The Exchanges

Q And it wasn't just about the political stuff?

A Oh, no, it was primarily about trying to, like, what I wanted to do from the start, but there was no direct way to do it, because at the time, in 2015, I don't know if you remember how bad the chatbots were, but they were really not very good. So for me, a lot of, like, I wanted to Already built bots to teach people stuff like how to breathe, how to meditate, how to exercise, uh, teach. There's like 40, 50 techniques that are very well established. Decades of research show that they work. And often as of these 40, 50, there's going to be two, three, four that appeal to a given person. You have to find the right one for that person. And, um, so we kind of know a lot of things that work very well, but how to deliver that in a format that's interactive, that's not boring. That's so I thought Chatbots would be a very good way to do this, but they have to not be boring. And at the time, the chatbots were not very good. So that's why that part of what I wanted to do, uh, in practice, it translated into me doing research on conversational AI to try to make the bots better. Uh, so that was one of the main things I wanted to do. But for all the parts about, um, misinformation and polarization, that's something where I felt like, oh, I can do something right away. And then I looked at the research and like, oh, actually, no, not really. But so that was already, um, Part of what I want yo…

AI assessment note: “Oh, no, it was primarily about trying to, like, what I wanted to do”

Q interesting that that is your perspective because we just saw some research come out. That shows that the number one use case of chatbots today has gone from idea generation to companionship and therapy. And it seems like you're not taking that path and you're trying to have an alternative where you're trying to spark human connection as opposed to have the bot be the connector itself. Why is that?

A So it's first is I'm completely okay with just about teaching you lots of techniques. I think it can work very well, but I think a lot, if you want to, uh, enlarge that first circle of people who can just learn from, uh, first thing, lots of people can learn just from books. Those people have already learned those techniques if they're interested in it. Like I, I read so many books on this. If you read all the books, you know, a lot of things already, but, but that's going to be a narrow set. Then you have some people who will learn just from about a little game, a thing like you can think Duolingo style. Okay. Perhaps. If something is more, like, bite-sized, I can learn, and I still enjoy doing it single-player mode. And the thing that has been shown time and again in research is that you need some kind of accountability, and accountability works better if you have a human reference in the loop, someone who keeps you to your commitments, someone who checks in on you, and someone who, like, for example, if I have an appointment with a doctor or a personal trainer or a friend, I don't just, like, Not show up and that's okay. If I don't show up, I have to warn them or I have to, uh, I have to like give some kind of, of, of, of excuse or warning or something with a bot. If you don't show up, it's like nothing happens. You can, you, you don't have this whole social expectation thin…

AI assessment note: “accountability works better if you have a human reference in the loop”

Q Yes. So, so you have a company that is attempting to make this happen, to sort of nudge people into moments that are into habits that would make them more fulfilled. It's funded by OpenAI. I think that's public, right? We can say that. Um, tell us a little bit about what you're trying to do with ThrivePAL.

A So it's funded by OpenAI and General Catalyst, mostly, and a bunch of other people. What we're trying to do is, um, is to give people the tools that are very science-backed that we know work and help people, um, discover them and experiment with them and find what works for them. So those kind of tools, they've existed for a long time, but not everyone's going to be receptive to the same one. So a very well-known one is meditation, for example. If you look at meditation, Um, you're going to find lots and lots of research that shows that it can do great things for people. It doesn't mean that everyone is going to find that meditation is the answer to all their problems. It means it's worth trying, it's worth learning about, it's worth playing with and experimenting with, and if it works for you, you keep it. If it doesn't, you try something else. Um, so there's a bunch of these things that are powerful and work for some people. Um, and some things That even if they don't work for you, you kind of have to find a way to do that. Otherwise you're not going to function so well because until we get uploaded to cloud or whatever, we live in bodies, uh, how to take off your body, for example, exercise, physical things.

AI assessment note: “What we're trying to do is, um, is to give people the tools”

Q So how are you going to do this?

A So, uh, first thing is, uh, I think there's several levels of learning. The first one is even knowing that's the thing. So for example, if I tell you I have groundbreaking information, if you smoke, you might increase your chance of cancer. You're like, oh, I know that everyone knows that. That wasn't the case a few decades ago. So first getting the information out, there's some things like the thing around the role of muscle for metabolism, where I think the information is not out yet. So that's level one is like putting the information out for some of these things. Um, like even before you finding it's painful to do something, knowing what it is that you should do and aim to do. So, um, like, yeah, step one. So this, uh, for the microbiome, I think there's lots of things where people are not aware of how important it is to take care of the bugs in their, in their gut. Like the fact that 70% of your immune system is in your gut and that it reacts so much to what you eat and That perhaps if you have a choice between taking this antibiotic and not, and you're just taking, like, to be sure, like, just wait a bit, know that there's a downside, that you're going to completely wipe out all the bugs, uh, in your gut, and it can cause a lot of downstream problems. So, uh, I think there's a lot of information that still needs to get out. Level one. Level two is knowing, once you know, …

AI assessment note: “first thing is, uh, I think there's several levels of learning. The first one is”

Q And so how do you, how do you, um, anticipate that? And how do you plan for that as someone who's building an app yourself?

A Yeah, no, no, for sure. Um, so first thing is ChatGPT knows lots of good things, but ChatGPT is very permissive and not so opinionated necessarily about what's good. And so we give a lot more structure as to like, okay, these are the things that you need to think about. This is how you organize your thing. These are good techniques. Like we give a lot more of, um, uh, we have a strong stance on what the good directions are. Um, so that's kind of. Way more curated and same, especially for the game where these are all going to be what we provide. And if you go into our app, there's this thing called like instant insight, which are little techniques that you can learn about. I decide what's in there. Um, so it's more like this choice, this editorial choice. So it's like editorialized chat GPT and also the way I really ask it to be science backed. And I check like all the literature myself because I've read so many papers I've published in neuroscience and like theoretical neuroscience, these kind of things about how the brain works. So making sure that, uh, this is reliable, which ChatDpt is very happy giving you bogus information sometimes, because some people believe it's true, um, and science is complicated, so I trust myself way more. So I, I do that work to, um, to ensure that what we put in there, um, is in the right direction.

AI assessment note: “it's more like this choice, this editorial choice. So it's like editorialized chat GPT”

Q And amplification is the right word because there's scale online, right?

A Well, I mean, especially for me, it's not just online is, uh, because, um, like what I'm, I'm 40 now, so 40 something. So, um, I grew up with the beginning of the internet when I would go on like, let's get like net scale and have a model and we're like, and then you would like start to load and then you would like go back, have a snack, come back like 20 minutes later and you had like 2000 of the pitch loaded. It's like that was online at the beginning. Then we had a bunch of things. And you had to be at home, right? So I think for me, what really changed is the mobile phone because suddenly you're always online. So it's not just online versus offline is always versus in boxes.

AI assessment note: “what really changed is the mobile phone because suddenly you're always online”