How Bots, Deepfakes and AI Agents Are Forcing a New Internet Identity Layer | Alex Blania on a16z

1. 0:00 – 4:07

   Introduction

   1. BHBen Horowitz0:00

      How do you prove somebody is human?

   2. ABAlex Blania0:02

      It is a surprisingly hard problem.

   3. BHBen Horowitz0:03

      I think that people are gonna start getting accused of being bots. [chuckles]

   4. ABAlex Blania0:08

      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. Like-

   5. BHBen Horowitz0:17

      The AIs are really good at programming humans, much better than humans are at programming AIs.

   6. ABAlex Blania0:22

      [chuckles] Absolutely. An AI will be able to have a GitHub account and will be able to post and also attest to five other AIs that these are in fact humans, and even though they're not. Honestly, if you don't take it serious now...

   7. ETErik Torenberg0:38

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

   8. ABAlex Blania0:40

      Thanks for having me.

   9. ETErik Torenberg0:42

      So proof of human is having a moment right now. Why don't you first give a b-a background for people who are unfamiliar, what is the moment that's happening, and h-how did we get here?

   10. BHBen Horowitz0:50

       Yeah, and what is proof, proof of human?

   11. ABAlex Blania0:53

       Proof of human, as the name suggests, is, you know, do you know if you interact with a human or, like, something else on the internet? And I actually think the, the kinds of questions that we're now asking is, are you inter-interacting with a human, uh, an agent on behalf of a human, or just an agent? Like, I think these are, like, roughly the three-

   12. BHBen Horowitz1:12

       Yeah

   13. ABAlex Blania1:12

       ... the three areas that we wanna split apart.

   14. BHBen Horowitz1:16

       Well, and, and describe a little bit the difference between just an agent and an agent acting on [chuckles] behalf of a human. How do you see that distinction?

   15. ABAlex Blania1:23

       Yeah. So, um, quickly explaining just the term proof of human-

   16. BHBen Horowitz1:28

       Mm-hmm

   17. ABAlex Blania1:28

       ... and I think what is hard about it, and then I will, I will explain how that fits into, into an agent on behalf of a human. So, um, what proof of human really means is that, uh, you know, every individual that interacts on a platform has only one, ideally one account or-

   18. BHBen Horowitz1:44

       Mm-hmm

   19. ABAlex Blania1:44

       ... you know, a limited number of accounts, and stays the owner of that account. Like that, that's, that's kind of the property that you're looking for.

   20. BHBen Horowitz1:51

       Mm-hmm.

   21. ABAlex Blania1:51

       So, like, you're looking for a, a initial verification, uh, that ideally should be, you know, something like anonymous or very-

   22. BHBen Horowitz1:59

       Mm-hmm

   23. ABAlex Blania1:59

       ... extremely privacy preserving, and then on-ongoing authentication that the same person remains in control of the account. Um, and then there is, like, some secondary properties that I think are good to have. But that actually tells you that the really hard thing is, is uniqueness. Like-

   24. BHBen Horowitz2:13

       Mm-hmm

   25. ABAlex Blania2:13

       ... like, what, what is happening on a platform like Twitter right now is that there's all these accounts, you know, all these, all these bots that are in the replies, um, that, you know, there's probably one human sitting somewhere and, and sending out ten thousands or, like, hundred thousands-

   26. BHBen Horowitz2:27

       Right

   27. ABAlex Blania2:27

       ... of AIs. And there's this catch-up game where, like, uh, you know, Twitter and X are trying to just find them and block probably millions a day of these.

   28. BHBen Horowitz2:38

       [chuckles] Which is what? Like, a, a one-hundredth of the, [chuckles] of the bots?

   29. ABAlex Blania2:43

       That, that's right. That's how it feels like. Um, and, and then agent on behalf of human, I think, like, how that will look like is, uh, you know, I as a-- Like, I think all of us will have agents. You know, it's unclear how that will look like. Is this gonna be-

   30. BHBen Horowitz2:58

       Mm-hmm
2. 4:07 – 9:05

   Three Big Ideas People Were Interested In

   1. ABAlex Blania4:07

      And, and so that makes it really, really hard because, uh, back then when we started the company, there were, like, roughly three big ideas that people were interested in. Um, one was this idea of a web of trust or, like, related ideas. So this idea that you, you look how someone behaves on the internet or did behave in the past. So, like, usually a combination of you have these certain number of accounts, uh, that you h- you know, you own since a couple of years, and then you post regularly or you comment regularly to GitHub. Like, these were the ki- the kinds of things that people were using. And then, let's say, all three of us have them, and then I attest also that, you know, I know you in the real world, and I attest to you that I know you in the real world, and that's how you would build a certain graph. And that was, like, a very hot idea back then for this. Um, but we disregarded it basically immediately because we assumed that, you know, eventually everything that is j-just digital and AI will be able to do as well.

   2. BHBen Horowitz5:09

      [chuckles] Yep.

   3. ABAlex Blania5:10

      Like [chuckles] so-

   4. BHBen Horowitz5:10

      Yeah. No, we're gonna-

   5. ABAlex Blania5:11

      So, like, an AI will be-

   6. BHBen Horowitz5:11

      We're there.

   7. ABAlex Blania5:12

      Yeah, exactly. So an AI will be able to have a GitHub account, and will be able to post and own an account, and, like, also attest to five other AIs that these are in fact humans, and even though they're not.

   8. BHBen Horowitz5:21

      [chuckles]

   9. ABAlex Blania5:21

      So, uh, so, you know, there was, there was area number one. Area number two was to just, you know, uh, use government IDs-

   10. BHBen Horowitz5:30

       Mm-hmm

   11. ABAlex Blania5:30

       ... for everything, which, uh, we just also immediately disregarded for a couple of reasons. One is that, you know, uh, I think, you know, it's strictly better if the government would not control such an infrastructure in terms-

   12. BHBen Horowitz5:43

       Mm-hmm

   13. ABAlex Blania5:43

       ... of free speech and actually breaking that apart, but then also-

   14. BHBen Horowitz5:47

       Right, you lose anonymity-

   15. ABAlex Blania5:49

       You-

   16. BHBen Horowitz5:49

       ... instantly, right?

   17. ABAlex Blania5:50

       You could hypothetically-

   18. BHBen Horowitz5:51

       Yeah

   19. ABAlex Blania5:51

       ... set up a system that maybe preserves it, but it's-

   20. BHBen Horowitz5:53

       Mm-hmm

   21. ABAlex Blania5:53

       ... very hard to do. And then the second thing is also, um, you know, the government id-identity system is just not built for that.

   22. BHBen Horowitz6:01

       Mm-hmm.

   23. ABAlex Blania6:01

       And, uh, and, and what is so hard about this problem is it's going to be a global problem.And so it doesn't really matter if, you know, one government maybe has the perfect infrastructure. For example, Singapore is like an example of a c- of a, of a, of a government that has, you know, perfect infrastructure all around.

   24. BHBen Horowitz6:18

       Mm-hmm.

   25. ABAlex Blania6:18

       But that barely doesn't, doesn't matter because, you know-

   26. BHBen Horowitz6:21

       Yeah

   27. ABAlex Blania6:21

       ... for example, I don't know, Meta is a global product with three billion users-

   28. BHBen Horowitz6:24

       Yeah

   29. ABAlex Blania6:24

       ... and with a lot of other countries and-

   30. BHBen Horowitz6:25

       [laughs] Yeah, S-Singapore has-
3. 9:05 – 15:21

   The Orb Verification Piece

   1. ABAlex Blania9:05

      you know, on the verification piece, um, that's, you know, we, we, we've went down, if you know World, you know that we've built this thing called an Orb. So-

   2. BHBen Horowitz9:12

      The Orb, yeah.

   3. ABAlex Blania9:13

      You know, it's, it's doing a lot of things to prevent these kinds of attacks. So it's, for example, it has multiple sensors in the, you know, electromagnetic spectrum-

   4. BHBen Horowitz9:22

      Mm-hmm

   5. ABAlex Blania9:22

      ... to just make sure that you cannot show a display to it, and it, and it would-

   6. BHBen Horowitz9:26

      Mm-hmm

   7. ABAlex Blania9:26

      ... recognize that. Um, so I think on that side, we've, you know, we've, we've got it handled. On the, on the consumer side, like, you know, that should then reauthenticate, it turns out to be much harder because, uh, you would need to trust the phone in some sense.

   8. BHBen Horowitz9:42

      Mm.

   9. ABAlex Blania9:43

      Uh, because l- what we actually do in that moment is when you verify with an Orb, we... Not only do we check, uh, your uniqueness in a fully anonymous and privacy-preserving way, and we should talk about that, but also we send to your phone a signed face image that you then can later use-

   10. BHBen Horowitz9:58

       Mm-hmm

   11. ABAlex Blania9:58

       ... to reauthenticate against it.

   12. BHBen Horowitz9:59

       Right.

   13. ABAlex Blania10:00

       Um, and, you know, with a new iPhone, you can have meaningful amount-

   14. BHBen Horowitz10:04

       Mm

   15. ABAlex Blania10:04

       ... of trust against that, but with old Android phones, basically not. And so-

   16. BHBen Horowitz10:08

       Oh, [laughs] yeah, yeah, yeah.

   17. ABAlex Blania10:09

       Yeah, you know, because, like, you can just, uh, you can just show a, a deepfake essentially, either through a display-

   18. BHBen Horowitz10:14

       Mm

   19. ABAlex Blania10:14

       ... or just directly injected in the camera stream. So, um, that's a problem. And so i-it's gonna be a mix of, uh, you know, if you have a new enough, let's say, iPhone or a general phone, um, then you can just reauthenticate against that-

   20. BHBen Horowitz10:28

       Mm-hmm

   21. ABAlex Blania10:28

       ... uh, picture that you took on verification. Otherwise, you would probably have to even go back to an Orb somewhat frequently, um-

   22. BHBen Horowitz10:34

       Mm

   23. ABAlex Blania10:34

       ... like, let's say a couple times a year if you just-

   24. BHBen Horowitz10:36

       Oh, I see

   25. ABAlex Blania10:36

       ... you d- you don't have the-

   26. BHBen Horowitz10:37

       Right, to reauthenticate.

   27. ABAlex Blania10:38

       Yeah, yeah. That's right.

   28. BHBen Horowitz10:39

       Interesting. And then, you know, one of the things, so one of the kind of incorrect criticisms of the approach early was, "Oh my God, they've got my eyeball." [laughs] Um, you know, now they're, you know, they, they somehow have, uh, access to my privacy, and they're gonna, you know, do all these things to me, and, and that's my access, and then they can, they, uh, World's coin can, um, impersonate me and all these kinds of things. But that's not the case. And, um, so that was also, like, a non-trivial engineering problem.

   29. ABAlex Blania11:13

       That was, that was very much non-trivial. Um, so actually, I think one point on iris that I think people don't appreciate enough, and that's a bet we took back then, but it was essentially that iris will turn out to be super normal as a, as a modality-

   30. BHBen Horowitz11:26

       Mm
4. 15:21 – 29:21

   Social Media Bots: PSYOPs and Propaganda

   1. ABAlex Blania15:21

      cool.

   2. BHBen Horowitz15:21

      You know, social media is one kind of vector of, you know, things that were annoying and are now becoming overwhelming in terms of just bots. You know, particularly with PSYOPs, propaganda, all these kinds of things. What are some of the other, um, you know, uses of bots that are gonna be kind of impossible to live with if we don't get the proof of human in the future?

   3. ABAlex Blania15:48

      Yeah, actually, I think the, the simple model I have for it is every moment on the internet, uh, that is primarily about humans interacting with each other, you know-

   4. BHBen Horowitz15:59

      Mm-hmm

   5. ABAlex Blania15:59

      ... or, or, or even indirectly interacting-

   6. BHBen Horowitz16:00

      Yeah

   7. ABAlex Blania16:00

      ... with each other. So, uh, you know, you can, you can start with simple ones like dating, you know?

   8. BHBen Horowitz16:05

      Yeah.

   9. ABAlex Blania16:05

      That really matters. [laughs]

   10. BHBen Horowitz16:07

       [laughs] Yeah.

   11. ABAlex Blania16:08

       That one is-- the other side is in fact a person. Um-

   12. BHBen Horowitz16:11

       Yeah, well, the... [laughs]

   13. ABAlex Blania16:12

       Got bad news for listeners.

   14. BHBen Horowitz16:13

       Well, a-a-and the person who you expect it to be.

   15. ABAlex Blania16:17

       Yeah, yeah.

   16. BHBen Horowitz16:18

       [laughs]

   17. ABAlex Blania16:18

       Yeah, we had these problems even before, uh-

   18. BHBen Horowitz16:19

       The whole catfish thing.

   19. ABAlex Blania16:20

       Yeah, exactly. Yeah, yeah. So that, that, uh, that's, that's an obvious one. Um, and, and so for example, Tinder's already using it for that reason.

   20. BHBen Horowitz16:27

       Yeah.

   21. ABAlex Blania16:27

       Um, I, I think-

   22. BHBen Horowitz16:29

       And what, what's the, uh, the Tinder use case, so?

   23. ABAlex Blania16:33

       So we started-

   24. BHBen Horowitz16:33

       Yeah

   25. ABAlex Blania16:34

       ... we, we started in Japan, uh, and [clears throat] like as, as a test, as a test market, and it's, it's essentially exactly what we just discussed. It is, um, if you verified with an Orb, you get a little badge that, you know, signals to other people that you are in fact a human, so it ha-high, has a high level of verification. Um, and then also, um, I don't think that's live yet, but what will come next is that you're actually the person you claim to be. So-

   26. BHBen Horowitz16:59

       Mm

   27. ABAlex Blania16:59

       ... meaning you have a World ID that is associated to the kind of profile pictures that you use. Um, so you just run a quick check-

   28. BHBen Horowitz17:06

       Mm-hmm

   29. ABAlex Blania17:07

       ... that, uh, this is all correct.

   30. BHBen Horowitz17:08

       Yeah.
5. 29:21 – 36:44

   We Had Proof of Personhood for the Longest Time

   1. ABAlex Blania29:21

      F- we had, actually, we had, we had proof of personhood for the longest time. It's even here in this, on this brief.

   2. ETErik Torenberg29:25

      Yeah. Yeah.

   3. ABAlex Blania29:26

      But then, uh, at some point we were like, "Shit, well, at some point AIs will have personhood too," so. [both laugh] Uh, so like, that's not gonna fly, so.

   4. BHBen Horowitz29:37

      Yeah, but they're not gonna have retinas for a long time.

   5. ABAlex Blania29:39

      [chuckles] Yeah. That's actually-

   6. BHBen Horowitz29:41

      Although that's coming eventually.

   7. ABAlex Blania29:42

      It was, it was actually really funny. It was like some of the, some of the OpenAI people, uh, that I met were like, "Man, Alex, this is gonna, this is gonna be so dark. Like, people will hate you for, like, not giving personhood to AIs." And I was like, "Jesus." [both laugh] All right, let's, let's, let's call it proof of human then. Um-

   8. ETErik Torenberg29:59

      That's funny

   9. ABAlex Blania30:00

      ... so that, that's how it changed. Um, but then actually, so then I would say like last year, so post-- Then there was like a big shift post ChatGPT. Like, people were now like... That was like the AI suddenly got real to people. And then actually I think... And so that's when people started talking to us. But still we're not like, you know, like it's a future problem. It's probably a couple of years out. Like, we don't really care about it. Let's stay in touch. Like, that was like the common response. And then, uh-You know, and well, but you also, you had a couple CEOs that really believed it and were, like, willing to take the long-term bet, um, to, to give them credit. But I think the second big shift was actually Claude Bots, uh, and Mode Book recently.

   10. SPSpeaker30:43

       Yeah. [chuckles]

   11. ABAlex Blania30:45

       Just because-

   12. SPSpeaker30:45

       Yeah. That, that kind of means, like, the, the cow is way out of the barn.

   13. ABAlex Blania30:51

       [laughs] Yeah. Yeah, and, and so, like, honestly, if you don't take it serious now-

   14. SPSpeaker30:56

       Yeah

   15. ABAlex Blania30:57

       ... then I think you just, you, you should get a different job or something.

   16. SPSpeaker30:59

       Yeah. There you go.

   17. ABAlex Blania31:01

       Like, you're not-

   18. SPSpeaker31:01

       Yeah

   19. ABAlex Blania31:01

       ... they're just not, not thinking about problems in the right way. Like, it's-- And so that's, that was, like, the moment when many, many people started reaching out, and now, now it feels like much more of an executional problem, not, not any more a-

   20. SPSpeaker31:12

       Market risk

   21. ABAlex Blania31:12

       ... like a market risk or like a thesis problem or s- like, like just a... And which is still a big fucking problem.

   22. SPSpeaker31:19

       Yeah.

   23. ABAlex Blania31:19

       It's like, how do you, how do [chuckles] how do you get fifty thousand devices out there? How, how do you make it cheap enough? How do you make it economic? H- like, you know, how do, how do you meet all these three things at the same time is still a very hard problem.

   24. SPSpeaker31:30

       How do you normalize the behavior, et cetera.

   25. ABAlex Blania31:31

       That's right.

   26. SPSpeaker31:32

       So people aren't weirded out in a Starbucks or something.

   27. ABAlex Blania31:34

       Although I, I think that's now gonna be-

   28. SPSpeaker31:37

       Totally get used to

   29. ABAlex Blania31:38

       ... I, just because I think people will hate the alternative so much.

   30. SPSpeaker31:42

       Yeah.
6. 36:44 – 40:09

   Next Year Go-to-Market Is Focused on the US

   1. ETErik Torenberg36:44

      okay, next year go-to-market is focused on the, on the US. Uh, say more about how, 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?

   2. ABAlex Blania36:56

      Basically, a month ago, we entered a very different phase as a project, where I do believe many of the platforms that we are now integrating with will really, you know, bring a lot of users to our platform, and that changes, you know, how you think about it entirely. Like, if you have a, if you have a platform of a, a billion users, um, sending users to you, then it's really just all about, like, how do you meet that demand? It's like, you know... And that's, that's, that's what we're now entering. And, and so, um, yeah, so I think the response is first. Um, I think you will see, and we're already working on it, but you will see a lot of really large platforms that you know integrate, uh, in the, in the near term future. I think that will, just to set expectations, I think that will be slow initially because it also should be, just to s- you know, to, to get... understand the product. It will be focused on certain geographies, like what we did with Tinder, where we started in Japan just to, you know, to, uh, to test the product and also to just normalize the concept. Uh, but that will happen. And then secondly, which is now becoming, like, one of the main priorities for me, is just how do you get this Orb distribution up? Which is, which is, you know, broadly speaking, there's a couple different dimensions to that. But one is, first of all, the product needs to work at scale, uh, you know, without supervision, which is, turns out to be much harder than you would think. It, you know, e-every engineering problem at scale turns out to be much more complicated than you would think, because, you know, fighting for 1% of improvement in quality is this clusterfuck of, you know, all these dependencies to come together. So that's, I think that's, like, one of the biggest engineering focuses right now. But then second, um, you need to find places to deploy them at. And, and the way to think about it is there are large scale distribution partnerships. That could be something like Walmart, you know, or if you, you know, if you're very ambitious, it could be something like Starbucks. Um, or it, it can just be you go to one of, you know, hip coffee shops, and you just, you just put it there. Or, you know, and then it, you could go, you could eventually even go to the DMV and just put it right there. So that's the problem we're currently trying to, trying to puzzle together. Um, and, you know, it's gonna be some, some of all of that. I think there's gonna be some large scale distribution partnerships, many one-off coffee shops. Oh, actually, one thing that we will, uh, we will launch soon, and the team is gonna hate that I'm saying this now, but, uh, it's gonna be Orb on demand. So-

   3. BHBen Horowitz39:32

      [laughs] Orb on demand.

   4. ABAlex Blania39:34

      Yeah. So in, so in the Bay Area. Just because actually it's such a, it's such a gnarly problem to, you know, to get an Orb to truly everyone.

   5. ETErik Torenberg39:42

      Yeah.

   6. ABAlex Blania39:42

      You know, it's like to, to get that, the CapEx is insane.

   7. ETErik Torenberg39:46

      Yeah.

   8. ABAlex Blania39:46

      So it's actually, it's actually much cheaper and easier to just put an Orb on a motorbike-

   9. BHBen Horowitz39:52

      [laughs]

   10. ABAlex Blania39:52

       ... and drive it to you. As, as, as crazy [laughs] as, as crazy as it sounds. So, like, in, in places like the Bay Area or New York, y- you will just be able to say like, "Yeah, I wanna verify now."

   11. ETErik Torenberg40:03

       Huh.

   12. ABAlex Blania40:04

       And 50 minutes later, there's, an Orb comes to, to your door-

   13. BHBen Horowitz40:07

       [laughs]

   14. ABAlex Blania40:07

       ... and you can, you can verify. And,

7. 40:09 – 41:56

   Different Levels of Verification

   1. ABAlex Blania40:09

      uh-

   2. BHBen Horowitz40:09

      Did you ever think about, uh, I don't know, this is probably a terrible idea, but, um, having kind of different levels, like we know you're a uni- unique human, or like, eh, this guy may be a unique human-

   3. ABAlex Blania40:22

      [laughs]

   4. BHBen Horowitz40:22

      ... 'cause he's done it on his iPhone, and it's not-

   5. ABAlex Blania40:24

      Gradations to it

   6. BHBen Horowitz40:24

      ... quite the, the same, but-

   7. ABAlex Blania40:26

      Yeah, yeah. We, we have that. So actually we, um, you know, ge- generally we just have the, you know, we have the principle of, you know, what- whatever could be useful for this problem-

   8. BHBen Horowitz40:35

      Yeah

   9. ABAlex Blania40:35

      ... we just build it.

   10. BHBen Horowitz40:37

       [laughs]

   11. ABAlex Blania40:38

       And, and, uh, a- and so we, we have something called face check that, that does that.

   12. BHBen Horowitz40:44

       Mm-hmm.

   13. ABAlex Blania40:44

       So it uses, it uses face, uh, from the camera. It still uses multi-party computation-

   14. BHBen Horowitz40:49

       Mm-hmm

   15. ABAlex Blania40:49

       ... what we've built for the entire system, so you're still anonymous.

   16. BHBen Horowitz40:52

       Mm-hmm.

   17. ABAlex Blania40:52

       Um, and, you know, it, of course, reaches way less accuracy. So, uh, you know, as a system, you will know something along the lines of, well, this is, you know, at least one person cannot create 100 accounts. Maybe it's just 10 or 20.

   18. BHBen Horowitz41:08

       Right.

   19. ABAlex Blania41:08

       So like, it's like a, at least it's some measure of rate limiting. Um, and I do think, just to set a disclaimer, I think with deepfakes and, you know, all this stuff, I think that will fundamentally break. So it's a-

   20. BHBen Horowitz41:20

       Mm-hmm. Temp solution

   21. ABAlex Blania41:21

       ... it's a, it's a, it's a temporary solution that I think can get us to scale. That's kind of how I think about it. Uh, we also actually use government IDs, uh, similarly, where like we, we use, uh, just the ones that have an NFC ID chip.

   22. BHBen Horowitz41:33

       Mm-hmm.

   23. ABAlex Blania41:34

       Um, and we use multi-party computation, so you remain anonymous, and platforms can choose to use that as well. Uh, but no one really did. It's just somehow they have this-

   24. BHBen Horowitz41:42

       Right

   25. ABAlex Blania41:42

       ... like, very negative stigma, which I think makes sense.

   26. BHBen Horowitz41:45

       Yeah.

   27. ABAlex Blania41:46

       Um, but yeah, ba- basically whatever could do it [laughs] we, we put it-

   28. BHBen Horowitz41:50

       Yeah. [laughs] By any means necessary.

   29. ABAlex Blania41:51

       That's right.

   30. BHBen Horowitz41:52

       Yeah. Yeah.

Episode duration: 42:11