机器人、深度伪造与 AI 智能体正迫使互联网建立新的身份层｜a16z 对话 Alex Blania

How do you prove somebody’s human? It is a surprisingly hard problem. I think people are going to start getting accused of being bots. What we currently see is less than 1% of what it will look like in probably a year or two.

The idea that AGI will lead to some very fundamental shift seems obvious.

AIs are really good at programming humans—much better than humans are at programming AIs.

Absolutely.

AI will be able to have a GitHub account, post, and attest to 5 other AIs that these are, in fact, humans, even though they’re not.

I say, if you don’t take it seriously now—

Alex, welcome to the podcast. Great to have you.

Thanks for having me.

Proof of human is having a moment right now. Why don’t you first give some background for people who are unfamiliar? What is the moment that’s happening, and how did we get here?

What is proof of human? Proof of human, as the name suggests, is: Do you know whether you’re interacting with a human or something else on the internet? I actually think that the kinds of questions we’re now asking are: Are you interacting with a human, an agent on behalf of a human, or just an agent? I think these are roughly the 3 areas that we want to split apart.

Describe a little bit the difference between just an agent and an agent acting on behalf of a human. How do you see that distinction?

I’ll quickly explain the term “proof of human” and what’s hard about it, and then I’ll explain how that fits into an agent on behalf of a human. What proof of humanity really means is that every individual who interacts on a platform has only one—ideally one—account, or a limited number of accounts, and remains the owner of that account. That’s the property you’re looking for.

You’re looking for an initial verification that ideally should be anonymous or extremely privacy-preserving, and then ongoing authentication that the same person remains in control of their account. There are some secondary properties that are good to have, but that tells you that the really hard thing is uniqueness.

What’s happening on a platform like Twitter right now is that there are all these accounts—all these bots in the replies. There’s probably 1 human sitting somewhere and sending out 10,000 or 100,000 AIs. There’s this catch-up game where Twitter and X are trying to find them and block probably millions a day of these.

Which is, like, 1/100th of the bots?

That’s right. That’s how it feels. And then, with an agent on behalf of a human, I think all of us will have agents. It’s unclear what that will look like. Is it going to be 1, or are there multiple ones, maybe with different tasks and even different types of characters?

I think it will then come down to: I approve a certain action of my agent. I give it certain rights. So, act on my behalf. Post to my X account. Post to my Instagram, for example.

But it’s my Instagram, and I’m a unique human that owns that.

That’s right. X or Instagram could decide whether that’s actually something they want as a platform.

Right. That’s how you could do it. That makes sense. So, how do you prove somebody’s human? It is a surprisingly hard problem.

Those agents are very clever. We started this company a couple of years ago, way before ChatGPT and before all of that. We took it as an assumption that eventually we would have AIs that both pass the Turing test—they can claim to be human, and you will not be able to tell them apart on the internet—and are highly agentic and just run around and do their own thing.

That makes it really hard, because back when we started the company, there were roughly 3 big ideas that people were interested in. One was the idea of a web of trust, or a related idea: You look at how someone behaves on the internet or has behaved in the past.

Usually, it was a combination of having a certain number of accounts that you’ve owned for a couple of years and then posting or commenting regularly on GitHub. Those were the kinds of things people were using. Let’s say all 3 of us have those accounts, and then I attest that I know you in the real world. That’s how you would build a certain graph.

That was a very hot idea back then, but we disregarded it basically immediately because we assumed that eventually everything that’s digital and that an AI can do, an AI will be able to do as well.

We’re there.

Exactly. An AI will be able to have a GitHub account, post, own an account, and attest to 5 other AIs that these are, in fact, humans, even though they’re not.

Area number 2 was to just use government IDs for everything, which we also immediately disregarded for a couple of reasons. I think it’s strictly better if the government would not control such an infrastructure in terms of free speech and actually breaking that apart.

Right. You lose anonymity instantly, right?

You could hypothetically set up a system that maybe preserves it, but it’s very hard to do. The other thing is that the government ID identity system is just not built for that.

What’s so hard about this problem is that it’s going to be a global problem. It doesn’t really matter if 1 government has the perfect infrastructure. For example, Singapore is an example of a government that has perfect infrastructure all around.

But that barely matters, because Meta is a global product with 3 billion users, and there are a lot of other countries.

Singapore is what, like, 2 million people or a million people?

Yeah. So, do you want to lock everyone else out? There’s a long list of other reasons why we disregarded that basically immediately.

The last one is biometrics, which immediately gives us this ick reaction. It even went further, because what’s so hard about this problem, as I mentioned in the beginning, is uniqueness.

In very simple words, you can describe the problem like this: What does Face ID do? Face ID checks that I’m the same person again when using my phone. It’s a one-to-one authentication. There’s an embedding stored on my phone; it takes a picture of my face, creates a new picture, compares it to the previous one, and if it’s close enough, I can use my phone.

That’s one-to-one: 1 embedding to 1 new embedding. To solve the proof-of-human problem, you need to distinguish 1 new individual from all previous individuals. You need to make sure that Ben is trying to sign up and that Ben did not sign up before.

Suddenly, it goes from one-to-one to one-to-N, and N is the size of your network, essentially, that you’re trying to prove it to. You can do the math and calculate how much mathematical entropy—how much information, just information-theoretically—you need to prove that.

It turns out to be a pretty high number because it’s an exponential problem. You can do the math and find out that things like faces or even fingerprints don’t work. You would basically hit a wall after tens of millions of users.

Right. And so then you end up with something like the iris, which is the muscle of your eye, that actually has enough entropy.

That’s unique enough. You also have to solve the problem that biometrics have historically been subject to replay attacks. I may not have your eyeball, but I’ve got enough information that I can run a replay attack on you.

It’s important to split up the problem into verification, which is essentially, in old terms, like getting your passport, and authentication, which is you showing your passport constantly for certain kinds of things.

On the verification piece, if you know Worldcoin, you know that we’ve built this thing called an Orb. It’s doing a lot of things to prevent these kinds of attacks. For example, it has multiple sensors in the electromagnetic spectrum to make sure that you can’t show it a display and have it recognize that.

On that side, we’ve got it handled. On the consumer side, where it should then reauthenticate, it turns out to be much harder, because you would need to trust the phone in some sense.

Mhm. Because what we actually do in that moment is, when you verify with an Orb, we not only check your uniqueness in a fully anonymous and privacy-preserving way—and we should talk about that—but we also send to your phone a signed face image that you can later use to reauthenticate against it. With a new iPhone, you can have a meaningful amount of trust against that, but with old Android phones, basically not.

Oh, yeah, yeah, yeah.

Because you can just show a deepfake, essentially, either through a display or directly inject it into the camera stream. So that’s the problem. It’s going to be a mix of: if you have a new enough iPhone or a newer phone in general, then you can just reauthenticate against the picture that you took on verification. Otherwise, you would probably have to go back to an Orb somewhat frequently, let’s say a couple of times a year.

I see. Right, to reauthenticate.

Yeah, that’s right.

Interesting. And then one of the incorrect criticisms of the approach early was, “Oh my God, they’ve got my eyeball.” Now they somehow have access to my privacy, and they’re going to do all these things to me. They have my iris, and then Worldcoin can impersonate me, and all these kinds of things. But that’s not the case, and that was also a nontrivial engineering problem.

It was very much nontrivial. Actually, I think one point about the iris that people don’t appreciate enough is that it was a bet we took back then: iris will turn out to be super normal as a modality, just because I think we will all wear AR and VR systems that do that. Apple already does it.

Yep.

Apple already has iris ID in the Vision Pro. So I think it’s going to become something that we use across many different devices and normalize in that sense.

So maybe that’s a general point.

But on the privacy piece, that took us a lot of time because, when we decided back then—with our assumptions, which was 6 years ago—that we would need a custom hardware device for biometrics, it was actually quite scary to come to that conclusion.

That’s an expensive conclusion.

It’s very expensive. Then there was the idea that you would need to distribute them all over the world. That just assumes that you would be able to somehow bring up billions of dollars and undertake a massive effort to roll this out across the world. But then there was also the privacy challenge: how could you build such a system with all the requirements that we care about? The 2 main high-level ideas for how to solve it were multi-party computation and zero-knowledge proofs.

Mhm.

Again, what is different from Face ID? Face ID can be very private because the embedding is stored on the phone. It doesn’t have to leave the phone ever, because it’s just you against you in the past. But to check uniqueness, you need to check against all previous people, so something needs to leave.

Yeah, something needs to leave and be compared to someone else.

And that’s a much harder challenge. How we approach that is with multi-party computation. In our case, when you verify with an Orb, we take all these pictures. They get computed on the device, and then they actually get split up into multiple pieces. For example, we take a picture of your iris, calculate an iris code, then break that iris code into multiple pieces and send it to multiple computers, such that there is no central database. No one actually has the information about you.

Right.

Then you do some clever tricks for how these different parties need to come together to do a computation that still leaves the pieces apart.

Right, right, right.

Yeah, so no one has the whole thing, and also during the computation no one has the whole thing. They do some clever interactions to come to the conclusion—

Like a zero-knowledge-proof kind of technique.

It’s very different, but in terms of the properties it achieves, it’s somewhat similar: no one knows anything about you, but you can together make a statement about you. You send it to this multi-party computation, and what comes back is, “Yes, that individual is unique.”

And the second thing we do is separate all of this from you with a zero-knowledge proof. Meaning, you have the secret on your phone, but no one else has it. No server has it; we don’t have it. Then you can later go back to this multi-party computation and say, “Hey, I have a secret that is part of that computation, and I am in fact unique.” Then you can prove that to a platform. You could go to a social network and prove that you’re a unique user to the social platform without us knowing anything about you or the social network knowing anything about you.

It’s just very counterintuitive that even though it uses biometrics, you preserve anonymity and extreme levels of privacy, which I think is super cool.

Social media is one kind of vector of things that were annoying and are now becoming overwhelming in terms of just bots, particularly with psyops, propaganda, and all these kinds of things. What are some of the other uses of bots that are going to be impossible to live with if we don’t get to proof of human in the future?

Actually, I think the simple model I have for it is that every moment on the internet that is primarily about humans interacting with each other—or even indirectly interacting with each other—is affected. You can start with simple ones like dating. It really matters whether the other side is in fact a person.

Yep. What if the other side is in fact a person?

And the person you expected them to be.

Well, I’ve got bad news for listeners.

We had this problem even before the whole catfishing thing.

Yeah, exactly.

That’s an obvious one. Tinder is already using it for that reason.

What’s the Tinder use case?

We started in Japan as a test market, and it’s essentially exactly what we just discussed. If you’ve verified with an Orb, you get a little badge that signals to other people that you are in fact a human. So it has a high level of verification.

What will come next is that you’re actually the person you claim to be. Meaning, you have a World ID that is associated with the profile pictures that you use, so you just run a quick check that this is all correct. You then know you’re not interacting with a bot, but also that you’re interacting with a fully authentic profile.

Yeah. Another fun one, because I think it’s somewhat counterintuitive, but I think it will be video conferencing.

Mhm. Because you already have deepfakes.

Yeah, I just don’t feel like going to this video conference with my deepfake on.

Actually, you raised it to me first, and that’s why we started building a product for it. It will actually start with very high-value users—for example, people like yourself who maybe manage a fund, where sometimes calls could be very high-value if they’re about borrowing money or—

Oh, yeah, yeah. Somebody can— It’s still slightly hypothetical because these things are not fully real time, and you can somehow—

They’re very close.

But they’re very close. And so I think, in a year from now, it’s just going to be a full commodity. It’s going to be super photorealistic and absolutely real time, and you will just not know anything anymore on these video calls. So I think that’s another one.

I think another one will be gaming. It’s fun, but it’s going to be gaming.

Yeah, because gamers really care—

Oh, yeah, yeah. Because gamers really care that they’re not playing an AI.

Holy cow, that’s frustrating. Especially if we bet money.

Yeah, exactly. You lose money, you train multiple hours a day to get really good at this thing, and then suddenly you get destroyed by an AI that is superhuman in every dimension.

Funny enough, I wonder what you think about this, because I don’t have a good mental model for it. Even the whole model for video platforms, I think, is about to break. There are a couple of dimensions to the problem, but one is that if the creation of content is becoming super scalable. For example, I heard about this one guy who created, I think, on the order of 100 videos a day on YouTube and made tens of thousands of dollars a month. All of them were fully AI-generated.

Yeah.

People just fell for it. So now the question is: is that actually something that YouTube wants to monetize that way?

Yeah, well, it’s interesting, right? They fell for it.

But maybe they liked it. They’re like, “That could be.” But it would sure be nice to know, “Okay, this is a human video, or this is an AI video.” Actually, my thesis about this is something along the lines of: I think there are categories of content that are clearly just fictional. Movies are that, you know? You don’t care that there’s any connection to reality. It’s just a fully fictional story. But if you think about something like TikTok, or all these kinds of things, people actually really care about them mostly because there is some connection to reality.

Yeah. Well, there’s reality and there’s a connection to a human, right? You can create a pretty good podcast—you can take a scientific paper and give it to Gemini and say, “Make this into a podcast,” and it’ll be a pretty entertaining podcast. It will be reality in that it came from some real thing, but you would like to know that.

You would like to know that. Yeah, I would like to know that. As an advertiser, you’d like to know: did a human watch it?

Or did an AI watch it? [laughter]

Yes, right. That’s the other thing: I created 100 AI videos, and I had 1 million AIs watch them. [laughter]

And then I made a lot of money off YouTube. Exactly. I actually saw that video today of a YouTube farm where there are thousands of phones that just watch videos all day for some reason.

Yeah, and that has zero value to YouTube advertisers. That’s actually a real problem for them. The whole creator-economy platform of the last decade—Substack, Spotify, and all the people who support artists, Patreon and other creators, YouTubers—they have a personal relationship with these people. It’s not just that they like the art. If they suddenly found out that they were bots, they might not want to support them in the same way.

You might not want to give them a big YouTube tip.

Yeah, I think there’s a certain subset of people who want to support actual people and feel like they’re having a real relationship.

Yeah. The thing that I think people don’t really get is that this should be obvious, but I don’t think people really understand the consequence of it. What we currently experience is a super, super tiny glimpse of what’s about to happen.

Yeah, right. It’s a glimpse.

It’s a glimpse. The cost of intelligence is dropping almost exponentially, and agentic capabilities are increasing in some superlinear form. We currently see less than 1% of what it will look like in probably a year or 2. And, second, these things will actually be superhuman in many ways. They’ll be perfectly able to understand you and talk to you in exactly the right way.

AIs are really good at programming humans. That’s much better than humans are at programming AIs.

Absolutely. There’s no question. I think that’s going to get quite scary, also. But at least if you know you’re the victim of a psyop, or that it’s a very advanced one done by an AI, that would be extremely useful to understand. There was one paper that I think you’ve got to read. It was about the Change My Mind subreddit, where the University of Zurich did this thing where they had AIs actually interact with Change My Mind.

Yeah.

They were superhuman in their ability to change minds because they went back to the profiles of the people posting and understood their political motivations, the way they talked, and then interacted in exactly the right way. [laughter] They just hit all the buttons.

Totally. Talk a little bit more about the state of the product and the business today. How many IDs are out there? Do you want to give us a little bit of an update and maybe talk about the evolution as well?

First of all, it’s a multisided problem. I think there are roughly 3 things that you have to consider. One is that you need platforms to use the technology—things like Reddit, X, or things like that. You need distribution of these devices, and I think the right mental model for it is: how many minutes does it take a person to reach such a device, on average?

Currently, if you took the global average, it would be a terrible number. It would be days or something, because many people would need to fly. How do we get that down to below 15 minutes across the US? That’s probably roughly 50,000 devices that you need to deploy. It’s not crazy, but it’s also not nothing. It’s hard to do.

The last one is: how does all of that come together into something that a lot of people really want to use? That’s a combination of the utility of all the subplatforms, essentially, but all of that layers on top. Maybe you can use a new Reddit account. Maybe you get a certain amount of a ChatGPT subscription for free. I think it’s going to be a combination of things, but you need to land all 3 at some point at the same time, which is hard to do.

We’re now at 18 million verified users and 40 million in total in the app. The biggest thing is that, because of the past administration and because we use crypto, we didn’t really invest in the US for a long time. That’s now the main shift that we’re going through.

Hopefully, we get the Clarity Act passed shortly.

Yeah, exactly. It would be really great to get clarity on that.

The big focus that we’re going through right now is to go all in on the US. Over the next year, 90% of the company’s effort is going to go toward the US. How do you get device distribution up? How do you eventually have this on every Starbucks, so it becomes super normal and people just use it every day?

On the platform side, we went through a very interesting experience personally. A couple of years ago, universally, people just made fun of us. That was the universal reaction, apart from a couple of other people who believed in it. In the press, the amount of fun being made of it just showed how shortsighted people are.

That’s right.

It’s like, you don’t think the bots are coming? What did you think when we first pitched it, actually? Even you must have thought, “This is crazy.”

Well, because you had the Orb. The Orb was so wild. “Okay, we’re going to scan people’s retinas, and that’s how we’re going to know they’re human,” and so forth. You pitched us 6.5 years ago—6 years ago.

Yeah, it was before COVID, because you were there with the Orb, right? AI just hadn’t happened yet.

But you could kind of see that there were bots. They were very crude compared to what they are now, but it seemed inevitable. At least at the time, it was so far from the future that we always worry about the timing of these things. But you were impressive enough, it was going to happen eventually, and it was an exciting enough idea that all those things got us to say, “Okay, we’re in.”

It wasn’t obvious that it was going to work in that timeframe. It seemed very unobvious for a long time.

How different was that pitch from what it ended up being? Talk a little bit about it.

It was actually pretty much exactly the same pitch. I think it’s the same thing. The device changed; they made it much more economical and convenient. But the initial instinct was right. Basically, everybody is going to have to prove they’re human. You’re either going to have to have some proof that you’re human in cyberspace, or it’s going to be a very bad world.

Yeah. I mean, the robots are going to get us. We’re done.

Right. And the second point was that when it becomes a big deal, we’ll be able to build one of the most valuable networks as a result. In a world of AI, having a human network is going to be incredibly important. You’ll need to prove that you’re human, but it will also have very strong network effects.

Even as you get into platforms, one of the platforms’ largest problems has been bots. You remember Elon backed out of buying Twitter because all the stats were based on bots.

Still, even knowing that, it was hard for them to get all the way to the future in their thinking and go, “Yeah, we need proof of human.”

Yeah, it’s kind of obvious.

People were like, “What does it even mean? What does proof of human even mean?” Did you have the language? When did you come up with the language “proof of human”?

We actually had “proof of personhood” for the longest time. It’s even here in this brief. But then, at some point, we were like, “Well, at some point, AIs will have personhood, too.” So that’s not going to fly.

Yeah, but they’re not going to have retinas for a long time.

That’s actually—oh, that’s coming eventually. It was really funny. Some of the OpenAI people I met were like, “Man, Alex, this is going to be so dark. People will hate you for not giving personhood to AIs.” And I was like, “Jesus.”

Let’s call it Proof of Human, then.

That’s funny. So that’s how it changed. I would say last year, post-ChatGPT, there was a big shift. That was when AI suddenly got real to people. That’s when people started talking to us, but it was still, “It’s a future problem. It’s probably a couple of years out. We don’t really care about it. Let’s stay in touch.” That was the common response.

You also had a couple of CEOs who really believed that and were willing to take the long-term bet, to give them credit. But I think the second big shift was actually Cloud Bots and Mobile Book recently.

Yeah.

That kind of means the cow is way out of the barn. Honestly, if you don’t take it seriously now, then I think you should get a different job or something. You’re just not thinking about problems in the right way. That was the moment when many, many people started reaching out.

Now it feels much more like an execution problem, not a market risk anymore—not a market-risk or thesis problem. It’s just: How do you get 50,000 devices out there? How do you make it cheap enough? How do you make it economic? How do you make all 3 of these things work at the same time?

It’s still a very hard problem. How do you normalize the behavior so people aren’t weirded out in a Starbucks or something? Although, I think people are going to get used to that. I’m saying that because I think people will hate the alternative so much.

Yeah.

I think people are going to take a lot more pride in being human, by the way, particularly online.

I think people are going to start getting accused of being bots.

Totally. It’s going to get really weird. Without a clear delineation, it’s going to be a mess. I don’t understand how somebody can think they’re going to have a social media platform that doesn’t distinguish between humans and bots. That seems absurd to me.

It’s absurd.

My guess is that over the next 2 months, we’ll see these platforms trying to use things like facial biometrics on the phone. I know it will break, so it’s fine, but I think we’ll go through that cycle now.

We just need to get to scale fast enough to meet the market for what comes after, which I think means something like the Orb is the only solution. Currently, there’s no real competition. I have not seen a competitor yet, and I think we’ll also see that.

It’s so ridiculous. It’s so ridiculous, and it is so hard to get to in terms of building it. Then there’s a massive network effect, which means people are starting 6 years behind you on that. But I’m sure they’ll come, because it’s such an obvious problem now.

What actually do you think about, as AI continues? What, in your mind, are the economic policies that we will need to implement, or directionally?

I think governments do have to figure out how to send citizens money. They’re good at taking money from citizens, but not the reverse. If you go back to COVID and the stimulus program, I think $400 billion was stolen. You would have liked to know that you were sending the money to unique humans—even if they weren’t citizens. As long as they were unique humans, that would have been good.

Yeah, I mean, the Social Security system, for example, is a mess in the U.S. It’s a total disaster.

We’re going to have to get to some kind of cryptographically strong way to identify who’s a citizen of what country. That’s going to be a really bad problem, I think. Otherwise, there’s no way to even have a democracy.

It’s pretty crude what they’re trying to do with the SAVE Act, but it’s not completely insane. How do you even know that the people who are voting are actual people, or living people, or anything? We genuinely don’t know now.

The whole mail-in ballot thing is built for a very different world. I don’t think that, in an AI world where you can have high-scale impersonation, combined with a broken Social Security system, you’re going to have the will of the people anymore. I think that’s going to be gone pretty fast.

We’re going to need some kind of cryptographically strong infrastructure for who’s who. Similarly, I think we’re going to have to be able to get people money much more efficiently than through this crazy apparatus of social programs that we have, just because of how lossy and fraudulent Social Security, Medicare, and all of these things are.

Medicare is so frustrating for people that they shot the CEO of UnitedHealthcare in mail, and people are happy about that—really happy. Think about how bad a system that is. The government spends a lot of money sending you money for your health care, but it does it in a super-inefficient way.

We have the technology to do that now. I think AI is going to make that problem so bad because of the ability to file fraudulent claims and create fake, you know, buy social. I mean, you can buy Social Security numbers on the black market. For those who don’t know, that’s an easy thing. That’s a real thing. Everybody’s Social Security number is for sale.

I agree with that.

I think proof of human is a piece of a very important puzzle where we have to upgrade the entire infrastructure, or we’re not going to be a democracy anymore. That would be my guess.

You said, “Okay, next year, go to market, focus on the U.S.” Say more about how you’re thinking about that. Is the incentive for people to do it because they get to use a set of services? Is there some other economic incentive, or how do you envision it?

Basically, a month ago, we entered a very different phase as a project. I do believe many of the platforms that we’re now integrating with will bring a lot of users to our platform, and that changes how you think about it entirely. If you have a platform with a billion users sending users to you, then it’s really all about how you meet that demand. That’s what we’re now entering.

First, you will see—and we’re already working on it—a lot of really large platforms integrate in the near term. To set expectations, I think it will be slow initially, because it should be, just to understand the product. It will be focused on certain geographies. With Tinder, we started in Japan just to test the product and normalize the concept, but that will happen.

Secondly, one of my main priorities now is: How do you get Orb distribution up? Broadly speaking, there are a couple of different dimensions to that. First of all, the product needs to work at scale without supervision, which turns out to be much harder than you would think.

Every engineering problem at scale turns out to be much more complicated than you would think, because fighting for 1% of improvement in quality involves all these dependencies that come together. That’s one of the biggest engineering focuses right now.

But then, second, you need to find places to deploy them. The way to think about it is that there are large-scale distribution partnerships. That could be something like Walmart, or, if you’re very ambitious, something like Starbucks. Or it can just be one-off hip coffee shops, where you put it there. Eventually, you could even go to the DMV and put it right there.

So, that's the problem we're currently trying to piece together. It's going to be some of all of that. I think there's going to be some large-scale distribution partnerships and many one-off coffee shops.

Well, actually, one thing that we will launch soon—and the team is going to hate that I'm saying this now—but it's going to be Orb on Demand.

Sounds good. [Laughter] Orb on Demand. Yeah, send it there.

It's just because it's such a gnarly problem to get an Orb to truly everyone. The capex is insane. So, it's actually much cheaper and easier to put an Orb on a motorbike and drive it to you, as crazy as it sounds. In places like the Bay Area or New York, you will just be able to say, “Yeah, I want to verify now.” And 50 minutes later, an Orb comes to you at work and you can verify.

Wow. Did you ever think about having different levels? Like, we know you're a unique human, or, hey, this guy may be a unique human because he's done it on his iPhone and it's not quite the same?

Yeah, yeah, we have that. Generally, we have the principle that whatever could be useful for this problem, we just build it. We have something called Face Check that does that. It uses facial data from the camera. It still uses multiparty computation that we've built for the entire system, so you're still anonymous.

It of course reaches way less accuracy. As a system, you'll know something along the lines of, “Well, at least one person cannot create 100 accounts.” Maybe it's just 10 or 20, so it's at least some measure of rate limiting. I do think, just as a disclaimer, that with deepfakes and all this stuff, that will fundamentally break.

So, it's a temporary solution that I think can get us to scale. That's kind of how I think about it. We also use government IDs similarly, but just the ones that have an NFC ID chip. We use multiparty computation, so you remain anonymous, and platforms can choose to use that as well. But no one really did. Some of them have a very negative stigma, which I think makes sense.

Yeah.

But yeah, basically, whatever could do it, by any means necessary.

That's right. Yeah. I don't know. Well, thanks so much for coming on the podcast. It's been great.

Yeah. Thank you. Thank you. Thanks for having me.