The a16z Show - Expert AI as a Healthcare Superpower

Episode Date: January 10, 2023

In this episode, Marc Andreessen and Vijay Pande discuss expert AI and its role in healthcare, bio, and more. Watch on Youtube: https://youtu.be/c7ScUDYSRYoSubscribe to Bio Eats World: https://podcasts.apple.com/us/podcast/bio-eats-world/id1529318900 Stay Updated: Find us on Twitter: https://twitter.com/a16zFind us on LinkedIn: https://www.linkedin.com/company/a16zSubscribe on your favorite podcast app: https://a16z.simplecast.com/Follow our host: https://twitter.com/stephsmithioPlease note that the content here is for informational purposes only; should NOT be taken as legal, business, tax, or investment advice or be used to evaluate any investment or security; and is not directed at any investors or potential investors in any a16z fund. For more details please see a16z.com/disclosures.   Stay Updated:Find a16z on YouTube: YouTubeFind a16z on XFind a16z on LinkedInListen to the a16z Show on SpotifyListen to the a16z Show on Apple PodcastsFollow our host: https://twitter.com/eriktorenberg Please note that the content here is for informational purposes only; should NOT be taken as legal, business, tax, or investment advice or be used to evaluate any investment or security; and is not directed at any investors or potential investors in any a16z fund. a16z and its affiliates may maintain investments in the companies discussed. For more details please see a16z.com/disclosures. Hosted by Simplecast, an AdsWizz company. See pcm.adswizz.com for information about our collection and use of personal data for advertising.

2020 was a year where many of us got to experience firsthand how AI is transforming the creative spheres, from writing to image generation. But those are far from the only arenas where AI is advancing. In this episode, A16Z Bio and Health founding general partner Vijay sits down with A16Z co-founder Mark Andresen, who infamously wrote his essay, Software is Eating the World, over a decade ago. In this conversation, Mark and VJ discuss what AI can and cannot do today, but also the tension of where it's advancing and what might stop it. They also explore the rapidly advancing technology

with a frame of what it can augment instead of replace, touching on the potential future of doctors, teachers, therapists, and more. I hope you enjoy this fascinating conversation. Hello, and welcome to BioEats World, a podcast to the intersection of bio, healthcare, and tech. I'm Olivia Webb, the editorial lead for Bio and Health at A16C. I'm very excited to share this episode because it features bio and health founding partner, BJ Ponday,

in conversation with A16Z co-founder Mark Andreessen on the topic of expert AI. You can also check out a video version of this podcast on A16Z's YouTube channel, and we'll only get in the show notes. In this episode, Mark and VJ have a lively discussion about the future of expert AI with regard to health care, but they also get into self-driving cars, screenplays, music, and the nature of consciousness itself. It's one of our longest episodes to date, but we could have gone longer. As VJ and Mark discuss, AI has the potential to change many industries.

So let's get started. Hey, I'm Vijay Ponday. I'm the founding general partner of the A16Z Bio and Health Fund. And I'm Mark Hendryson, a co-founder of A16C. Mark, thank you so much for joining us. Yeah, it's pretty great to be here. So, you know, you famously wrote about software eating the world. And that was basically, what, 10 plus years ago?

Yeah. And actually, that very much seems to have come to fruition. If you look at all these other industries, that software really wasn't so a part of, software has actually become a dominant part. But actually, this year's been kind of an amazing year for another type of software for AI. And I'm curious to sort of talk about the arc of what we think is going to happen in the future, based on what we've seen in the past, and really how this new technology is going to change everything, much like we've seen software change the last 10 years.

I'm curious what you think for just like this year. It's been kind of an amazing year. We always seem like not much happens in any given year, but 2022 seems to have been an amazing year for AI. Well, so Vladimir Lenin once said, there are decades in which nothing happens, and then there are weeks in which decades happen. Yes. And let's not hope that happens politically anymore, but it does happen, you know, in science and technology, it does happen.

There are sort of moments where things kind of hit critical mass. And, you know, this sort of AI machine learning revolution seems like that's what's happening right now. You know, it's been interesting to watch. You know, it's sort of like, it feels to me at least. It was like there was like a breakthrough moment in 2012, right? that had to do with images. Yeah. And then there was a lot of work, you know, subsequently that led to, you know,

things like the creation of self-driving cars, you know, based on that. And then there was some, it feels like some natural language breakthrough maybe three years ago. Yeah. And now that's really catalyzed into this, you know, kind of whole thing that we see happening around, you know, GPT and text generation. Yeah. And then, you know, even other applications, transcription, you know, is getting much better. All of a sudden, speech synthesis is getting much better.

And then now you've got this artistic revolution happening with image, image, you know, image creation. and now video creation is right next, coming up now really fast. And so it seems like one of those catalytic moments. And then it's like every week now, it seems like there's like fundamental breakthroughs, there's research papers, there's product releases coming out.

So it seems like a cascading thing. The way I think about it as a software person, sort of lifelong programmer, is that they're basically, in the fullness of time, it will appear, I think that there were kind of two different ways to write software. There was sort of the old Paderet of Rights software, which is sort of the classic Von Neumann machine,

you know, deterministic way. And the whole problem with writing software and the old model is like computers are hyper-literal. Yes. And so they do exactly what I tell them. Right. Every time they do something wrong, it's because you have instructed them in properly.

And it's your very humbling experience to learn as a young programmer that everything is your fault. And the machine will just sit and wait for you to fix the problem. Like it's not going to do that on its own. But then there's this new, there's this other way to write software. And this is actually due with, you know, having these AI systems and then having training data, training the systems, tweaking the systems, and the sort of, you know, capability that that,

the way I described it to kind of normies is that, you know, that sort of unlocks the ability for computers to more and more interact with the real world. Yes. And with the messiness of the real world, right, and the probabilistic nature of the real world. Yeah, well, it seems almost less like writing software, almost like training something. It's like when I think about machine learning and image recognition, you talked about it felt like it was almost like training a dog, right? Like reinforcement learning is like, we'll give treats as it gets better.

But there's something different now. It feels like, I don't know, we've gone from like training. a dog to recognize a bird versus like a hot dog or hot dog or not hot dog or so on to actually something where feels closer like training a person or I don't know how you feel like when we talk about learning and training and Dita like what are we training what is where do you think we are in that arc of getting to like eventually how 9000 and so on well so I while this has been happening you know I have a you have a young young child so I have a seven-year-old now so as this stuff has all the mapping I've been simultaneously training and training now the seven-year-old yes

Yeah. You know, anybody who's had kids will recognize what, you know, kind of what I'm about to say. But, you know, it is really interesting watching little kids. The way I think about it, or at least look at I have who's great, it's like, you know, it's like everything, for the first few years, it's like every single thing he did was like a little applied physics experiment, which is like, let's see what happens if I drop this. Let's see what happens if I eat this. Let's see what happens if I do this to daddy. Yes. Right. And see what the response is.

Yes. They just run, experiment. And you can see it. You can see it very clearly when they're like learning how to walk. they're like running all these experiments about how to stand up and what to hold on to and they keep falling over and then at some point like the little neural network like actually figures it out it does learn off and the way they go right yeah um and so you know clearly it's like it's kind of you know it's a little bit eerie like you can see that a similar kind of thing happening you know like having said that like you know the human brain just like keeps developing um and then you know it ultimately you know it clearly has consciousness achieves you know higher levels of consciousness achieves higher levels of sort of self-knowledge you know reaches the daycart you know kind of you know stage where it is that has self-awareness, you know, clearly is very creative from an early age. I'm a little less convinced that the software technologies we have now are like on some linear path towards just like, quote, AGI, it's quite a quote, like consciousness, like, I guess.

It's hard for me to believe the consciousness is just simply like emergent from like higher-scale neural networks. That to me seems like a hand wave. Having said that, I have a lot of our friends who are pretty sure that that's what's going to happen. Yeah, actually, I feel that way as well. So I want to get to EGI in a bit. I mean, and also we can debate where the consciousness is an illusion as it is. But where we are now is kind of amazing. Like, people can take like GPT3, you can give it SAT exams, can do okay.

Actually, you can do quite well. Yeah, yeah, I can do it. So the one I saw that scored it like with 1,200? Yeah, yeah, yeah, something like that. Yeah, so it's not bad, right? I can do that work. That would get you in a lot of. That love this.

Yeah, I actually gave it like asking questions like to explain the derivation for the source trial radius, you know, the black hole radius to write a code. for, let's say, eight by eight, Tic-Tac-To, like random things that dude should never be able to do because it's not just memorizing it, it's generalizing, and it's getting that. But then also it actually seems to have some sort of weird hiccups, and that actually one thing that really does not seem to get is humor, you know.

So I'm kind of curious where you think it's going to go, because before we get to AGI, there are things that an average human can do pretty well that GPT3 can't, but then there's also what experts can do. And what I'm very curious about is actually we may get to some of the expert stuff first before it can do even something like humor. The irony is that something like humor that we take for granted might actually be really hard and other areas might be easier.

Well, the ultimate example of the things that can't do, like it can't like pack your suitcase. Yeah, yeah. Like there's no robot that will pack your suitcase. Yeah. Like if you try to get her like make an helmet, like it, it'll shred your clothes and they're not, you know. Yeah. So it could drive your car, but it can't pack your car. but it can't back your suitcase.

So you can't do your lottery. So there are these interesting kind of twists. So I would describe a little bit as followed us, which is I think that this generation of AI that we have is impressive. It is a little bit of a sleight of hand. Yes. Which maybe we talk about. But I also think, actually, to your point,

human consciousness is human intelligence. Yes, is also a little bit of a slide of hand. Yeah. It would be slightly different slides of hand. So the slate of hand that you see when you're using GPT or one of these image generation things is it's not literally creating new information. Like, what it's doing is, it is, it doesn't have, like, it has no opinion.

Yes. It has no, like, point of view. It has no, like, it's not sitting there, like, thinking on its own, coming up with some new thing. What is doing is, it's basically training, you know, ideally what it's doing is, is trading on the sum total of all existing human knowledge. Yeah. So for text generation, it's training on all existing human text, right?

And so it plays back at you, basically, projections from the sort of, you know, assembled composite. Yes, the text. And so you ask it to do the 8 by 8, like, no, probably somebody on the internet at some point wrote some paper. I think that's a little more than that because I asked like 56 by 56 or 101 by 101.

It asked some sense of generalization. Yeah. But I'll bet we can check this. I'll bet if we Google long enough, I'll bet we could find a paper that described a general purpose algorithm for, you know, multi-eval. Oh, that may be, right?

Yeah, yeah, yeah. He's like, somebody did. Yeah, I've done the same key way. I haven't write sign-filled scripts. And sometimes they're really funny, and sometimes they're just like the information. Yeah, I went for curb, but it's the same idea. It's not yet.

But like, look, there are a lot of jokes on the internet. Right. And so you'd have to kind of, you could kind of go back and kind of say, okay, it probably like pluck these jokes. Or by the way, maybe there was a paper somewhere where the articulated a general theory of humor, right? Because this has been, humor's been studied as a thing. And maybe there's like a general thing. And maybe there's like a general thing. And maybe there's like a general thing. Well, it could be too, like all sitcoms might be the same sitcom. Well, at some level, right? Well, so you're being an example. So I also have to do like dramatic screenplays. Yeah, like dramatic hostage play. Yeah.

But it's quite good at those, like three. You can say like write a three-act screen play. And yet it will do it. And it will have the proper, like, set up and resolution and so forth. But, yeah, there are systems for, like, screenwriting in Hollywood where they have, like, three acts. Yes, yes, yes. It's all Rocky or it's all Star Wars. Yeah. Well, so, so actually, it's really interesting that maybe what we think is magical when humans do it isn't actually all that magical either.

So that's what I was going to say. So then the human's light of hand as like, you know, is there actually free will? Is there actually creativity happening upstairs? Yeah. Yes, out of the way, if there is, is it everybody? Is there really a thousand types of movies or is there, like, one latent space of the monument. And basically what's happening.

I think the theory, you know, I'm kind of making this up, but I think the theory would be the hero with a thousand faces or the idea of the Jungian hero's journey. Yes, just sort of the basis for all of these plots. Star Wars and Harry Potter and everything else.

Yes. It'd be, you know, a somebody with your background might say that basically it's sort of algorithm for surfing human neurochemistry. Yes. Yes. Right. And it's generating different like neurasso sort of neuron,

you know, neurasical responses to like, you know, fear and anxiety and love and all these other things. I've always been fascinated there's this thing in psychology called core affect theory oh that one i don't oh yeah this is great so we so okay so what i hit would say all these despair like really these different emotions yeah great core affect theory says no we don't oh yeah yes or no good bet good or bad yeah and then then we either have like a positive like uh they either have like a positive uh neural response or a

negative neural response and then it's either high intensity or low intensity and then you just basically and so it's like wistfulness is like you know just slightly negative but like you know despair is like extremely negative. Yeah. So it's all two by two. It's a two by two. And yes, and we're more basic organisms

than we think. And then we just, we retro, you know, and we're very, one of the things is very known is humans are very good at creating a story

to justify whatever happens, right? Um, and so we create these stories, these scripts are of this idea of an emotion, but it's basically just justifying the neural response.

And so the, the cynical view would be like having an ice cream cone in the hot day and we have falling of love as far like the same thing. Yes. Well, maybe neurochemically maybe they are.

Well, This comes into play in, like, you know, drug abuse, right? Yeah. Yeah. Things that generate an opioid response. Yeah. Like, some people get an opioid response from alcohol. Yeah.

Right. And their former pro-alcoholism, people don't get that response. So it's literally a neurochemical thing. So, yeah, look, maybe we're bundles of neurochemistry to a much deeper extent than, or much simpler extent than we want to believe. Yeah. Having said that, you know, again. Oh, oh, and then that takes me the other thing on AI, which is, I do, you know,

one of the ways that people are testing AI is with the so-called Turing test risk. Yes, and the simplified form of the Turing test is you're chatting with somebody that may be a human or maybe a bot and you chat for 20 minutes and if you can you guess better than randomist with a human or a bot. You know, my take on that is the Turing, you know, Alan Turing was a genius but the Turing test is malformed. Yes. Humans are too easy to trick.

Yeah, yeah, yeah. But that's too low of a bar because tricking a person is not that hard and does not prove anything other than you tricked the person. Yes. Like I think, and this is relevant because I think, you know, things like GPD are about to pass the Turing test. Yes, yes. Yes, any other property.

Probably right. Yeah. Yeah. Yeah. And so I think it's going to turn out that that was too lightweight of a test. Yes. Well, here's my favorite example for why I know GPT is not self-aware. If you ask it if it's self-aware and you ask it to elaborate on how it became self-aware, it will happily tell you. Yes. And by the way, if you ask it, if how it's going to feel, if you turn it off, it's going to tell you, please don't.

Yeah, yeah, yeah, yeah. If you ask it to explain to you why it's not self-aware. Yes. It will very happily do that too. It does not have a. different opinion that helps those two outcomes. Yeah.

Whereas every living, you know, every conscious while, even non-conscious living organism has a very different response

to those two scenarios. It's been amazing because in some ways I feel like it's as much been interesting to study the AI as the AI is reflected for us to study yourself. You know,

and I think we are sort of seeing that the magician has certain tricks, whether it's an AI magician and a human magician. I was going through this education process. Curious though,

like it feels like, you know, so like GPD can get into high school, getting into college, let's say. What, like, what would it take for it to get its Ph.D.? You know, and, like, I think that's where the sort of dramatic stuff is to come. Yeah, yeah. Well, again, exactly to your point.

This is stressed, I haven't asked the question the other way, which is that, well, okay, what does it take to get a PhD? What does it take for you? You would get a, like, yeah. How are the universities doing? Yes, yes, yes, yes, yeah. Yeah. How are they doing quality control of their own Bs? Yes, so how many people are getting PhDs today that we would say are, like, actually valid,

like, you know, whatever. actual accomplishments. Yeah. Yeah. By the way, people who got, you know, professors 100 years ago, like, how would they score the SDs that are being granted today? Yeah, they say, well, I answered that. You know, would they say the bar is lower?

Lower. Lower. If they would say the bar is dramatically lower. Yes. Right. Yeah. And so, you know, the answer might be we have lower at the bar. But the same thing for college admissions.

Like, you know, what does it take to get to college? What does it take to finish college? Yeah. And, you know, the education says, well, this is coming up a lot right now because it's like, okay, GPT can auto-generate, like, you know, essays, right? And so student essays. And so it's like, okay, the grading method of assign an essay and grade the result, like, is probably not going to work anymore. But it's like, was that ever actually, like, just because we thought that that was education, was that actually education?

Like, was that actually teaching anybody anything? Like, actually, I'm sure someone's going to take that to apply to colleges. Oh, yeah, yeah, absolutely. Yeah, college applications are basically done. Yeah. At least at the extent that you believe the college applications were in a legitimate wage as evaluate anybody in the first place. Yes. Like, yes, it's now.

And star. I'd be more skeptical that they were ever useful at first one. Yeah. Right. Yeah. Well, so in the PhD, let's talk about, like, at least that old school mentality of a PhD of some advanced learning where you become an expert in something. Right. You know, I think that's the thing where... What do you mean by expert? Let's say the ability to be in the top 0.1% of humanity of, let's say, designing a drug or building doing something.

Yeah, yeah, yeah, yeah. Is that what they teach? Yeah, no, that's... That is my goal. It wasn't aware of people. It wasn't aware of that far as the drug. I think it is. Is there some time? Or at least that's what you have to do eventually when you get out. Right.

You know, and you have to apply it. I think it's where I think it is, one of the things about being an expert in my mind is that something that is the difference between bad, good, and great can be really close. Like, I could probably write a piece of music, but no one would think it's all that great, you know. And then you could have someone who's a good musician, but not a great one. And then you have like a genius, like a Mozart or a Led Zeppelin or whatever, a peculiar. genre, you know. And I think where we aren't there yet is that when the difference between good and great is so close, or like, I don't know if I remember from a smile tap, there's a fine line between Berlin and stupid, you know, I think that is where I think it hasn't really hit yet.

And that if you look at the jokes, the jokes are just kind of, okay, the screenplays it makes are not like brilliant screenplays. I think it could get into college, but could it win best screenplay, you know? And so that's this part where I think we're there, we're not there yet. You know, but that I think we're getting there. So name a great music composer generated by a music PhD program. Yeah, yeah, yeah, I've been the last years. Yeah, name one. Yeah, I'm thinking more in the scientific side of things.

But, yeah, I don't think, probably the PhD program, and that space is probably not intended to generate music. Okay, yeah. Name one great screenplay written by a PhD in drama. Yeah, yeah. So that's an interesting point. But I think what I'm getting at is still, like, the ability to do something.

And so, and the education part, we can talk about how they learn. Okay. Because I think in the case of the screenplay or the music you're talking about, they still have to learn something, right? You know, or do you think they just innately sort of knew how to write a screenplay? I don't know. Yeah. I assume there's a process where they write a screenplay, it's kind of mediocre.

Oh, yeah, yeah, yeah, yeah. And then they get critiqued or they critique themselves and then it improves and improves and improves. Well, the screenplay, okay, so the divorce is divorce for the education. Yeah, yeah, yeah. Yeah. The test of the screenplay, the test for a screenplay, does it sell? Yeah, so screenplays are subject to market discipline. Yeah, right?

Yeah, so question number one for screenplay is, does it sell jazz studio? Yeah, and then the test number two is when the movie comes out and the TV show comes out, is that anybody watch it. Yeah, like it. Yes, do they finish it. Yeah, yeah. Yeah, one of the fun things that Netflix will now tell people who make film and TV is they actually tell them for the first time whether anybody's actually finishing. Yeah, yeah, death. Yeah, just all those stats are kind of monblogging.

Right, you know. Yeah. A lot of movies, like, and, you know, people go to the theater and they feel, you know, they feel, you know, tested in and they don't want to leave in the middle, but it's very easy to punch out. It turns out a lot of screenplays. You know, this is something that professional screenwriters will tell you, like, it can't ever sag. Yeah. Yeah, just as one example, because people will stop watching.

So, yeah, so screenwriting is subject to market test, popular music. Yeah, for market tests. But in the classical music, which I'm a huge fan of, is no one is subject to market tests. Right. It's thoroughly subsidized. Yes, that's interesting. Right. It's not in the free market anymore.

Yeah. Or maybe the equivalents are movie music is, you know, Yeah, so movie music is subject to market tests. Right. And it's probably the modern classical. It is the modern, yeah, it is modern classical. Yeah, for that reason. So, yeah, the market test is real.

But, yeah, let me grant your point. So let's build on what, yeah, grant your point. Like, let's use the, we use the term paste. Yeah, yeah. Or just ability to do something hard. Well, ability, so, okay, so ability to do something hard and create, let's say, create something hard. Yes, yes, create something complicated.

And then also the ability to judge. Yeah, right. And you critically like to start with judging your own work. Yeah, and probably they're for the ability to prove. And then they're for the ability to prove, right? So, yeah, I think that there's, yeah, so there is something about taste. Yeah. Like, I tend to think this stuff all has, like, aesthetic.

Yes. Properly constructed math. Yes. So, Fermilet or software program has aesthetic. Oh, 100%. Right. Actual design has aesthetic property. Physics, you know, all of it.

Yeah. Yeah. So there's something about taste that's like some combination of quantitative, militative. Yeah. Like, it's great startup. It is like from a mediocre one this tastes. Right, yeah, exactly. And like there's certain signals, like there's certain methods and certain signals.

It's not necessarily reducible to another rhythm. It's more of like a composite, you know. It's sort of the foundational knowledge to bind with some scope of experience combined with some kind of ineffable characteristic of genesis. Well, we associate an aesthetic with it, but I wonder whether that's also just our emotional connection to it. Right. You know, because I think we have this good right or wrong or more right or more wrong, like a gradient. Like, yeah, that's the right direction.

But a lot of is also whether something is elegant versus just a hack. You can tell whether these great things are just simple and powerful rather than like some complicated machine to do something that, you know, you know there's going to eventually fall apart or that. You think about that's true in physics or in go to market or in music. It has all that both that sort of complexity and a simplicity at the same time. But I'm curious, like, so when AI gets to that point, Which I think that's a when, not an if.

Okay. Yeah, yeah, yeah. So why would you say, why wouldn't it get there? Because, like, do we even understand how it works? People. Maybe we don't have to. Well, maybe we don't have to. So this is where I described, this is like the AGI.

This is where I go the hand wave. Really? It's sort of the same thing. The embedded assumption that it's a wet is that it will be an emergent process that will sort of unlock as a consequence of greater, greater levels of yeah. Maybe. Yeah.

Yeah. Yeah. One way of looking at that is, yes. That is what's going to happen. That's human context of submerges like that. Yeah. They have to see what's going to happen. The other pressure on that is it's just a mess.

It's a hand wave. It's a hand wave and I think it's a cold cope. And the cope would be, okay, so let me ask you a question every time. Yeah. Yeah. What is the sub-specialty of human biology and medicine that most understands the nature of human consciousness today? Oh, I don't think there's a, there is one, right? There is one.

Anesthesiology. Okay. Which is poorly understood. But they know how to turn it off. Yeah. And they know how to turn it back on. Yes. They've got the on off switch.

That's all we got. That's all we have. Like, we collectively have been studying this question of human consciousness for a very long time. We have very advanced technologies today, functional MRI, like all this stuff. But that speaks to that there's a field I would love to see create it, which is molecular psychology. Yeah. Okay. Yeah, where you can start to probe this, a little more than on up.

Okay. And molecular. So, so, and there's some... Literal or metaphorical, what I say? Quite literal. Like, it's a play, like, molecular biology was this big thing in the 80s. Or if I, like, can bring, like, chemistry of small molecules to, like, Polket biology, or chemical biology as well.

And if we could use, like, small molecules to maybe perturb more than just on off, but like perturbs things, we can start to understand the brain a little bit because reading is one thing, but like poking and sort of perturbing and then seeing the result is usually how we do any sort of experiments?

Would you view that? Is that a chemical? Would it be a chemical experimentation? It would be electrical? It could be either one. It could be any of that, but probably be some combination of those things. Neuroly and it's like on a track and theory

to enable any of some of this. Yeah. So like, look, we just don't, Okay, so here me the counter argument is what we just, we don't know how human consciousness works. We actually, I actually, I, I didn't go into the field, but I didn't go in the field. I actually was going to be what I was going to study in school 30 years ago. But I looked at the field at the time and I was like, they don't have a clue. Yeah, I'm going to spend my entire career.

So you wanted to go into consciousness. The time of the college instance was, yeah, the hot thing. Yeah, yeah, yeah. But that was like the expert systems. Expert systems. Well, there are early neural networks. And then a lot of it got into brain chemistry and like we're going to figure this stuff out and we're going to learn how to build, you know, And it's just like they didn't know them, as far as I know, they don't know now.

And so the counter argument would be this is all just like massive cope for the fact that we actually, we don't understand that. So we don't understand how to do it. And so all we can do is hand wave and kind of just say, well, it's just going to be a virginate. And it's like, no, it's not. And we're going to be sitting here 30 years from now and we're still not going to have any more knowledge. You know, barring other scientific breakthroughs of the kind that you're talking about. Yeah. What's interesting is if you think about that time, we had neural nets, but they were all single-were, basically.

And then they couldn't even do XOR. You know, you couldn't even do some simple things because you, you couldn't even do some simple things. you needed deeper networks to get at them. And you couldn't have deep networks then because we didn't have the computational power. And so the space was pretty dormant for a while, you know, AI until like we started going to having the, basically just the computational power from GPUs

and other things that we would go deep. And then you could feed the data through. So it is possible that we sort of have a point where we sort of saturate the compute that we have now. We get to as much as we can get to. And that may get to close to AI. Maybe not. And then it takes another like 30 years

to get to the next. sort of breakthroughs to get there. But, okay, so I would pull back from there. So AGI is the fun thing. There is a sort of step back, which is to pick a domain. And you know the domains I think a lot about life sciences, the diet and drugs, doing health care, like seeing if you can do a, pick a diagnosis, can you suggest a drug? In those areas, now we're talking about much more limited domain.

So we're not talking about we don't need to go all the way to consciousness for that necessarily. You can have something that's more limited. in that limited domain, right now it seems like generally AI isn't quite far enough yet to be able to like, yeah, I don't see the examples quite yet. Yeah, well, we'll see. I mean, so what's the counter?

And I know you, especially think about health care a lot. Yeah. Yeah. Well, so the first thing is whenever your score, well, let's talk about medical diagnosis, which is kind of just low-hanging fruit question because everybody experiences it. So to start up front, you have to ask a question up front,

which is like, is the goal, what's the threshold? Is the threshold perfect, or is the threshold better than human? Yeah, that's a great point. Yeah, right. And by the way, this is a topic that comes up all the time with self-driving cars, right? Which is, is it perfect? It will never make a mistake or is it just going to be better, better than human. And the way, the self-driving cars score this is accidents per thousand miles driven. And self-driving cars are already lower than human drivers. And humans are already living cars. And humans may actually be getting worse than one location. Every other car gets trained on how to do that. And then, of course, the machines have the characteristic. They get better universally. Right. So a car has one mishap and one location. Every other car gets trained on how to to deal with that in the future where, you know, the learning happens across the entire system. And so, like, I think you can make a serious argument that, like, basically, self-driving cars are already better than people on a relative basis. And therefore, like, morally, you could even go so far as to say human drivers should be outlaw today.

Right. Like, if you have the alternative, if you can have the self-driving car, then, yeah. Like, the utilitarian argument would be you should obviously ban human drivers today because the machine-driven stuff is already better. Probably, by the way, the same is true for airplanes. right now we're not actually going to do that and there are other considerations involved and so forth but like you know logically speaking you should at least think about that as a possibility and i think you should think about that as a possibility i think for medical diagnosis which is you know and here the test is very simple which is well i here at least express two tests

test number one is the absolute test which is if i feed in a set of symptoms it generates the correct diagnosis 100% of the time deterministically guaranteed that's a high bar the other is i do that with the algorithm and then i go to 100 doctors human doctors and I get back out of different responses and then let's compare, right? And then let's track over time and says, you compare to the media. Yeah, right?

And like, how good is the media doctor at doing the diagnosis? I'm like, I don't know what your experience has been. Well, and the meeting doctor may be smart, but also may be overloaded, maybe exhausted, may have like 12 other patients. 15 minutes. Yeah, I know a lot of experiences, 15 minutes.

There's the thing you're, like, experts in these areas tend to either, like, be, like, doctors themselves or they, like, know a lot of doctors, where they have like they're, you know, they work in the industry, they make money, they have a concierge doctor who spends a lot of time with them and does house calls. The median health care experience is 15 minutes in somebody's, you know, Harry's schedule with the doctor

that may or may not ever see you again and has very limited data. And there's one known algorithm, which is that they come up with their diagnosis, they come up for the treatment, you go with that, that doesn't work, you repeat. And while not sick and while still sick and not dead, you just repeat.

And then I think many of us have been through that. Well, and then there's all the other sort of things. So then there's like drug interaction. You know, is any one doctor tracking all the interactions of your drugs? Then there's this other issue which is, okay, they give the prescription. Is there actually compliance for taking the prescription? Does the doctor actually know whether you're taking the prescription? Compliance is one of the biggest disasters.

Right. But that means like the ability for an immediate doctor to even evaluate the success of a treatment, they may actually may not be able to do it because they may not have the data on compliance. And so like you look at the existing, I don't know, for me, you look at the existing system by which this all happens. It's very similar to looking at the existing system by which people actually drug cars, which is like, oh my God, this is not good. Like, this is really not good. And we kind of fool ourselves

and I believe it that it's good because it kind of feels good. We don't really want to look behind the curtain, but we look behind the curtain. And it's pretty horrifying. Yeah. And so from that standpoint, if you follow that logic,

then it says, okay, if the machine could do a better job, you know, if the machine was twice as good at just like listing symptoms, giving the response during the prescription doing the follow-up. Yeah. I mean, how far, I don't know if you've done this, but you plug in a list of symptoms.

I've been playing with it, too. Yeah, yeah, yeah, yeah. Yeah, yeah. I mean, because it does have access. I mean, it has to the collective medical. Yeah, and if it does that, now it can't. You know, it could be filled with all the EMRs, all the medical records and so on, and then it could sort of learn from that as well.

Well, then the other question, I'm sure you thought about, but like, okay, so the medical field moves. Yeah. In the existing system, the media doctor has to read all the papers. Yeah, yeah, yeah, which never happens. Right, no one has that for that. Yeah, right. Yeah, and then there's continuing education, but still it's not the same. Well, here's an example.

Do you like your GP? would you rather do you want any NGP or an old GP? Probably, yeah, right? Well, presumably the old GP has more experience. So they have more pattern matching over time. Yeah, with patients. The NGP has probably more up on the current science. Yeah, yeah, yeah.

Okay. And then it's like, okay, do you really want to have to make that tradeoff? Yeah, or can the machine actually have both of those? Exactly. Well, that's the thing, is that like you talked about how, like, can it beat, let's say, how does it do compared to 100 doctors? When the 100 doctors collaborate, presumably that's the ideal situation, right? I mean,

or, you know, that sounds horrifying. No, no, no, I mean, that's the wisdom

of the crowd. No, that's perfect. It could go, well, it could go either way, but it's like to beddy. Usually.

That's the Soviet method. Usually when you actually, when you pool it, you can, or at least maybe it's how you collaborate. Have you really found human beings to make better decisions in groups than they do as individuals?

That's a good question. Yeah. In your entire life? Yeah. Oh, yeah, yeah. The full, the serious answer is, the wisdom of crowds,

madness are crowds. Are flip sides of the same coin, right? When are you harnessing the wisdom? When are you descending in a madness? Or even just, you know, mediocrity. Yeah, and very specific tasks. Groups can do well.

But otherwise, it's like one big group project from high school. Yeah, which is like a... Well, so generally, what generally would have with people in groups is the social conformance kicks in. Yeah, and so people want... There's a well-known kind of thing. There's a lot of group polarization, which is you take a group of people who are inclined slightly to one side of the medical spectrum. You put them together and let them talk for three-arms. Yes.

They all come out much more radical. Yes. Yes. Right. Is it self-reinforced? Yes. Yes, yes. Right.

Well, so maybe that's a really interesting thing because you can imagine training AI to do, have these different aspects. And its collaboration with the other versions of it would be very different. Yeah, it could be very different. I mean, yeah, maybe it should do like this effectively in Monte Carlo. Yeah, yeah, yeah. Right, right, right. Run the same inputs 100 times.

Yes, yes. Right. Yeah. Well, okay, so I either will never get there or we're already there now. Right. But I think in 10 years, it does seem especially, maybe we hit like another winter, but it seems like things are accelerating so much. This seems pretty real.

It seems pretty real. What do you think society needs to do to change? Because there's like all these things we were talking about. And this seems bigger than like just the revolution of software over the last 20 years or internet from last 20 years. Because we're talking about how it changes government, how it changes regulatory, how it changes education. I mean, I don't even know where you want to start with that.

But I think that's something where it may take us 10 years just culturally to be able to get ready for this thing that may arrive in 10 years or may already be here. Yeah, I don't know where you want to start. Yeah, so where I would start is we've already fallen into, I think, we have deliberately kind of fallen into the trap already, which is we've only been using a single kind of example. And we've used it both in our discussions on medicine and also in education, which is basically a something is people are doing something today and then maybe the machine can do it instead. That's an important thing and that's, you know, we're thinking about. But the way the technological impact actually plays out in human society is not just that.

The way it plays out is it lets and basically revisit more fundamental assumptions. Yeah. Or what's not being done today. Well, it's not being done today. Yeah. All of a sudden becomes possible. And this comes, this always comes up in any sort of discussion about employment. Yeah.

Yeah. People doing jobs versus machine doing jobs. People get worried about technological displacement of jobs. But technological displacement of jobs, like technology never actually creates unemployment. Technology only ever creates jobs in that. Yeah. And the reason for that is technology makes possible things that were not possible before. Yes.

Which is growth. And so specifically, for example, the role of the doctor, you know, it's like, okay, the doctor of the future is probably not going to be doing the same. Right. We have a term in the IT, break fix. This is kind of what doctors, you know, the core motion of a lot of doctors. As he said, diagnose, prescribe, diagnose, prescribe.

And doctor's bugging. Yeah. To be bugging. Yeah, exactly. Doctors of the future probably, like the technologically empower doctor carest from now is highly unlikely to be spending their day doing that. Yes.

They are probably going to be spent on their day doing things that are actually much more important than that. Yes. Right. And so, for example, maybe they have more time, right, with patients because the machine is a time-saving device. Maybe they have more data to draw on, you know, to be able to make their decisions. You know, they've got the machine as a partner in making decisions.

Maybe they're able to spend more time in their conversation with the patient talking about psychological issues as compared to just physical issues. And as, you know, a lot of medical conditions involve, you know, two sides of that are behavioral issues. Well, as you know, like a lot of primary medical issues today are consequence of different behaviors. Yes. And maybe doctors should be spending more time of behaviors.

Yes. And it speaks to compliance as well as other issues. Awesome. Yeah, well, I mean, compliance is behavioral issue. Like, why don't people do this or that? Right. But then also there's all the behavioral health issues, right? Which is probably one of the biggest catastrophes that we have coming out of COVID.

Yeah, exactly. Right, yeah, exactly. Maybe doctors should be, you know, maybe the doctor in the future will be more of a life coach, of which there will be a pharmacological, you know, sort of a biological or pharmacological component, right? But maybe it's like, maybe it's more of, you know, sort of the holistic, you know, medicine. And so, you know, maybe the doctor in the future is just as much, is actually a much more important and, you know, sort of fundamental figure in your life than he is, than he or she is today.

Yeah, that sounds great. So if I'm a doctor, that's where I would want to be, like, and yet towards, right? And then that's probably a bigger and more important market, right? And then in terms of, like, the size of that industry will probably expand, you know, kind of correspondingly. I think the same thing is true in education. Like, you know, the teacher 10 or 20 years from now, I hope is not doing the same things the teacher is doing today.

I hope they're doing much better things. Right. So, for example, one-to-one tutoring. There's basically the thick education example. Like, there's only one, in the last, like, 50 years, there's basically only one-known education intervention at scale that actually improves outcomes after, you know, thousands of experiments. Yeah, it's one-to-one tutoring.

Yes. Which is very ancient, actually. Which is very ancient, right, which is the ritual form of the age. Yeah, it's literally how people used to get educated. And so maybe this industrial, you know, the education system we have today is an artifact of the industrial age. If the industrial age components of it become automated, the teacher becomes freed up to actually work more one-to-old,

one of students, the result might actually be a significant breakthrough in how education works. Although the ways you're describing, you can imagine also like AI doing one-on-one. Well, yeah, they were the intensive. They're part of that. But also, yeah, and maybe the AI is the one-on-one. And maybe in that case, the teacher is supervising the AI. Right. And maybe the teacher is making sure that the AI is, like, on the right track and doing the right things and is able to kind of sit at the control panel and watch all that happen.

Well, that speaks to something really interesting because I think we're probably a little nervous, at least short-term, to just unleash this and, like, not pay attention to it. And so you'll have the doctor using this as a tool but keeping an eye on it. You'll have the teacher maybe scaling dramatically for all this one-on-one, but keeping an eye on it. Do you think that's actually the way it's going to? I mean, this is kind of how all technologies work. Yeah, yeah, yeah. So it's sort of, another way to think about it is you could imagine two acronyms for AI's and artificial intelligence, which kind of implies your placement. Yeah.

The one I actually like much better is augmented intelligence. Yes, which is like the old Doug Engelbart idea. Yes. And augmented intelligence is, you know, it's another example, the term would be Steve Jobs, a bicycle, your mind. Right, right. Yeah, we're, you know, a bullet train for your mind. It's like, yeah. Yeah. Right. And so the augmentation, right? And so the way, if you just look at the history of new technologies, the way it plays out, it's every afraid is going to be a replacement and it turns out it's an augmentation. Yes.

Yeah. So you take a human being and you give them the technological tools. They have, therefore, are much more productive. Yeah. Like a factory versus like, like, artisan with their tools. Yeah. Exactly. Or like, you know, you know, the dream of like an, you know, an isoskeleton. Yeah. Yeah. You know, the dream, you know, any of these things. Yeah. I mean, look, artists are much more productive today, with digital tools that they were with just, you know, painting canvas. Yeah, yeah. And by the way, even artists that still work on painting canvas are much more productive

today because they can tell their promise to much larger audience online. Or, like, my favorite thing for art is, like, you know, photography comes online and that dramatically changes art because being photorealistic isn't that interesting anymore. But so that creates moderate. Yeah, which actually is maybe even more expressive. And just taking a picture. And so now I can make pictures with AI all the time. So where does that shove art, maybe to a more interesting place?

And the artists of history, the artists were not happy about things. introduction. Yes. It originally is a threat. Of course, yeah. But it transformed. Yeah, it turned out to be, it turned out. Yeah, it turned out. The market for art is much larger today than it was.

That's interesting before the introduction of photography. I mean, we call it different things. We call it things like TV shows and so forth. But like the market for creative expression is much, much larger than it used to be. By the way, music, same thing, right? I mean, you know, recorded music was originally a threat. It used to be musician who composed and perform, right? And then, you know, to have music in your home, you'd have to hire a musician to come into

your home. You know, photographs were a threat to that. But photographs made the music. industry much, much larger, so people who were good at making music all of a sudden had a much bigger market. Yeah. So I think AI is going to play out in a very similar way. Like, there are people who will argue, know AI's different because I just keep climbing

that slatter, it will replace everything. I actually think it's going to be, basically it's the ultimate superpower. It's the ultimate pairing. We were talking about great screenplays and scripts. A good example. If I'm a Hollywood screenwriter today, like GPT is my best friend. And I'm just sitting there all day long and I'm just saying, you know, playing out, and it's like, okay, I reach this plot point, dot, dot, dot, give me a list of, like,

like 10 ideas for what to do. It's like, oh, okay, that's an interesting one. Well, I'll give you an example of how this could work. So, Matt Manman, it's one of my favorite shows. Matthew Wider, you know, ran that show. And he was always praised. He was like, wow, that show was so unpredictable. Like, you know, you never knew where it was going.

And he said, yeah, well, the technique we had in the writer's room was, in any given time, we had to figure out what happened next in the plot. Yeah. We would bring up with the five sort of things, obviously. Five obvious things did rule all those dudes. So GPT would be obvious things, and you rule those out? Yeah, exactly. So it pushes creativity.

All of a sudden, every individual screenwriter could do that without having to have a whole writer's urban to brainstorm. And just plug that in. It gives it back to you in two seconds. You're just like, okay, not those things. I'm going to do something else. And now I am more creative than I was before. Wait, your comment about music is really interesting because now we've got Spotify. So we got everything in your pocket.

Can imagine like the AI Spotify, which is like the doctor, the personal trainer, the educator, like all those different things in my pocket available right now for whatever I need to do. Yeah, that's right. Yeah. And with the human escalation path, right? Yeah, it's like, yeah, the AI therapist or whatever, but yeah, with the thing, it would be like, well, okay, yeah, yeah, especially if it gets really serious to escalate immediately. Yeah, that's right, yeah, yeah. Okay, so what's, what's going to hold us back? What do we need to change? So I think it's mostly fear. So this is where maybe I'm a radical on it. Yeah, you know, this is usually where people start talking about, like regulation. I think it's, like, we have these, we have these fear-driven reactions. I always think about, there's this deep-seated myth in human societies, the Prometheus myth, right? Yeah, yeah. And the Prometheus myth is all about. new technology, right? And the Prometheus myth is like basically this new technology of fire. Right. And you know, fire is one of these classic technologies where like it can be used for good.

Can burn life. Right. Or it can be used very bad. Right. And yeah, it can destroy your whole world. And so, you know, Prometheus famously, you know, goes and retrieves, you know, fire from the gods. And his punishment for it is to be chained to a rock and if his liver pecked out every day for the rest of eternity. So embedded in there is like the anxiety about the new technology. And then the rivaled into technology and maybe is like, you know, the fear, right, is that. It's that. The fear, right, is that. isn't that bad and the person who does that should be punished. And so I always find that myth kind of plays out over and over again and all these discussions about regulation that this stuff, you know, especially it's the gods who punish it, right? The existing gods. Yes, yes.

Well, on behalf of, on behalf of existence. But, yeah. Yeah, so, yeah, I think generally it's this, it's just you get these fears. If you look at the history of, we talked about some of this, if you look at the history of new technologies, yeah, generally have these fears. Yes. Every step along the way,

if technology has been created with some prediction that is going to upend the social order and cause the, you know. Well, it does up into some degree. And it will do that. But generally speaking, in a positive way, on balance. Yeah. You know, technology is why we live much better lives today. Certainly people now would not want what people had 50 years ago.

Nobody would make that, yes. Right. And you could do, you could go back in time infinitum and be like that. Nobody would ever make the trade. Yes. Nobody would ever make a trade to go back in time. Yeah. Yeah.

And it's, yeah, right. That's literally it's because you would not want to lose the technologies of the yes. Yes. Yes. Because you have today. So I think that's true. So I actually think, like, fear may be the, to rip off FDR, fear may be the actual biggest threat.

Yeah. Fear leads to the kind of, you know, reach for regulation. Yes. I'm a skeptic. I don't, it's like, I don't know. Regulating math. Because we really need to regulate math. Well, but it's not going to look like regulating math, right?

It's kind of look like regulating this superpower. That's what they're going to say. Yeah, right. But then the actual implementation is, you know, risk. Regulating algebra. Yeah. Yeah, regulating algebra. Regulating linear algebra.

Yeah. Really going to regulate linear algebra. Yeah. multiplication. Yeah, really seriously. Yeah. And then even if we do, are we going to possibly do it in a way that makes any sense? Yeah. Well, okay, but it won't, obviously won't look like that. It will be saying, well, we can't have computer drive cars. Right. Or like, what, what's the, how do you give the computer a test? Yeah. Or how do you know? Like, okay, you make this, I'll be the cynic. So, okay, you make this claim that the computer AI is better than human. Like, how do I know that? Because that, well, yes, it turns out because the cars are driving. Yeah. So, yeah.

So there was, okay, so here was a new ride. Okay, so here's how that played out in self-driving cars. Yes, there was one category of company that said, we're going to basically wait until it's perfect. Yes, it's going to basically different of the melody. We're going to work with the regulators. We're going to do this stuff. Yes.

They're not driving. And they're not on the road. They're still not on the road. There's another category of company that said, you know what, let's evolve out of basically the cruise control. And, you know, it's crew control, and then it's radar. And you get humans driving with it, and you label data. Exactly.

And you don't expect the car to drive itself in for the very beginning. the car is like an autopilot kind of thing, the expectations you pay attention. Like, you know, Tesla's the company I'm alluding to. And if you turn on full self-driving on Tesla, you're still told, like, you're not supposed to be watching a movie. You're supposed to be actually paying attention. And the car will, like, alert you when it's time to pay attention.

But, you know, notwithstanding that, Tesla has been climbing the ladder on self-driving car functionality capability. They do new software releases push live to car at night anytime they want. Those new releases are not being tested by any federal... It's a whatever it's not...

these things. There's no actual test happening. Yeah. And that has led to incredible progress, including, as you said, clearly in the data, this is now different. Because you can't drivers. You can't make it work just magically. Right. It has to do happen gradually.

Right. Because it's actually much like medicine is entering into a complex system, a lot of variables. Yeah. In the real world. Like medicine, too, it's like life or death. You know, it's just serious. But, yeah. But, yeah, we go back to how we started the conversation.

Yeah. The wait for permission saying, the might. binary zero over one, wait for permission, wait for perfection, thing, versus the incremental, let's get better and better and better and better. And the threshold is, is it better than humans? Is it an end up improvement? I mean, clearly in self-driving cars, that second approach is the approach that's working. And you just, you know, observe. And do you think you get to the tipping point where, look, let's look at the statistics? We have, because we have all this happening right now. We have the statistics. And it's like so much better than humans, what wouldn't we do?

Yeah, exactly. Right. And then at some point, the morality tips where it's like, well, obviously, we have to go in this direction. Yeah, it's just obviously better. Yeah. I suspect we're going to get their medicine pretty quick. Yeah, yeah, yeah. I'm an optimist on that. And again, I'm not an optimist because I think that's going to be perfect. I'm optimistic because I think the status quo's not that good.

Yeah, yeah. Well, that might be like you start empowering doctors. You get them tools. They start using them. And start empowering patients. Patients start using them. And actually here, I think it's even different than a car because you know, on a road.

It's your body or whatever. And actually, patients are driving their own health care more than ever. I think COVID was another sort of tailwind there. So maybe you start, maybe it's just about developing the tools and giving them out. Well, here would be an example, so let's use our screenwriting example of the plant in medicine, which is, you know, a given set of conditions, there may be many possible diagnoses. An experience I've had is there's a set of symptoms. Yes.

One doctor comes up with one diagnosis. Another doctor comes up with a different diagnosis. You read the literature and it's like actually both of those diagnoses in theory are, but like for some reason the one guy only thought of the one, the other. Yeah, only thought of the other. Yeah. So a way for doctors to start using this technology today would be plug in the symptoms, give me five possible diagnoses.

Yes. Okay. Oh, I didn't even realize, right, that, you know, because maybe this is a new thing. since, you know, I went to medical school or something. I didn't realize diagnosis number three was an option. I just go look at that. Yes, yes, right? And so the doctor is still doing the diagnosis.

It's your screenplay example. Your augments. As a doctor, you're augmentation. Yeah, in that case is luring you to things that you should know, but no. Yeah. Yeah. Yeah. I mean, that's interesting.

It's almost like having a mentor or just someone to riff with. Yeah, that's right. Yeah, yeah, yeah. And they're right. It's great thing is, it is a machine. It will riff with you as much as much more. Yes. I'm like, you know, sit there at three in them.

Yeah, yeah, it's happy to get bored. doesn't get hired. Yes. By the way, and then it also has the advantage. It has all the up-to-date information. Yes. Yes.

And all the outcomes. And when it makes a mistake, it actually can learn from it whether actually from being, other than being like devastated by it or emotionally reacting to it. Right. Right. And like self-driving cars,

if it makes me, if some other doctor in some other state had a patient last week and made a mistake and they fixed the mistake, it will not make the mistake. You can hear a patient. Yeah. Yeah. Yeah.

So, I mean, so you think, and so that is a very different regulatory play than we have seen. in the history of health care. Well, I think that's just, well, you tell me, I think that's just going to happen. So here's that everybody knows. I'll give you a couple things. Everybody knows that patients should not be on Google.

Yes. Everybody knows every patient now does that. It's called Dr. Google. It's literally how it's called in the field. Right. And there's no way, like, you're not practically speaking. Yeah, I like regulate that out of these. Yeah, yeah, yeah.

That's going to happen. I think doctors using these new tools as augmented is something that they can just do. It doesn't require approval. So the ship is where I think so. Yeah, yeah. And by the way, patients, using GPT, if it hasn't started, it's going to start imminently. Yeah, yeah, probably.

So the patients are going to show up with the results of GPT queries, and the doctors are going to have to respond to that. And so they're going to end up being in this world whether they want to be or not. But that's actually really interesting because as a patient, and I probably know just about medicine to be dangerous to myself, but like I show up with the doctor and I have all of that thought out. Basically, that might equalize the patients, you know, such that they can actually come much more educated and come from much more thoughtful

and they become much more in the process as well. Yeah. Okay, so what goes wrong? Double-ed sort. I mean, as a doctor, do you want your patient? Yeah. You want a patient more educated or us educated. They may just be humaning me, but I think they want some more.

Maybe with you that, yeah. With me, they might look a little bit more sideways. Actually, but if it was really helpful, I think they would. I think it's just about how good it is, right? Okay, so what goes wrong? I mean, look, then the big thing that goes, I mean, look, the big thing, I think two things go. So one is just the expectation of perfection, right? And look, it's very easy to generate the negative headline.

It's very easy to set off the scare of the moral panic, basically. Right? It was a single essence goes wrong, and it gets extrapolated. You know, we talk a lot about phyatomide. Like, you know, it would be very easy to have that kind of mode. Or like the person on a bike that got hit by a Tesla or something like that. I think it was biking across a freeway. Right.

Exactly. And so, like, a human probably hit him too. Yeah, that's right. Yeah. Oh, well, that's a good point. Yeah, yeah. The trolley problem. Yeah.

You know, the trolley problem spent in the press a little more recently. Yeah, it's a chance out that Sad Maconeman Free was an expert in the trial. problem. Okay. It shows you that it's not the route to the morality is. Yes. As it's been marketed,

but yeah, the trolley problem, the problem gets always mooted, the trolley problem gets always mooted about who's self-driving cars, which is, you know, you have a choice between killing, you know,

it's like, I don't know, five grandmas or one little kid or all this different, like, you have to pull a lever in the side. But, like, human drivers don't. No, no, no. New drivers never make that sound.

No, they have gas or brake. Yes, right? And they have, I'm going to hit the car in front of me. Yeah, not yet hit the car right of me. It's never this elaborate thing. It's always a very simple thing.

And so it's not a question of whether the machine can ideally solve this sort of, you know, idealized complex problem. It's kind of kit the brakes faster, right, when it's about to crash on the car directly in front of it. And so properly, logically, kind of containing the expectation here to actual real world and not having this spin off into these like basically fantasy narratives that you can then criticize. Yeah, yeah, the absolute limits. And then, yeah, look, I think just the generalized fear, right? And what I always have to remind myself is like, you know, I'm like I say, I was on the software developer by background. It's like, okay, I can actually, like the algorithms that do that, like, you know, can I tell you ever ask for how they work? No, like do they, do I understand how they work?

Do I understand the basic foundations? Do I understand the basic math? Yes. Yes. This is why I make the comment about regulating math. Yes. Yes. As somebody who's not a coder, right, this whole, this, all this stuff is weird.

It's like weird math. Yeah. Yeah. And so there is a, yeah. I have to remind myself to be patient and tolerant of people who don't understand the mechanics of what's happening. And that said, I think the people who are going to be ready, and I think they also have to get in a mechanic to try to understand us, and there's always slippage there.

Yeah, so what's the antidote to fear? Is it optimism? Is it education? I mean, ideally, ideally, it's, yeah, ideally it's cultural orientation towards new technology, and then ideally it's education of people learning and kind of, or, you know, the CP Snow, two cultures. Yes, yes, yes, across coming together and kind of educating each other. Honestly, a big part of it also, I think, is when things become a feta company.

Yeah, I mean, this is what Tesla's done himself driving cars. Yeah. Like, if it's just happening. Yeah, right? Because who would want to go back? Like, he wants to go back. Like, the system adapts. Right.

And so there was this famous Uber fought all these regulatory wars and all the cities that they were in because it was not technically allowed. Yeah, taxi limbo charges in the beginning. So one of the things they did early on was they just made sure that there were always lots of Uber cars available around state houses and city halls. Yeah. And so whenever somebody, you know, so you literally have somebody who's like in, you know, sort of giving this, like, roaring speech, you know,

city hall about shutting down Uber and then they would come out and they'd have to get home really fast and Uber would show up 20 seconds later, right? It's like at some point it just was like taken for granted and then at that point if you just said literally are we going to take Uber away, people would have said no, we can't and it's over and that's what happened. And then later that happened is they changed the laws to accommodate that behavior. And so I actually think part of it here is just like having these tools, okay here's a thing, here's a good news thing. These tools are becoming widely available up front, right? So like 50 years ago, a new technology like this would have been like deployed in the government first and then in the big companies and then

years later in the form of something individual people could use. The model today is like it's just online. Yeah. Like GPD's online right now. Yes. Well, the future of paint is really intriguing because from an engineer point of view, it's the engineer's dream that if we make it good enough, such that it can get to a point where people just love it and it's helpful and it does what it needs to do, the rest will take

care of itself. Yeah. I mean, I kind of think that's mostly how things out. I mean, yeah. No, that's a beautiful feature. Yeah. So now, look, having said that, healthcare is very sophisticated, right? There's lots of regulations.

There's lots of payment, right? All these things. So I saw this thing on the front of the other day. Yeah, it blew my mind, right? Because this whole time I've been thinking in terms of, like, you know, diagnosis and stuff in my life. So this doctor posted a video and I think I saw that. We saw this. So that first a video and he said, look, he said the problem it is for whatever diagnosing.

So whenever, like, I do the diagnosis. I do the prescription. Then it's a question of whether or not I can get the insurance company to reimburse to pay for the thing. Yeah. To do that for anything even slightly out of the ordinary, I have to write a letter, the doctor, just to write a letter of the insurance company. And that letter needs to be in a specific format and it needs to make the case, right? Make the case. And it needs to have the scientific citations. Yeah. And if I do the letter really well, it's going to get paid for. And if I don't do the letter really well, it's not going to get paid for it. It's going to matter, you know,

to the possibly the life of the patient. Yes. And so he's like, it turns out GPD is really good writing this letter to sit. With the references. With the references, yes, with the scientific references, like full on. Yeah. And so you've got this. So that's another way I think about it is you've got this bureaucratic process, which is legitimate and required and yet to exist. And that data needs to be submitted. And honestly, it does not matter to that process whether that document is written by human or machine. Yeah, yeah, yeah. But all of a sudden, if every doctor in the world is really good at writing correctly, properly, yes, letters, then as a sudden, it goes to the thing. All of a sudden, that doctor now is another, you know,

whatever, four hours a week to take care of patients. Yeah. Like, that's the kind of thing that I think is going to happen more uniquely. And that, what's interesting about that example is you can imagine that example having a big impact on the efficiency of the health care system today. without any regulatory changes. Yes. Without, within the system. Within the system. Actually within the system.

Yeah. And so, and that was the one where it's just like, oh, in retrospect, that's obvious. I just haven't thought about it. Yeah. One guy thinks about it. All the other doctors start to do that. The whole system upgrades, stuff function, you know, one time. Yeah.

That kind of thing, I think, is a real possibility. Yeah, and that could be because someone's working within the system, you can have the transformation immediately. But then eventually someone has to read all those letters. But someone has to validate them. It's probably, you know, some sort of NLP on the other side. Well, that's right. It's corresponding. So there's, we have this company,

this company called Do Not Pay. Yeah. Yeah, she's this company. It's an app that sort of acts like a bot. Yeah, no, I'd use the app. It's for people to try it. And it basically, it will basically get you, it started to get you out of, it was started to get you out of like basically

fake done sort of BS traffic tickets. And then it's a, he did this thing a while ago where it will unsubscribe you for, you know, all these consumer subscription services like Comcast or whatever. Like, they all make it hard to, like, ever turn off the subscription. And so he has this way to, the bot will do it for you. And so he just started using AI and the bot. And so the way a lot of consumer subscription companies work is you can't,

you can't actually unsubscribe online. You have to call an 800 number and you have to argue with a person. And there's actually this thing in these companies called Save Teams, where they're actually paid specifically to prevent you from unsubscribing. And they'll try to cut special deals with you and they'll try to talk you out of it. And so he has this thing wired up where now he has an AI generated text with then text to speech. Oh, it's just talking. And it talks, it talks, it talks, it talks to the customer service person

into the line. And it basically, with infinite patience. Yes. And so it will just sit and he will just argue like, no, I am actually going to us to arrive for, yes, no, I'm not running. Yes, no, no, no, no offer. No, no, no, no. Right. Exactly.

Until finally the other guy, by the other guy gives up and says, okay, fine, I'll stop charging you. And so it's like, okay, you know, it was a precondition of the system that that worked the way that it did. it was a burden on people to have to deal with that. A, I can now step in and equalize the power imbalance between the customer and the company. And presumably that will change the system. Yeah, well, one would think. Well, one would help, right. And to your point, like, Step 1 for changing the system might admit retaliation,

which is all of a sudden the save teams will be bots. And so maybe the boss will be arguing with the bots. But at least it gets you out of this kind of Kafka-esque thing you're in today, where when you deal with these big companies, you're dealing with this giant bureaucracy, your individual dealing with the giant bureaucracy. At least it, like, equalizes the power. Well, that's kind of amazing.

And that will be the spark for changing things. Because once you're in that sort of system, like, we got to do better than it. Yeah, this is crazy. Right. Yeah, yeah. Yeah. And especially with bonds on both sides, now we can finally say, well, let's do an API on both sides. So let's do something smart on both sides.

Yeah. Yeah. Yeah. Well, Mark, I mean, that's such a sort of beautiful, optimistic view of how this could go, right? Because the future we're talking about is actually much more engineer driven that if an engineer can build this, and it really, really works. It really helps patients. It really changes things. It will get adopted.

As it gets adopted, cultural work around it and will love it. And we'll love it. And then the future will just be right in front of us. Yeah. Patients are going to get a vote. Yes. Doctors are going to get a vote. Yeah.

And, you know, it's an industry native of people. A world man, people will get a vote. Yeah. Beautiful. Thank you so much for joining. Thank you for joining BioEats World. BioEats World is hosted and produced by me, Olivia Webb, with the help of the bio and health.

team at A16Z and edited by Phil Hegeseth. BioWeets World is part of the A16Z podcast network. If you have questions about the episode or want to suggest topics for a future episode, please email BioWeetsworld at A16Z.com. Last but not least, if you're enjoying BioWeets World, please leave us a rating and review wherever you listen to podcasts. Please note that the content here is for informational purposes only, should not be taken as legal, business, tax, or investment advice, or be used to evaluate any investment or security and is not directed at any investors or potential investors in any A16Z fund.

For more details, please see A16Z.com slash disclosures.