MONEY REIMAGINED: The Dichotomy of AI | MIT Professor Sandy Pentland Examines Whether It Poses a Threat or Opportunity to Humanity

This episode of Money Reimagined is sponsored by CBO Digital.

You're listening to CoinDesk's Money Reimagined with Michael Casey and Sheila Warren.

Hello, and welcome to Money Reimagined. I'm Michael Casey. Unfortunately, we're doing a lot of this at the moment, but my co-host, who I always love to be with, Sheila, is She's just a very busy lady at the moment, so haven't been able to get her on the show. That's not to say we don't have a fantastic show. I hope, I'm pretty certain of it today because we have the one and only Alex, otherwise known as Sandy Pentland. I'll get to him in a moment. This is Money Reimagined. As I said, you can listen to us weekly on the CoinDesk Podcast Network or wherever you get your podcasts, and we would love to hear from you. So tell us what you think of this or any other episode. You can email us at podcast@coinDesk.com. Podcasts@coindesk.com. Make sure you use the subject line Money Reimagined. As I said, we are actually joined by Professor Alex, otherwise known as Sandy. That's who he'll be referred to from here on in. Sandy Pentland, who directs the MIT Connection Science at MIT's Media Lab. Sandy has been involved in data science for a very long time. One of the most noted and most oft-cited computational scientists in the world. I had the pleasure of working with Sandy when I was at the MIT Media Lab Digital Currency Initiative. And I just thought it would be really worth bringing him on today because Sandy has, I think, really been at the forefront of thinking through how do human beings deal with the ownership of their data and how we work through the challenges of the Web2 world with that. And now as we move into an AI age where he's also spent a lot of time investigating and quite at this very moment starting to confront some real issues, Sandy, around the governance of AI, it would just be great to have a conversation about it. So first of all, before we get into that, how are you? Thank you very much for joining us.

Doing fine. Glad to be here. It's a good topic, so let's do it.

All right, let's do it. So The news I think that's a good peg here is, of course, the fracas at OpenAI, which I think the mainstream media covered as if it were a kind of a coup, if you like, or an attempted coup from the board, then the survival of Sam Altman. And I think, as a person from the media, we all love blood sports. And so, there was some drama there that inevitably became part of the story. Like, is he in? Is he out? Yeah. What was it that had upset the board so much and how did they handle it and so forth? And clearly a big show of support for Altman amongst the employees kind of coming to his rescue. And it looks like there would've been a mutiny and they had to come back. So it all looked as if the board had overplayed their hand and ultimately had failed in whatever it was trying to do there. Some of the board members who'd voted against Sam Altman were removed, new ones were brought in. And there's sort of something of a new governance model. But the key point, as far as I'm concerned, is that OpenAI was created at the very beginning with a real acknowledgement of something that we're now talking about more and more. And that is that this rapid-moving machine learning, and particularly this type of generative AI, has the potential to evolve out of our control into AGI, artificial general intelligence. And that if it did, there may be a moment in which We humans get sort of superseded by it and that we fall victim to this thing and that there's real dangers involved. Right. We've been hearing a lot from these people. And therefore OpenAI was created with this structure where it was initially a nonprofit, that it would be free from the financial incentives of a typical profit, for-profit company, and therefore be able to engage in this research in a way without being compelled to make money. And that would be a good thing. And then they realized that they'd have to compete. And so They spun out a for-profit company, OpenAI Global, so they could raise money from venture capitalists and start to offer incentives to their own employees. And the original nonprofit board, that company became the major shareholder of this. And so there was still this structure of oversight, but it had already, it seemed to me, had been thrown straight into this sort of like now for-profit conflict. And a misalignment was starting to happen. I don't know what it was that upset the board. We've heard various stories about articles that were written and arguments with different people. Whatever it was, that structure didn't— it seems the real risk here is that didn't work, that if there is supposed to be some oversight, it didn't happen. And I've always thought that the biggest problem we face, not only with AI, but with Web2, is the sort of centralized control. of these platforms that really built upon the human labor of everyone else, the data, the labor, the content, and that there's a misalignment between this sort of very centralized control of that and the profit motives of the companies that run that. And somehow we've got to fix out that realignment. So I don't know, I feel as if whatever they tried to do at OpenAI just didn't work to solve that problem. And what we're left with is the problem. So before we get into what you see as the solution, is my read on that right? What's your takeaway from the drop-in?

Well, you're putting an awful lot There's a lot packed into the question there. They're good questions, they're right on. But I think the first thing to talk about is, you know, what's about the governance of AI, because that seems to be the core of the battle. And then the second thing is, well, how do you avoid centralization, which is a perfectly good— I mean, that's a really key question. But I think you want to take those as 2 topics, not as—

Let's do that. Let's start with that first one.

Yeah. My read on this, and, you know, like I've had Sam over to talk, and I know Some of the other people, is that this is a battle between people who think that AI poses an existential threat to humanity. And, you know, you can sort of see that, but, you know, there's many things that pose an existential threat to humanity, and there's a standard way to get around that, or to deal with it, let's just say. And that is that you audit things, right? You keep a continuous running audit. That's publicly available. We do this with money, right? Everybody has to have those quarterly reports. We report the taxes, you know, the audit firms come in and it doesn't always work, but it works pretty well. And what it also does is scare people enough that they don't do really outrageous things. And when they do, they get caught really pretty quickly, right? So how do you audit AI? Well, you just keep a track of what it does. And each Each time there's a decision, each time there's a way, you write it down and you ask, how does this compare to the sort of way we want it to do? So in other words, is this affecting Black people different than white people, women different than men? Is it saying a lot of stupid things? You know, you can keep track of that stuff. It's not even expensive. And then there ought to be the ability for people to sort of audit it. Right now, what you have to do is you have to have like lawsuits and discovery, and it's this You know, you have to invest a lot of money and time and dig it out of their, you know, cold hands to be able to answer, is this actually working correctly or not? And you just have to make that visible and auditable. That's the solution that we've had for almost everything that we've done. And the advantage of that is we don't know what sort of troubles AI can get into. crazy junk people are going to do with it. And so you can't write laws beforehand to control it. What you have to do is say, okay, we know about normal bad things. We know about harms. That's why we have all those laws. That's why we— okay. And we have to ask, well, is this AI causing harms? And that's an auditing question. So, you know, I work with the EU regulators, with the US people, with other sort of countries. You know, people are coming around to this idea. There are some things that are sort of obviously you don't want to do, but mostly we're in sort of new territory. And what you really ought to do is keep track of stuff and make that publicly available so that people can say, hey, wait a second, that's not fair, that's not right, as opposed to having to, you know, raise a $100 million fund to go sue the jokers to whatever. So the OpenAI thing was, in my view, sort of a bunch of people that were sort of AI doomers, and they just didn't want to see stuff happen at all. And Sam, who's maybe a little bit too forward, right, in what they're doing. And, you know, I'm happy to have people do things, but I want to know what they're doing, and I want to have that something that's guaranteed. We have the technology to do it.

There's a couple of concepts that I think immediately come to mind when you say audits and transparency. I mean, they were part of the OpenAI story, but less so. One is open source code, right?

I mean— Open source is a beginning, but only a beginning,

right?

Because it's what data went into it. And in fact, even then you don't understand. You have to actually look at what it actually does.

Right. So I think there's like, again, you're calling for transparency, which I love. I think that that's, you know, that again, I think I like this point that we don't have, we don't know what harms or good for that matter could come out of this. And so—

It's not just transparency, it's also accountability.

Right.

So when you see something bad, you can like nail the guy to the wall, right?

Yeah. So, but that again, it feels like a very different structure to what we currently have.

Yeah, no, no, but it's not a hard thing to do.

Okay.

And we ought to insist that they do that.

Pieces that need to go into it. Like when you talk about an audit, I mean, we in the crypto space immediately think of a blockchain and so forth.

That's a reasonable way to begin doing it. Yeah.

Well, let's talk about that because I feel like one of the massive challenges I would have thought, if not for a blockchain, is that there are so many inputs. The sources of that data are so varied given that they're basically hoovering up it all from the breadth of what's on the internet.

Well, so that's the first thing. We have a project that's a big open source project. Similar to sort of like Mozilla-type things called the Provenance Project, where what we've done is curate all the training data and put metadata on it about, you know, what's the quality, who owns it, what's the copyright. And so that people can now begin training these things with assurance that they're doing things that are legal and not stealing somebody's pajamas, right? Mm-hmm. Or that they're using stuff that was done by forced labor or what. I mean, there's lots of ways you can go wrong on the training. And so what we're doing is putting together this huge thing. It's really many, many pentabytes of data. It's enough to really begin doing things so that people can train things and keep track of what they trained on so that you can tell, right?

Does it need to go right to the source of that activity? You talked about forced labor, whatever it is, to be Are we talking about making sure that every bit of export that's happening, in this case data, is being generated in a way that can be tracked? And therefore, how do you go back to—

Well, so the Provenance Project is an open source, open— it's like Wikipedia or something like that, right? It's a cooperative thing. And what we're doing is we're Putting together all the data sources that have a verifiable provenance, right? So here's this thing, I can show you how I made it and you can use it. That's the sort of thing. So it's not like going in there and messing with every byte, right? Right. And then volunteers sort of look through it and we're building tools to guarantee that it is what it claims to be.

Hmm.

Right? So that's sort of to get things off the ground. But then the other thing that you need is you need this notion of, okay, so I'm less concerned with like chatbots and stuff like that. But what I am concerned with is when they produce outputs, right? And you know the category of person that's receiving these outputs, that you can ask, is it fair? Is it saying different things to men than women? Is it saying different things to these people versus, you know, is it making medical decisions? So there's a really good one to keep track of really tightly, right? Point is, is that, you know, with money, everybody has books. You keep track of what comes in, what goes out. You add 'em up, make sure they all add up. You have to produce reports about it, which can be automatic. Things like Quicken and stuff do that automagically. So that you know that nothing's wrong. And if you do that same type of thing with most AI things, you can do a lot. So for instance, some countries, like I was just in Singapore, Singapore requires that you validate your AI mechanism or program before you release it. So they have a sandbox where they go in and they test it. And they test all the things that you claim that it does. And it tests it against all the sort of harms that you would imagine, and you get a stamp of approval. And if it's a thing that's a little bit dangerous, then you might have to do that regularly.

Who gets to decide what's dangerous in this case? It's essentially the regulator in each of these countries. And so Singapore's version of dangerous might be something that's different from—

Sure, people are going to disagree.

Right.

The way that is handled in most countries, right? And I think Singapore is an example of this, is that there's a first cut that the government tries, but really it's based off of court cases and liability. So yeah, if what's happening looks a lot like a standard harm, you're in trouble.

Right.

Right? Okay.

Yeah, I mean, we do. That's right. We have an enormous amount of body of case law that tells us what is harm and what isn't, That's right.

Yeah. Yeah.

No, no, this is not new around the world. Right.

And all the technologies that we have from sort of chemicals to electricity to whatever, you know, have to avoid those harms.

Right. I think where it gets complicated in this is maybe perhaps it's overthought, but I understand where it comes from, is that people see it as a free speech issue. Right. Is there a point in which somebody's saying this is harmful speech when it, when it might just be free speech? Right.

So that's bringing in a different thing. So my attitude is not necessarily something that everybody will agree with, but I think there need to be both platforms that have verified identity. For instance, now most crypto, you have to like produce verified identity. It's not attached when you do transactions. It's anonymous still. But if something really bad happens, the cops can go find you.

It's a backup. Or, you know, yeah.

Yeah, and—

Break the glass opportunity.

Break the glass sort of thing. So, and different societies define the glass a little differently. And there can also be fora and other exchanges where it really is genuinely anonymous. And, you know, the question is gonna be for people, do you wanna have any sort of protections or not? If you're on a fully anonymous thing, you know, it's caveat emptor. If you're on something where there is glass that can be broken, then you have the protection that, you know, something really bad happens, you can get it fixed or—

I like the idea of having, you know, a certain amount of choice to individuals as to— but so long as they've got information about the risks involved, right? I also like this idea that you would have a voluntary proffering up of provenance information and metadata so that like, okay, I know that this is safe. I've got no idea what's in that, but at least I know this is safe. And you by default create almost like a market incentive to create—

Yeah, that's right. It's a question of what's your risk tolerance and so forth. The only thing that is a little sort of problematic there is that that puts a burden of understanding on individuals. You have to understand what's dangerous and what's not. And of course, con artists come up with all sorts of creative ways. Which gets to another thing we were going to talk about, which is data cooperatives or cooperatives, right?

I'm very keen to talk about that, but before we do, I want to get into one other thing. And then maybe I can find a segue out of this. But like when you say, you know, people won't necessarily know to be able to decide, they need information to understand what it is. I've always liked the fact that SSL certificates have that very simple thing now. Yeah. Just the lockbox at the top of your URL and you know, okay, this one I can trust. That one I can't, right? And that's become like ingrained in human behavior. And we just, we don't have to know how the actual SSL process works. I certainly don't. You probably do. But in any case, you know, I can trust it because this is a standard and I know that these standards are being used. And so I feel like that's kind of where things go.

No, no, that's a perfect example. And one of the things that we are advocating for and think ought to happen is those certificates ought to, they chain back, right? Where did they come from, right? And they can chain back to things like, is this a verified identity or not? Not who it is, just that you can count that this is not some AI bot spamming you, right?

Right.

So, it's like, you know, a real human or something like that. And currently, this is only like a trivial change, right? But platforms and providers don't fill in that slot at the moment. And if they did, then the SSL chain would Tell you, oh, this thing originated from a bot and that one didn't. Right.

Okay. Yeah.

Yep. That's provenance. It's like,

okay. Pretty valuable.

Yeah. Yeah. Pretty valuable. And that ought to be something that every browser, every other thing, sort of like you can set it. I don't want to see bots. Right.

And you're not going to catch all of them. Fakes. Right. It was, you know, there's those sorts of ideas.

Right. That's right. So you're not going to catch them all. Nothing's perfect. You know, money fraud, we've been trying to fight for millennia, still happens. But you can catch 90, 95% of it and you can, can reduce the risk of it.

Yeah, I think it's a great alternative way of looking at it.

Do you have a trusted partner for your crypto trading? CBOE Digital will introduce financially settled margin futures on Bitcoin and Ether January 11th, 2024, with physically delivered contracts to follow, listed and cleared on CBOE's US-regulated exchange and clearinghouse, and complemented by a liquid crypto spot market for greater ease and access. We invite you to learn more about this and all applicable risk disclosures at cboedigital.com/coindesk. That's cboedigital.com/coindesk.

How does blockchain work in this? So, you know, you talk about chaining this certificate going back to its source. Are we talking about a distributed ledger model? Is it a public blockchain? What is the structure that allows us to keep that track record with what is really a massive, potentially massive amount of data? pool of data that you're working with?

Well, I mean, I can tell you what I see people doing, okay? Like, we work with a number of different countries, and one of the things that they're most interested in is distributed health data, right? Because health data, really valuable. Everybody understands that, but it's highly private, and you don't want to concentrate it in various ways. And so you might think that that's a blockchain type of thing. It's really more of a Web3 type of a thing. But unfortunately, those words have been tainted in various ways. And so when I talk to the people, right, the national leaders and stuff, I tell them that it's a distributed federated system, okay? So everybody keeps their own ledger and then they have transactions that they record with other people. And so you can actually find out what happened. You could trace it back, but everybody's keeping a ledger of what their things is and what they got from other people.

It's like a series of—

And people like that. Right. People buy them.

Permissioned ledgers then?

Yeah. So this is for permissioned ledgers, right? You can make it for, in general, when you're talking about security, you're talking about highly valuable things, there's some level of permission that's involved. Absolutely.

Right. Okay. Certainly.

So like, for instance, with crypto exchanges again, right? You know, it may look synonymous, but underneath that there are actual credentials.

Well, certainly now because the external laws in place that have come into force, KYC requirements and so forth. And so, yeah, none of us can escape the The heavy hand of identity requirements and whether that's a good or bad thing.

Well, but it's not even that heavy-handed. I mean, come on, right? You know, so you do this once when you set up your account and then you're free.

Right. And there are also interesting ways to take that break the glass model you talk about where it gets more and more abstracted and there's a lot of anonymity that you can bring to bear to sort of—

Yeah.

I

mean, one of the things that we're looking at and have sort of pushed in various places is You know, most places have client-attorney privilege, okay? So that's like, you know, you can talk to your lawyer, right? And it doesn't end up in court except in extreme cases. Well, you know, you can use that sort of privilege to guarantee your anonymity. So it's still there. There's still, you know, identity, but it, you know, they have to bring out the big guns to be able to get there.

Deep door, you know, in a way, right? So let's move to data ownership and data collectors because Because I think to bring it back to AI and this sort of risk or this opportunity almost to humanity, it's a big large language model approach to things with training on massive datasets, right? It all seems to get abstracted away, but underneath it, there are all these human beings who have generated data and information and are putting it out there into the ether largely through social media. There's an interesting philosophical question. whether it's copyright or not, it's philosophical. How on earth, you know, we're being used for this in a way. Like, so how do we get to a world in which we're empowering people? As you know, you, you're actually, uh, a fellow for Project Liberty, which is, you know, founded by, um, Frank McCourt, who the cat is out of the bag 'cause CoinDesk ran an op-ed by him a couple of days ago in which it was mentioned that I'm writing a book with Frank. And so full disclosure, I'm very interested in this topic for the— because of a book I'm working on with Frank. Sure, sure. And you've been gracious to help out with that. But like, that's that core issue, data ownership, data control. Walk us through what that world could or should look like.

The key thing I think is to realize that data is not just this thing that is exhaust from what we do. This is a core means of production. It's like money or labor. You have to treat it seriously. You have to keep track of it. We don't like just throw money out on the streets, right? Unless we're crazy. We shouldn't be just throwing data out on there. And we've talked about a couple ways to do this through sort of distributed ledger technology. That's great. But you have to be, it gets really complicated. I mean, even like someone like yourself who's incredibly sophisticated would have a hard time knowing what's really the right thing to do.

My sophistication reaches only a certain point and then it doesn't get much higher.

Like everybody, right? You know, and it changes.

Yeah, very limited. In our capacities. Yeah.

And so, you know, in the case of other means of production like labor or capital, what we did is we started building cooperatives or associations, labor unions, cooperative banks, things like that. So we could bring our resources together to hire somebody whose full-time job is paying attention to it. And they act like a lawyer for us. They're a fiduciary and they help us manage it. And when you do that, then suddenly, you know, you can get sort of world-class service without giving up your rights. 'Cause, you know, your bank doesn't own your money, it merely manages your money. Same thing with, you know, okay.

So— But that very idea, by the way, that you have a right to that data is a really important starting point.

That's right, that's the key, right?

And that's really, that doesn't exist at this stage, really.

Well, it's begun.

Right, yes. But you've been trying to make this case, but it doesn't function like that very, it doesn't function As if we have our rights to that data, I think is the big question.

That's right. Well, the thing is that things like GDPR and the California law initialize that sort of notion of the right to your own data. People don't take it broadly enough. So like, the things I say should also be included there. It's not just my NFTs or cash. But those laws didn't operationalize it. They just said, oh, there's sort of a right to it, right? And so now the next battle is, is how do you actually operationalize that so that you get the value from that? And that's where ideas like co-ops come in. We did a legal interesting thing, Data Rights Protocol. The Consumer Reports, working with them, we just released a thing called Permission Slip, which weaponizes your rights to own your data using legal mechanisms so that it's not just an API call. Hmm. Not this big thing where you have to like call them up and sue them and, you know, so, so it's moving in that direction. It could move a lot faster, but, you know, these things take time.

So, so one thing I like about the data collective idea that you've got this, you know, is, is the sort of the power in numbers, the, the idea like, you know, because there's this whole conversation about whether or not you, you have much bargaining power with just your data. Uh, I, I personally think that that's a little bit of a ruse because It's constructed around, yes, you don't have much power with the individual data on a fully aggregated Web2 structure of what data value is, right? But in a world in which that data was all pushed back into the hands of individuals, there's all sorts of ways that I could imagine an individual could use their data constructively. That's right. Nonetheless, the idea that we could think through different structures of organization around this And then you mentioned to me sometime, like, this is actually— some people say this is like a 4th factor of production. We live in a world where data is so important. Walk us through that.

Well, so we've had these sort of revolutions in being able to think about ourselves with each other and with the environment. You know, probably the first one was the notion of land ownership, you know, so titles and things like that. And there's a lot of infrastructure around that, including a distributed ledger, right? I mean, we all have like, that's what that is. And it's a blockchain. The title that says, oh yeah, I bought this from Sam who bought it from Alice. That's a blockchain, right? It's just an old-fashioned version of it.

Old-fashioned version, yes.

You know, and then with labor, when that became commodified in the early 1800s, you know, we formed labor unions and we pushed back and we got labor laws. And none of this is perfect, but it's a lot better than it would've been. And the same thing happened with money. So it used to be that there was, like in the US, there were just a few big banks that sort of owned everything. There wasn't even a central government bank. And so agricultural cooperatives started their own banks.

Hmm.

And, you know, there's a lot of political battling and pushing around there, but what it did is it destroyed, it distributed power over money across—

Which is capital, like I suppose the 3rd factor of production. That's right.

And so now data is the same thing. And, you know, it's a means of production. And I'm not the only one to say so. I mean, I've been saying this for quite a while, but recently Secretary General Xi of China got up there and said, you know, data is a principal means of production like labor and capital. And if you think about that, you know, here's

the loudest

voice in Marxism, which is the battle between Right.

I kind of wish it had come out of the mouth of somebody else, but it's a very interesting—

Yeah, yeah, yeah, yeah, yeah. Well, no, but you ought to pay— if it came out of somebody else, you might say, oh, but coming out of his mouth—

Right. In the sense it matters and that it makes you think hard about what this is.

You really ought to think about that.

The means of production conversation is a very established way to think about, whether you're a Marxist or not, an interesting way to think about the structure of the economy and where power and how you regulate and control that. So the idea, so a data collective, like, so what would I do? I would like get together with some of my buddies. How do I form a data collective?

Well, the way collectives are always formed, I mean, for real, is that there's some people that are either very frightened about it and they volunteer some time or somebody who thinks that this would be a good way to You know, not get rich, but, you know, to make a living is helping people with this. And they set up something that's like setting up a bank or setting up a labor union or a food cooperative. These are all the same type of structures. They have a little bit of law. There's a lot of lawyers willing to help, and we've actually put together legal templates for doing this sort of thing. And, you know, the simplest version is just helping people understand what to do. Here's Here's the way you set all your settings and here's the way you do it. But there's much more sophisticated versions. So for instance, there's a company, just to mention this, there's a company in England called Privatar, which if you're a member of that cooperative, nobody ever gets to see your data. What they do is they stand between you and the person who wants your advertising data. And they pretend to be you. And that's legal because they have the legal right to represent you.

Okay.

So Facebook thinks it's seeing X, Y, and Z, but actually that's what your cooperative tells Facebook.

That's very interesting. They're legally obscuring things and—

That's right.

Able to maybe bargain on your behalf kind of almost.

Exactly. Legally obscuring things. And the trick for them, right, is that what they do is they take the people in the cooperative that are sort of similar and they maybe say, Here's the average person in that group. Facebook doesn't know this. So in other words, you don't get ads for things that you can't possibly use because what Facebook thinks about you is sort of right, but it's an average.

It knows you're going to split up with your partner apparently before you do, so it does know a lot

of things. Well, not anymore, it won't.

No, because you're not going to tell them.

So all that sort of stuff. That's really good, right? And so, you know, basically what you're doing is you're taking control of your data and the way you interact with these things. So you just may not tell some people things, right?

You mentioned something last time we spoke that I thought was very interesting as well about like how you would imagine— you could imagine some of these data collectors forming around geographic lines that there might be even—

I think that's the obvious thing to do. Yeah.

Which I think, I think what's really interesting about that is that like it, it puts a physical dimension to this, this, this very, this sort of very real problem we face about the physical lives we lead as humans and this sort of all-powerful digital domain that we're operating in. And if we were able to structure our relationship with that domain through something that is inherently defined by a physical geographic space. With all of the ancillary interests that go with that, my school, my, you know, whether or not my roads are being swept. Those issues may— yeah, I kind of— I don't quite know where I'm going with this.

There's 2 things that are really critical there. One is that if these co-ops were physically located, I mean, this is the way we do government. This is the way we do schools. It's the way we used to do banks and labor unions. Then they're, first of all, intrinsically distributed.

Right.

Okay, so you asked the distributed question. So these are gonna be like communities, small communities. And then the other thing that you mentioned, which is exactly correct, is that people who live together have a lot of shared interests. Okay? And that means the way their data is used is going to have a lot of similarities, not perfect similarities, but, you know, if you use the same drugstore and you have the same doctor and you have the same same corrupt politician and they don't plow your streets or whatever it is. You share all of that and you ought to be able to leverage that with the data, right?

This is fascinating because you think about where the incentives are now misaligned when somebody else has got a direct, all-knowing, hidden control over your data. That's right. And they— I mean, the thing that I think, just in the process of writing this book, would be We've been talking about how social media, as constructed by these big platforms with sort of secretive algorithms, corrals people into these environments where they often become toxic and divided. And the arguments around the struggle in Gaza at the moment, of course, is a classic example. It's almost impossible to find any interest, to find some sort of consensus, right? Because immediately people are dragged into these things. But if I'm signaling to the world that what I really care about most in the world right now is my potholes or this or that, and yes, I have an opinion about Gaza, but this is the thing that really matters. And then it's sort of like you're opting into a different level of conversation. One would hope that you don't get dragged in where everybody is assumed to have— That's right. a common interest in something that really is a long way away from them and really shouldn't possibly be dragged into this debate over this thing, right? Not just saying everybody shouldn't have opinions.

It emphasizes what we all want,

right? And I think this is the thing. We weaponize division because we can— we being these platforms— can exploit that misplaced interest.

And don't let them do that.

I

mean, if you look at your community, you share so many things With the people. You want good medical, you want good schools, you want safety, you know, and you as a community will have opinions about that and need to act and so forth. And that shouldn't be in the hands of somebody far away, somebody with different interests.

Or somebody feeding you an algorithm. You know, like it's— Yeah. Yeah,

no, I think this is— No, no, that's known as local governance. It's local community empowerment. And community empowerment starts with data. Mm-hmm. I mean, nowadays, it used to, you know, the caricature of everyone goes to the town hall and talks. Well, you know, things are a little more people and a little rusher and, you know, things like that. And so now it actually really starts with, you know, knowing about your community, knowing what's happening in your community, knowing, you know, what are the attitudes and views about the community. The schooling, the policing, et cetera. And then being able to have discussions about that. And that's, all those things are around ownership of data, sharing of data, discussions around data. And that ought to be something that is definitely like in the hands of the community.

I like the way you said town hall because I was also thinking of town square, right? These are these concepts that we've, through the the centuries have worked on as safe public spaces where local conversations happen without the manipulation. Of course, there's always political pressure and so forth, but we've often used the town square as an analogy for a place where people can have a discussion about what they care about. The idea that the internet would be this open, free town square-like environment where people could discuss things was, I think, part of the sort of romantic beginnings of what it was supposed to be. But then these massive centralized, you know, data controlling systems took charge of it and it ceased to be an open town square.

That's right. And so, I mean, way, way back in the day, the only people who had the internet were actually all very similar. They were academics who were part of universities and things like that. And they actually shared a great deal. They were pretty close to What you'd call a community. But as that grew, that, you know, the commonalities fell apart and people came in and exploited that. And what we need to do is go back to that sort of notion of community as a real thing and not Facebook friends or any of that sort of BS, right? Sort of things where we have commonality. And we saw so much of that, for instance, during COVID where communities had opinions and wanted to do things about, but were powerless because they didn't know what was happening to them. Right. They didn't have control of their data, and they didn't have the ability to discuss it and cause things to happen. So one of the things that happens if you control data, you can control what people do with it. And it's not necessarily a peaceful process, but at least you have the leverage to make your point, and you can back it up with facts. And that's transformative when it happens, right?

Now, we began talking about OpenAI and that issue, and you talked about how you thought that these sort of doomsayers were coming in. And I think it's interesting point I'd like to dig into a little bit further here because Vitalik Buterin just came out. I dunno if you saw a couple of days ago with a really very, I think very thoughtful blog post about, it was kind of an answer to Marc Andreessen and kind of an answer to the doomsayers, right? Andreessen put out his techno utopian thing and I thought, I actually, to be honest, took real offense to it because I think he was strawmanning things and suggesting that those of us who are critiquing the sort of Silicon Valley dominance of these platforms are all Luddites and anti-technology, which is really just not the case at all, that it is quite possible and quite reasonable to be pro-technology, pro-change, but to also point out that we need to have different structures, right? Sure. So I thought Vitalik did a very good job actually of trying to get away from the doomsayers and saying, no, I think we should actually lean into the embrace of Whether it is blockchain, Web3, AI, all of these very transformative things and let them do the things they can do, but at the same time, just be very constructive around how do we protect against human harm? Where do you put yourself in that? And I think it's pretty clear that you came out of the MIT Media Lab, which is like a symbol of the wonderful things technology can do. So yeah.

Well, so look, First of all, let me just sort of point out that all this stuff with AI is a lot older than you think. So around 1980, you began to get the first sort of enterprise systems. At the time, they were AI systems.

Yeah.

They were using machine learning, they were using algorithms, and what it did is it made it cheaper to manage things centrally. And as a consequence, all the community banks disappeared. You no longer have community management of schools. You no longer have clinics in communities. They're all centralized. So the AI that has been out there now for 40 years, right, has centralized our society in such a way that communities are left powerless.

Interesting.

They don't know their data because it's centralized and managed centrally. It's run by algorithms. I mean, you think it's not, it is. You know, we're there. I mean, don't think about, oh, the future. No, it's there. It's been there. And a lot of the problems that we have come from that centralization. And one of the key things is that when you're centralized, you don't have the ability to see what's happening to you. You don't have a voice about it. But the other thing is, is they tend to make rules that are uniform for everybody. Well, that works for the average person, but it doesn't work for anybody who's a little bit different. Okay? So a lot of the inequality and things we see, that comes from centralization.

Mm-hmm.

It comes in from making the rules match the mean, but not for your community or mine. So that's thing one to sort of say. And then thing two is, look, I think that the cure to that is you have to empower communities. You have to give them the ability to know their situation, to compare to other people in an honest way, a data-driven way so that there's evidence, right? And to make their voice made in that way. And so we need to have things that are institutions that support that. And that's where labor unions started. That's where community banks started. That's where all sorts of things started is that Communities taking their voice and weighing into the valley. So if you go back to Andreessen and so forth, right? So the big thing is that at some level, having this data lets you actually be effective.

Hmm.

I mean, we have a census so that we know where the people are. Nobody questions that that's good. Of course it's good. Otherwise, you wouldn't be able to allocate resources in the correct way, right? The reason we know about global warming is data, right? I mean, come on, folks. Now, I'm speaking as somebody who's on the board of directors for the UN's Global Partnership for Sustainable Development Data, but if you care about society, you have to be able to measure it in some way. Mm-hmm. Right? You know, what is the poverty rate? What is the number of people? How, you know, what is the level of pollution? What is the average temperature? If you can't measure those things, you can't make good decisions. You can't empower yourself. So I'm just arguing that people, communities ought to empower themselves to move maybe closer to that, you know, vision that Andreessen had, right? You know. But empowered in what way? That way they can keep track of it and old people are coming on.

Well, then you could argue, and I actually think it's just a truism, like if we just find a way to distribute power of that data, give people the tools they need to anonymize where possible, we could unlock so much more, right? I mean, think about all the medical data that we have that's all just tied up in all these liability concerns and HIPAA rules and so forth. And like if we actually put it into the hands of people to say, hey, I've got this genetic information that I think would be useful for the pursuit of a Parkinson's disease cure or whatever, we could unlock an enormous amount of information. A huge amount.

And we could discover problems in neighborhoods that are poisoning the kids or whatever, right? I mean,

so you're a data optimist, right? I think that's one way to put it. Like, you're a technology optimist, but you're really actually talking more than probably about how you have an optimistic view of how data could actually just change everything.

Again, it's like data is a primary means of production. I think communities should have money too, right? That sounds like a pretty good idea. They should have land. That sounds good. But they should also have data because that's how things get done correctly, right? And so they have to be empowered to do that.

So we're running a bit out of time. So I'm just gonna— just one thing that you said that just reminded me, and you talked about this sort of centralization forces that have happened as our algorithms have grown. And Preston McAfee, who was an economist and was, I think he's at Google, is a leading economist there. He once said something at an event that I thought was fascinating. I heard him say that, look, centralization from an engineer's perspective and a computer scientist's is often seen as a good thing in the sense that it's like very efficient.

More efficient.

Very efficient. You can get from A to B really quickly. But he said, if you put, if you just left everything into the hands of those engineers, They would very efficiently build the Soviet Union, right?

That's right. That's the whole idea of the Soviet Union.

It's like, magic control straight from the top. Everything gets done, planned and delivered. Look at that, right? And then it's a great way to describe things because it's hardly a utopia, right? And so like this idea that we need to throw friction into the system to bring forth new ideas, innovations, empowerment, everything else, And out of that create not so much efficiency, but just sort of an ever-growing pool of different creative things. It's just a different way of thinking about the power of decentralization.

So I have a bone to pick with that. I mean, what you said is good. The core thing is that efficiency depends on sort of what you're measuring. So if you have an aggregate uniform rule, it is more efficient on average. It's just wrong for most people. Okay?

Yes. What

is the thing about blockchains and distributed systems is now you can optimize all the local things and get something that's even better than the global rule. So this is for the very first time because we have these very fast computers and there's data about everything and we have things like blockchain and we have those other sorts of things, we could actually find solutions that are neighborhood-centric, but globally are much more efficient and much more robust than the centralized systems. So this is a realization, this is like a fundamental mathematical thing, right? Is that you can actually do better from an efficiency point of view with local. Right.

Well, that's— That's new.

That is new.

And it sounds, it's like, you know, because it throws out, if you're right, not even the critique. I mean, even people within who are supportive of the technology will say, yes, blockchains are inherently inefficient, but that's a good thing because we don't want decentralized power. You're saying that from this global perspective, actually the end result is something that's more efficient for everybody.

So it's like the health systems we're helping several countries build. They're a distributed federated thing. They don't all have blockchains, but they're distributed and they have distributed ledgers. So they're highly efficient in that sort of like, you know, no data mining. But what's really interesting is they, all the communities have better service. So the average is better too.

Okay.

Right. That's the key thing. And then you got to figure out how to make the computers do it in a sensible way. Right. But one of the things that has to be there is local control, distributed ledgers. All

right. I feel like we could go and unpack even more, but unfortunately, Sandy Pentland, we are out of time. So I'm just going to thank you for what was a new, thoroughly fascinating conversation. More to be had. Thank you so much for joining. Pleasure. And it was, uh, there's a lot more to do here. So thank you to you all for joining us as well for another edition of Money Reimagined. Uh, we'll be back next week. You can, of course, as we always mention, listen to us weekly on the CoinDesk Podcast Network or wherever you get your podcasts. And we would love to hear from you. So tell us what you think. Email us at podcast@coindesk.com, subject line Money Reimagined. Bye for now.