DataFleets is grateful to host Marty Chavez and moderator Kenn Cukier as they discuss the increasing role of APIs in the tech economy, the need for regulation in Big Tech, and the promise of privacy-enhancing technologies to unlock data for the benefit of society.
Kenn Cukier: Hello and welcome to DataFleets’ fireside chat! I am so pleased to be chatting today with Marty Chavez. My name is Kenneth Cukier. I'm a journalist, and I'm a book author on books on data and society and technology.
And Marty Chavez is an interesting fellow. He started out studying biochemistry. He got a computer science degree and a PhD in computer science from Stanford. He was an entrepreneur, then spent 25 years on Wall Street looking [over] Goldman Sachs and actually being one of the forces that revitalized how the investment bank thought about the world and also thought about technology and thought about helping their partners succeed with technology. Since retiring from the bank last year, he's been thinking long and hard about where he's going to be putting the next chapter of his life.
And it's been in a whole slew of companies - large companies, large banks, as well as startup companies. So where Marty goes intellectually, and as an investor, really is sort of a lens into the future. So it's something that I want to talk to you about, Marty. Thank you so much for joining us.
Marty Chavez: I'm delighted to be here, Kenn. Thank you for that gracious intro.
Kenn: Absolutely. As a beginning, let's just talk about what you see as the biggest technology trends going on right now, before we drill down into some of them and what they mean.
Marty: Well, I wish I had thought of the term. It was popularized by Marc Andreessen: software eating the world. I think there's really no pithier and more accurate description of what's going on. So the way I think about what's happening is you take a business process or scientific process that used to happen exclusively in reality, in the physical world. And now you can make that same process happen in an alternate or virtual world. And being able to do that to a process or a business transforms it in some interesting ways that are fascinating and are hard to predict, but across the board, that's the theme that's going on. Doesn't really matter what business you're talking about. How is that business being reenvisioned? How are its workflows being recreated? And how is that all happening in software? That's the overarching theme.
Kenn: Okay. So actually, let's dive down a little bit deeper into that because I think there's more to it than how you've characterized it. Here's what I mean by that: if you think of Moore's law as a frame by which the industry can sort of have an end point to understand itself, a distant point. That worked from the Sixties and Seventies the Eighties. And then when Marc Andreessen - Nineties and Two-thousands as well, of course - But when Marc Andreessen came up with his idea of “Software eats the world”, that was another frame. But that really characterized, almost, the Post-9/11 moment, the Two-thousands, the Two-thousand-tens.
Kenn: But some things changed in the last decade, and I think you're pointing your finger to it. On one hand, it's big data, and it’s machine learning. On the other hand, it's the (we'll call it the) API economy. Where, what you're saying is everything is programmable. Develop that more for me.
Marty: Sure. So the thing that has changed on that continuous trend that you just described that has been profound in the last ten to twenty years is we've gotten to the second half of the chess board in compute power, right? That famous, maybe apocryphal, story about putting a grain of rice on the first square and doubling and so on. It doesn't look like much, but by the time you get to the second half of the chess board, you really start to notice it. So the thing that we're really noticing, Kenn, is that we can now with the techniques you described - big data, machine learning, and so on - you can construct extremely high fidelity mirror worlds, or models, of reality.
In the past, really the compute power was too low. So as an example, in 1990, when I was a doctoral student at Stanford, I was working with a small group of people who had the modest ambition of training up a machine to make diagnoses in general internal medicine. Give us the symptoms and manifestations, we'll give you the differential diagnosis.
And the problem was that the compute power was just way too low. My whole dissertation was fast, approximate methods to get almost the right answer because the computers just couldn’t handle it. And indeed how I ended up on Wall Street is that (actually, this may sound odd) but the problems that Wall Street wanted to solve in the Nineties and [on] were really an awful lot easier than problems in life sciences. Wall Street had problems that were more tractable. It's not that hard to describe a foreign exchange trade compared to describing a human cell as it's living and metabolizing, right? So Wall Street had the money and the need to construct these mirror worlds. And by that I mean a software model that lets you ask what-if questions? Right?
It's one thing to build a model of reality. And then you can ask questions about the present, and that's super valuable. It's even more valuable if you can ask counterfactuals. Or ask about interventions. What if I did this? Or what if I did that? How would the world change? And if you can do that in software, and not have to perform that experiment in reality, then that changes the game.
So I could describe almost everything I worked on in Wall Street as losing billions of dollars in imagined simulations - without actually having to lose it in real life. Now, if you take that and apply it to all kinds of businesses, that's what we're able to do now, because as you said that Moore's Law has just kept going. So, problems that seemed unbearably difficult 30 years ago - like diagnosis in general internal medicine, like simulating an entire human cell or tissue or organ or organism or population - we're starting to be able to do that. That's the big thing that's happened in the last few years.
Kenn: Okay, so this raises a whole lot of tricky questions. I mean, if there's something great to play for, grandiose - the mirror worlds can actually help us improve the mirror worlds, and we can take what we learn in the simulation and bring it back to this one and improve our world.
Kenn: But of course not everyone can do it because it takes a lot of money, and it takes a lot of know-how, and I don't know how the companies are going to be able to meet that. And the ones that do might even concentrate the economy even more
Marty: Well, and certainly there is a kind of winner-takes-all, right? We've all read about OpenAI’s GPT-3 and its billions - more than a hundred billion - tuneable parameters. And that it costs tens of millions of dollars in electricity just to train up a model. So, that's a problem. But then you alluded to something else that to my mind is one of the most important things going on in the world. Might sound a little bit esoteric, or even geeky, but it is the API economy, right? So we don't know how, for instance, OpenAI is going to make GPT-3 available, exactly. But there's certainly a lot of discussion about giving API access to it. You'll have to pay for it, but you won't have to spend the tens of millions of dollars of electricity just to train it up. I won't go so far as to say that that will democratize AI and machine learning, but the possibility is there.
Kenn: But it's pretty grandiose. In economic theory, we had Coase who told us about transaction costs and asked the question of, “What does a firm keep within its four walls? And how does it interact with others?” And of course, the whole point about the API economy is that it's, there's a sort of, collaboratory and combinatorial capitalism that has much less friction now. So what are the opportunities?
Marty: That’s super exciting. Well, I'm glad you asked that. I just taught a course at Stanford in the spring. It was an amazing experience, and it was about software eating finance, specifically. And the central thesis that I have is - and we could debate, it's not like it's guaranteed to be right - but the old dichotomies that we always used to think of finance - the buy-side versus the sell side, a market infrastructure provider such as an exchange versus a user of the exchange, people who distribute market data, such as Bloomberg versus people who consume it. That's how we thought about finance since forever. I think that is all giving way to a new API economy, and in this economy there are going to be a lot of participants in the financial system. Some will be banks. Some will not be banks at all. And everybody needs to be a world class provider of some product or service.
You don't get to have a hundred of these providers, but you maybe get to have three to five. And then everybody needs to be an astute consumer of services provided by others. And so you're really just consuming a lot of APIs, and you're producing a smaller number of APIs. And so it's the orchestration and the management of risks, because of course, what you don't want is a single point of failure - one of your APIs that's being provided to you goes out, and then your whole service stops. You can't do that.
And then another thing that I find fascinating on your point on transaction costs is look at Amazon, who's really been the seminal pioneer in this API economy. Here's one narrative of Amazon: they are looking at every expense line item that they have and asking whether they can put an API around it and turn the expense line into a revenue line, starting with Amazon Web Services. But why stop there? They’ll go onto logistics and every other aspect of its business.
So you really are seeing the economy being turned around through APIs... in a way that has existed in the past, you know, we standardized tire sizes - that's a kind of API - but now we're doing it across the board.
Kenn: Yes. And we had a role for the state - the great project of the 19th century was the era of standardization. And so what the middle C was in Oregon [and what it was in another community] in Germany - it was vastly different than that. So we had to standardize it, and we had international institutions. One in Paris that standardized the mutual register. We had one that standardized telecommunication standards. It's a UN body, the world's first actually international treaty organization was an international telegraphic union.
Marty: I didn’t know that. That’s awesome.
Kenn: Okay, so the question is who is the platform controller? What private sector company is going to standardize and be the platform for these APIs, the iTunes for APIs? Who's going to own that?
Marty: Yes, so that's a...a lot of different people are going to own that. I actually pine for those days, I wasn't around, but when there were international standard setting bodies that everybody respected. There still are, right? You can look at, for instance, the 4G and 5G standard. That is an international collaboration under the auspices of, I think, the same organization that you mentioned. Still at it.
And then of course there's the internet standards. And so there's the requests for comments, the RFCs, is how many of the standards that we all take for granted now...TCPIP, later on WiFi, HTTP, HTML5... all of these are not so much controlled by a private company, but they're done through a collaboration.
You could actually go further and say some of the great defects of the setup have led to some of the problems that we now have. So, the standard-setting bodies for the internet, for whatever reason - just because they couldn't get there, or thought it was too hard, or maybe thought it wasn't necessary - we do not have a standard for identity and we do not have a standard for money. And those are probably really big misses.
Kenn: Well, those are great. Let me leave identity for a second. Of course, money...one standard could obviously be the central banks that issue it, or it could be blockchain, right? Bitcoin. But my interest was actually who runs the platform for the API economy, whether there'll be an iTunes, whether it will actually be decentralized, or will actually you see a gatekeeper there? What do you think?
Marty: I think we're going to be wildly decentralized. There are some companies that are striving to be marketplaces for APIs, and you might see some of those things take off. I think, Kenn, what generally happens - and this is certainly the strategy that we were working on at Goldman Sachs, where as you mentioned I worked for many years - and again, this is a point of view, not necessarily right or wrong, but I would say just take your business and encapsulate it in APIs, or important parts of it, and put the APIs out there and beautifully document them. Make them really easy to discover. This is what Stripe has done absolutely brilliantly, right? And then encourage people to use them and actually encourage your competitors to copy it. If you, a competitor, can go make something that is exactly compatible with my API, but under the hood, your implementation is completely different, I encourage you to do that. Because if I get to set the API standard, I've won.
I think that is the playbook of, for instance, Amazon Web Services. They're not waiting for anybody to define a standard. They're saying, what would data and compute on demand look like? What would be a great API for that? And let's make that happen. At Goldman, we would look at, for instance, transaction banking services or trading and risk management and liquidity provision services and document the API for everybody. And if an arch-rival simply copied it, that is great, because it means all of the clients are already tied into that API. They've built their platform on top of our platform, and that's creating a sticky and recurring and stable interaction with the clients. That's the dream. That's what winning looks like in the API economy.
Kenn: Okay, so from the point of view of a business process and an interaction with consumers, I see the API economy. Now there's a whole other aspect of technology and business in the world where we're in a sort of black hole, or lost at sea, whichever metaphor fits. And that is all of the data that companies have that they're not using themselves or interacting with with other partners. And the reason why is because of the difficulty of using data, and particularly personal data. And we're obviously chatting on a fireside chat sponsored by DataFleets, which has an answer to this. So let's talk about what those answers look like and what the future looks like.
Marty: You might've seen the book, I think it's called AI Superpowers by Kai-Fu Lee, and it's two different visions of what AI could look like, right? And one of them - and this would be the Chinese-led vision - there's an awful lot of data on everything that every individual has clicked on, on the web or on a phone, right? Consider that, that data set. And if your philosophy is maybe the state owns that data, or it's in the commons, or everybody can use it in an untrammeled way, and data privacy is not important (and that's an extreme description of it).
But just imagine that for a second. Well, that is a dream for machine learning, as you know, right? The problem that machine learning has evolved to solve nearly perfectly today is the labeling problem. If we've got a huge data set, and it's been labeled one way or the other. So as an example, we’ve got a billion images and people have labeled whether those images contain cats or not because people like posting photos of cats on the internet. Well, now we've got a great training set. These are cats. These are not cats. If you can frame a problem that way the machine learning algorithms we have can do a brilliant job at taking a new image they've never seen and telling you whether there's a cat. They can tell you what breed of cat. They can tell you how old the cat is. They can tell you all kinds of things, right?
So if you've got a huge data set and untrammeled access to it, and data privacy is not important...the concern is, is that going to lead to winning in this world of machine learning algorithms? And so what is the West going to do where presumably at least we say data privacy is important. I’m constantly surprised at how much of our behavior we give away to companies to just monitor, monetize for themselves in return for some services they provide. So the question is, can the West with appropriate data privacy safeguards still learn from this vast trove of data? Is that possible?
Kenn: That's the question. That's a great question. What's your answer?
Marty: The answer is it is possible.
Kenn: But what needs to happen for that? For us to enjoy the real data economy and all the benefits that accrue from it? Because, from my vantage point, every time a person goes into the hospital, we collect lots of data. We typically don't store it, save it, share it, reuse it, aggregate, and learn from it. We should be doing it. It's heinous. It's almost barbaric that we're letting the data go to waste. And I see that across the entire economy. We have some of the world's best minds and the world's best AI companies working on games, and trying to win at games, because they can collect the data because they can't meaningfully get access to the data that really matters to improve people's lives. That’s a scandal.
Marty: It is a scandal. And it's a scandal, not just because we have some respect for data privacy, but also because the data is just locked inside silos with really no good way to get at it. And so it's a mix of those two drivers that's the problem...so what companies will need, and this is where DataFleets comes in, and there are many companies working on this problem.
Companies need easy ways to do all of this fancy encryption. Very few people are going to actually be able to solve the problems and key management and do all the calculations to come up with these algorithms. But the good news is they don't have to, right? Just like I can get in a car and drive it by pressing on the pedal and turning the wheel - I don't really have to know what's going on under the hood. That's what companies like DataFleets are doing.
Kenn: Now, to get the standards that we need and to allow the data economy and the API economy to flourish, we need regulation because regulation can both restrict, but it can also enable. Washington DC is taking forays now, as well as around the world, at regulating the big tech platforms. It's not certain they know what they need to regulate or how. But at least they're now wise to the fact that they need to do something - that, actually, just looking from the sidelines isn't an answer. So let me turn to you and ask, what does wise regulation for the 21st century technology look like?
Marty: So, Kenn, I had a fascinating career experience, which I view as a peak career experience, which was the Dodd-Frank Act got passed, and then I was asked by my company to own the implementation of the Dodd-Frank Act for the trading business. That experience changed my life, and I think it’s a pattern for what can and must happen with tech. So every industry gets to a point where - not every industry, but lots of industries as there’s a whole pattern in history of this - get to a point where there's an oligopolistic, maybe monopolistic, concentration of market power.
And then there's always the question about does that lead to anti-competitive practices? And I am certain that that time has arrived for big tech. If you look at the current generation of companies, not all of them, but one way to describe them is they've happened upon this vast sea of data, which is human behavior. And nobody seemed to lay claim to that data. And so they said, “Oh, it's just there like a new world. And we are declaring it for ourselves and we are going to monetize it.” And there are essentially no constraints on that.
I think the analogy to financial regulation is there was a long period of time when there was deregulation. And the idea was let's just trust banks to look after their own risk. And that will be sufficient. And we don't need a regulator to tell banks what to do. And as we saw that didn't work out too well. And with Dodd-Frank, there's been - you know, my one concern about Dodd-Frank was it was an attempt to do about a million different things that got written into even more rules and regulations - I don't know that all that was necessary or useful, but there was something extremely useful embedded in Dodd-Frank, which goes back to where we started the conversation on software and modeling your reality and creating a mirror world in software of your business, that the Dodd-Frank Act and the Federal Reserve actually required banks...they didn't say, “You know, you might want to simulate yourselves nine quarters in into the future.” They said, “You will simulate yourselves upon request nine quarters in the future. And we'll give you a bad scenario, a severely adverse scenario, and you must demonstrate to us that even in that scenario, you still have enough capital to continue to perform your critical function of lending.” That changed the game completely. It is entirely why in the current crisis, you really haven't heard too much about banks. You've heard about struggles in almost every other industry, but the banks were forced by the regulation to be resilient. And the banks were forced to internalize a systemic downside scenario, meaning they have to have capital to absorb the loss amongst themselves without turning to the taxpayer for more capital.
And so I think you're going to see a similar pattern apply to regulation and big tech. And I got lots of ideas about how to do it. My first advice to the big tech firms is, do not resist the tide of regulation, right? You actually have to go and partner with the politicians and the regulators, which is hard because they will be appropriately suspicious of your motives, right? “We're here to help” - they're not going to be sure about that. But only these companies deeply understand their business, and the regulators are all smart people, but they need the companies to teach them how the business actually works. This is what Wall Street did.
Kenn: Okay, so that's so interesting. I don't want to take too much time on this, but I do want to probe it a little bit more to say that you've given us a structure or a form with which to think about it, but not the substance. We could be here all day talking about substance, but is there any one or two things that you think, substantially, regulation should look like?
Marty: Absolutely. So, first of all, I would say monetizing human behavior belongs to the commons. It does not belong to any one company. And any company that wants to access the click streams, the aggregated, anonymous click streams of people has to pay the commons for that. And the government can use that revenue...
Kenn: Wait, wait, wait, wait, wait. Let me stop. I'm a company. I've hired some of the smartest data people in the world to tell me what signals are relevant and which ones aren't. I'm going to invest a pile of money to be really smart and collect that data [and] another pile of money to make sure the data is clean. And then you're going to tell me that I have to give away this data to some commons, to my competitors. That's crazy.
Marty: Well, okay. Um, so, so right now...
Kenn: Because I'll just explain why, because it would chill, it would chill innovation. I wouldn't become a smarter firm that's willing to make those investments to learn what's relevant and what's not [in] understanding consumer behavior, if I wasn't able to monetize that myself and get my competitors away from it.
Marty: I am not at all worried about the ability of Facebook and Google to pay the commons for access to individual data and still derive insights and still clean it and still find the signal better than anybody else and still serve up those targeted ads. So just right now, they're getting it for free. And so this could be the source of new revenues for land grants, for education, like the land grant universities for education. That's one thing.
Second: opt-in national digital identity. Right? So on Twitter, there are some people with a little blue check mark, right? And so the idea would be everybody gets a national digital identity. And if you want to post to some kind of alter ego, you can. But it'll be clear to everybody that it's not a real person or you're somehow hiding your actual identity.
And then here's another one, which is a corporate tax surcharge on targeted digital ads - limitations on the viral multiplication of tweets. So, everybody has a right to post something, but they do not have a right to have that amplified again and again and again virally in newsfeeds. Right? So we're going to need better content moderation as well. So when you put all of these things together, you can start to see a new world where there's a little less polarization, a little less trolling, a lot more accountability.
And then of course the big one, and you'll know this extremely well, Kenn, is the Communications Decency Act, right? So, there is right now this huge arbitrage where somehow a few companies are deemed not to be publishers and not to have any accountability at all for what's carried on their platforms. And we could ask, did we get the result we wanted with that huge exemption? Or would we want to limit it in some ways? So that's an outline for what we might do.
Kenn: Okay. Very interesting. I think for a lot of business executives, that's very helpful and very clarifying, but a lot of them also are thinking about their business and their own personal careers in a time of COVID and the post COVID economy. What advice would you give business leaders, both professionally managing organizations, as well as personally, as executives in terms of what they need to know - a frame, if you will - that they can apply to their work and to their personal lives in this new world that we're going to be entering into as we sort of as COVID ebbs, but all of the problems associated with it still linger?
Marty: So, being a software guy, I keep coming back to the same idea, which is every business needs to have a model of itself. And for a lot of businesses that might just be a spreadsheet somewhere. Unfortunately, those spreadsheets are generally historical and retrospective. What were the earnings last quarter? But companies also have some sense of what their earnings are going to be in the future. So for instance, every startup has or ought to have a spreadsheet model that shows the cash, the balance in the bank account of the company. And there are only a few ways to get that, get money into that bank account. You sell equity in your company to somebody and they give you some cash. You sell a product or service, and your customers give you some cash. And there are way more ways for money to go out of your bank accounts, right? So everybody wants to know if, if we don't have any more revenue coming in for n months, when do we run out of money? What is that n? Right? And if you have that model, now you've got a steering wheel and you know some of the things you can do to push that out into the future.
And here's the key point: building resiliency into your business. If you look at companies that thrived in the pandemic and companies that didn't, it's often said that it was the digital companies that thrived. Well, not every digital company thrived. And there were some old-school companies that have done pretty well, too.
And so I would say it's not being digital that means you're going to thrive into the future. That may just be necessary. That may be table stakes, but it's having resiliency. Do you have multiple vendors in your supply chain? Do you have cash to take you through six months or nine months or twelve months in a downside scenario? Do you have lending facilities in place? Do you have options to take parts of your business and reconfigure it? So, organizations, for instance, cultural institutions are figuring this out. Maybe they can't perform to a live audience, but if they have some beloved venue, maybe they can perform on that stage, a little distanced, and stream those videos and come up with some business model for actually receiving revenues for that. So the resilience is the key.
Kenn: What about a stress test for individuals? For executives who are trying to sort of improve their skills amid lockdown? It's easy to say, “Oh, you should learn to be a coder.” But what really would you suggest?
Marty: Yeah, well again, personal resilience and stress testing is something that I recommend to everybody. On coding, Kenn, I was coding since I was a kid. I love it. I find it therapeutic, but I know that that's not for everybody. And also, I don't think that everybody needs to be a coder, but I would say everybody needs to understand the algorithmic, data-driven approach to solving problems. And learning a little bit of coding is a great way to understand that.
And it's almost like writing a grammatical English sentence, right? In your line of business, that's an important thing for you to be able to do. And you do that at scale, and you do that all the time. But most people don't need to be a prose craftsman, and a master the way you do.
But they all need to know what grammatical English sense looks like, and they need to be able to produce it, even if writing is not their profession. So I would say absolutely the same thing about coding. And a lot of people are asking me, friends mostly, sometimes young people looking for mentoring or advice, “What should I do in this time?” I had a friend, for instance, who was in med school, but the med school clinical rotations were stopped during the pandemic and there was really nothing for him to do. And he was just at home with his parents and said, “How about I go to Coursera and take some courses?”
And if you've got lots of free time, maybe take a certificate course in something like data science or machine learning. Why not learn another skill employers find valuable? And that's a way of building up resilience. You might be able to apply that to something you do in the future.
And then also personal resilience. I found for myself just 20 minutes a day of Headspace meditation - or you don't even need Headspace - you can just count to 10 in your head and sit quietly for 20 minutes. And you know, it feels like a waste of time when you're starting off this practice, but what I found is it makes me much more resilient and much more able to think under duress and in all kinds of states of the world. And so that's a recommendation as well.
Kenn: That's great. Sadly, we are out of time. So I'm going to have to let go on that note, but I think it's the perfect note to close on Marty Chavez. It's been a great conversation. Thank you very much.
Marty: Great pleasure. Thank you, Kenn.
Kenn: And I want to thank all of our listeners who have been watching this, and I want to thank DataFleets who have put us together for this conversation. I want to remind people that they can share this piece of content with all of their friends and their peers. And I hope that they join other DataFleets fireside chats. Thanks a lot.