented symmetric multiprocessors. They invented network attached storage. They invented a lot of these things. I think the challenge that sun had is a couple of things. One is they built things for engineers. And I think that's a lesson that we all have to watch. Our end customer needs to, to be people, not engineers. Not that engineers aren't people, but you know what I mean? I think the other thing that they did, they didn't do well is they didn't know how to sell a lot of their technology. They didn't do a good job of transforming from the invention to the sales cycle in a lot of cases. And they did a couple of transitions. They transitioned successfully from being a desktop company to a server company. They became the dot and dot com, if you like. You know, that was a good transition, but they did attempt. They had an early day of cloud called Sungrid. I was involved in it. A bunch of people were on it. It was like EC2 on AWS. But they ran into this innovator's dilemma where they couldn't sell it well because of the transition from selling capital to selling service. The street loves recurring revenue. Wall Street. Right. But they hate a transition. They don't give you any, any, any break in a transition of a business model. Right. So, so they, they, they just. What have you done for me this quarter? And so sun had a lot of challenge moving from. I could sell a capital asset and recognize revenue immediately, large revenue. So sell a million dollar server. Recognize a million dollars of revenue to sell a server as a service for 15 cents an hour, which in the end ends up making more revenue, but starts off making a lot less revenue. And so I think it was a combination of not being able to manage that financial transition. I think there were other mistakes we made. I was an advocate for open sourcing Solaris X86 early, and we didn't. And I think Linux wouldn't exist if we'd open sourced Solaris X86 early. And that would have been a tremendous transformation because there was a lot of amazing things in Solaris. It's still an amazing operating system, just not heavily used anymore. Linux is reinventing a lot of the things that, you know, Solaris had containers back in the early 2000s. Right. Containers are all, you know, it had virtual machines, it had, you know, a trusted environment, it had, you know, all of these linear scalability. I mean, a huge number of things, you know, advanced threading systems that are, you know, still struggled in some of the operating systems today to get. But it should have been open source and it should have been on x86. Right. But it was very hard, I think, for sun to give up Spark and the advantages that they felt Spark had and to understand the value of open source at the time. They eventually did. But they open sourced Java, they open sourced their identity systems, they open sourced Solaris, they open sourced all those things. And it was great. And a lot of people have benefited from those things being open source still today, but they didn't do it soon enough.
- [65:07]
A
Well, Bill, this has been such a fascinating conversation. We covered so much public sector AI and government, Doge Robotics, Autonomous Quantum and Sun. So I would say we've done our work today. Great having you on. Please come back.
- [65:21]
B
Yeah, if you want me to come back and spend a day about quantum computing, happy to do that. And thanks again. It's been a great discussion.
- [65:28]
A
Thank you so much. All right, everybody, thank you for listening and we'll see you next time on Big Technology Podcast.