No Prep Needed with Noah Healy

And do you stream? What do you use for your programs, your podcast?

Um, uh, dang it, I can't remember right off the top of my head. It's, uh, it's, it's one of the multi-channel, um, uh, recording platforms, but there's like 3 of them, and I— the, the name of the one we use isn't springing to mind.

Um, no worries, no worries, we'll get that, we'll get that information for the podcast. All right, so we're doing it again, another second of today's No Prep Needed. Noah, have I met you before?

You have not.

I don't even know where you are with that beautiful bookcase in the back. I'm in Daytona Beach. Welcome to our live stream, No Prep Needed. This just shows you when you meet someone and you get to know them It's okay, you sit down and you have a conversation. The only thing we're doing is we're allowing each of you to be flies on the wall as Noah shares with me who he is. I can tell you a couple things. Again, I know he's bold and he's brave. He's not concerned because he actually has his own podcast, so he's been doing this for a while. And I love the fact that he trusted me enough to say yes, Adrian. So Noah, I want to give you the pleasure of introducing yourself. Because I don't know you.

Sure. Yeah. Well, Noah Healy, uh, I'm in Charlottesville, Virginia, uh, and as, as we've got going on the banner there, I've got a podcast called The Fourth Age, uh, where I discuss AI and its implications for our society, economy, and politics with, uh, my co-host Marty Weiner, who is the retired CTO of Reddit. And I'm also working on a patent I developed to change how marketplaces function and make make production radically more profitable than it has been for the last 800 years.

Okay, this is so fascinating to me. The only— like, it's weird, around last year around this time is when I first heard of like Jasper, ChatGPT, and AI. I, I'd never heard of it before, and I feel like at that point I was behind the, the game, right? When did— when, when was this all developed?

Well, technically, the The entire concept of the neural net goes back to the earliest days of computers themselves. One of the first proposals for how to actually build an artificial computational mechanism was sometimes called something like the synaptotron or something like that, where you would model human neural synapses. And the first time there was a period where it looked like that approach might actually be bearing some fruit was late '70s, but mostly early '80s. And those companies didn't really wind up producing very much of value. And so it was known as the AI winter. But in the late '90s, there was a minor but fairly public incident of IBM building the Deep Blue supercomputer and defeating the then world champion in chess, which was a bit of an outlier. It took a while before chess programs got to be actually better than people in general, other than purposely built supercomputers designed to defeat a single human being. However, a few different breakthroughs in essentially unrelated fields combined with a at the time fairly quixotic sinking of tens of millions of dollars into research led to these generative AIs like ChatGPT and Jasper and now Gemini and these other ones you've mentioned that sort of suddenly took the consciousness of the public by storm about 2 years ago now.

I feel like that I'm a year off. So chat boxes, or chatbots, they've been around for a while. They're AI.

They are what people generally think of as AI these days, and they're radically more sophisticated than the bots that existed before. But there's a bot known as ELIZA that goes back again to the '70s that was based on a speech pattern of a therapist. And so you could basically make any statement you wanted to Eliza. You could say, you know, "Hi, I'm Adrian Barker," and it would say, "How do you feel about being Adrian Barker?" And it would just keep doing that kind of, you know, therapy thing. And it could have conversations with people. It couldn't have detailed conversations or bring in facts And it was, it didn't hallucinate nearly as effectively as these modern chatbots do. But again, we're looking at technologies that have been showing off their existence for, in some cases, half a century. And what we're now seeing is that these things are moving from toy stage to kind of real stage, and we'll see if they can, they can break through to general use. I'd imagine they probably will.

That is so interesting. Actually, about this time last year, I, I got a fun app for myself and my daughter, and it was like a fake person that seemed real, and you could ask questions, and it was for the iPhone. And I remember thinking, this is like ridiculously cool. So All right, I have a question for you though. With— and I know you invented something, I'm like so excited, I get very excited, I'm ready to like push out 100 questions to you. But my one question is this: if there was— if we had 25 people in a room and we were all using, let's just say, ChatGPT, and we all asked the same question, why is the sky blue, would all 25 people get the same answer?

Probably not. In fact, the way many of these things are designed, essentially, certainly not. The one way to think about how these work, if you're familiar with sort of pachinko machines, the, the, there's a, a game that they show off in The Price is Right where somebody takes sort of a hockey puck and drops it down a bed of nails and it sort of bounces around. Plinko, that's it. Yeah, so some of the mathematics of how this thing decides what to do bears a very strong resemblance, um, that's, that's known as lattices in, in the mathematical term, uh, to how Plinko machines work. And because of how many arbitrary choices there are to make and the amount of arbitrariness that they plug into those things, um, that it will— it, it won't be circumstantially— basically, the 25 of you can't all ask the same question because they're asking from different devices at slightly different times. And as a result of this combination of other factors and their decision to design it so that it would be sensitive to various things in order to not just mechanically be slavishly repeating itself or hackable in that easy way. It's very unlikely that even two of you would really get the same response.

I hope everyone heard that because that is so interesting because I think that I use ChatGPT every day and my son and I tried it with the artwork too, with the DALL-E. Like I said, okay, let's— we typed in the same thing and we both did come out with completely different photos, but I was still like, well, what if there was more people in the room? Like, it wasn't enough of a test for me. But I love this because in my mind, people are like— so those that haven't used it are the first ones to say, oh my gosh, you know, it's going to be thinking for us, it's going to be taking over, which is my next question. But the reality is, I, I may use it, it may start off, but I'm still massaging it and taking time to make sure that it's giving me the right answers. And I don't feel like something like what if I'm looking for topics to write about, I feel like it's still coming from me. It's just helping me to dress it up a little bit. So like what would people say that, Noah? Oh, I don't like ChatGPT or any of that. It's going to take over the world. What's your answer?

Well, I think it's obviously the case that it will take over the world. However, I don't think that people are treating it in, the term that I use for the generative AI is that it's less consequential and more dangerous than people generally regard it as. So when industrialization was getting going and engines were happening and people stopped doing as much day labor because engines could do quite a lot of that for us, there was not a sort of general pushback that machines were taking over, but there were some serious concerns with urbanization. People looked at peasants getting up off the land and said to themselves, well, you know, 80% of the people working backbreaking farming, subsistence farming jobs is where all the food comes from. So if half of those people leave and go to the cities for these good-paying manufacturing jobs, then, you know, we'll have half as much food and half of us will starve to death. And that's a death spiral. So this is all terrible. Of course, we figured out how to use engines to make backbreaking farming labor more effective and produce a lot more food. Um, that's the kind of changes that we're likely to see. However, because these things are producing stuff that looks like or sounds like human opinions and ideas, the risks that we are currently up against is that our institutions, from our politics to our economics, are based around the idea that everyone that's making statements and having opinions is a person, and that the number of persons that exist is relatively static. It, you know, takes 18 years to turn a baby into an adult and so on. Um, I was talking to some recent college graduates who were highly successful and are now running a multi-million dollar company that they founded while they were in college. But I was pointing out to them that one of the reasons that sort of good jobs are available for people such as themselves is that they're relatively scarce. College graduates aren't the majority of people, and college graduates with the sort of gumption and go-getting to found their own multimillion-dollar companies are obviously a very small subfraction of the college graduates. Uh, but as has been demonstrated multiple times, ChatGPT can produce collegiate-level essays in seconds. Um, and, and that's just on the $20 a month plan. So in a world where there's a few, say, tens of millions of college graduates every year, there'll be a certain value attached to that. In a world where computers can manufacture trillions of college graduates every year, that value will naturally be radically lower. And the values in society that are being or going to be radically lowered are the ones at the tops of our societies. It's much easier for these things to simulate what CEOs do or what politicians do or what entertainers do. And, and it's a lot harder right now for them to simulate things like what cab drivers or garbage men do. Um, and so our society is going to have to change in ways that makes us able to use these capacities in ways that are actually valuable to society, because mostly what it's it's done so far and what it will do, um, until we look into these issues is just make it a lot noisier and crazier.

That is so— oh my gosh, you're brilliant. All right, here's another question. I, I'm doing my senior essay in college. I use ChatGPT for the whole thing. I then get a little nervous and I say to ChatGPT, hey, write like I'm a 19-year-old female and then add a few mistakes so that nobody really knows that it's real and make sure that I can't be found anywhere and help me to make it sound like my own. Will the plagiarism pick it up? How do the professors know when it is ChatGPT compared to that it's not? Like, what does that look like? Because I get very confused when I read that they can figure it out.

Well, in some cases, they'll be able to work it out. Because they might be able to build models that can detect whether or not there is plagiarism. As, as these models develop, there'll be things going on. OpenAI may— they certainly haven't announced that they're doing this, but any of these companies might have certain embedded characteristics that they decide to sign. For example, printers, due to regulation, have microprinting intrinsic in them so that The outputs of printers are traceable back to specific machines because of the microprinting on each page, uh, and that's also something that exists in computer chips. Um, my laptop has been unbound so that I can operate a real operating system on it, but the standard laptop computer chip has something called the Trusted Computing Platform, which was pushed through by a number of industry bigwigs that allows the chip on your machine to understand whether or not the software running on it belongs to an approved person, and if not, to not run that software. And that's basically a Microsoft digital rights management technique that is built in at the chip level so that customers can't pirate their stuff. And there's other people that were also interested in being able to control technology that nominally belongs to consumers. It would be possible for these companies to be behaving this way. They have not announced that they are doing so, but they've certainly done nothing to ensure us that they aren't at least that nefarious. But that's more or less the precise issue that you've put your finger on. Because if it were developed a method to work out that people were plagiarizing, that methodology itself would be the basis necessary to train new sets of generative AIs that could evade that, that would be able to do their jobs even better. And so I go to an example from my collegiate experience. I went to UVA, and we have an honor system which as part of its action, it's strongly discouraged for professors to even monitor students during the giving of exams. There's a handful of very notorious cases where it's actually done. Basically, sort of Physics for Jocks was sort of a notorious bed of cheats, but For the most part, students are trusted and encouraged to be trusted at every level of the university. One of my professors, who was the dean of undergrads at the time, felt that this was a perfectly decent thing to do and actually sort of wonderful, but on the other hand, training engineers didn't feel that it was perhaps wise to simply open the doors to cheating, and so he designed his exams so that they could not be in fact cheated on. And so his exams were open book, open note, week-long systems where you could solve the problems yourself or with sufficient research, maybe, well, find the sources of the problems and simply copy down the work. However, because you, you had effectively infinite scope and scale, the grading was extremely tight. He would grade down for misspellings or poor penmanship or basically any error at all because the object was to produce finished engineering work under general engineering conditions. And you could certainly start thinking along those lines in a world where It's possible to credibly pretend to be a skilled practitioner of any field because ChatGPT can do that for you. How would you go about interacting with people such that it doesn't matter whether or not they have the ability or ChatGPT is providing the ability or anything, you can instead distinguish whether or not the value being produced is real or, or, you know, simply a very convincing fake.

So when I was growing up, we would go downstairs, we would get the big huge dictionaries, right, our encyclopedias, and we would take the encyclopedia and, oh, I have to write about Abe Lincoln, okay. And then I would like, I pretty much took it word for word and then just switched things out because I figured, how would they know, right? I'm just a kid. They wouldn't really know that I'm using an encyclopedia. And but sometimes they would say, did you use an encyclopedia? And I didn't lie. So I was like, yes. But like, same kind of thing, right? Like, I always, I always felt that as long as you knew where, how to find the information, how much can you store in your brain anyway? So I mean, am I off in my way of thinking?

I, I would certainly tend to agree with you. I don't really see a moral differentiator between a person that has temporarily stored information in their hypothalamus and people who have temporarily stored information on their inner thigh, uh, in order to pass an exam. Um, uh, I understand that there is a very big practical difference in existing, uh, uh, you know, educational practice, and, and that's one of the points that I very strongly disagree with, and I think ChatGPT and some of these other generative systems is going to very much point up these problems. And it points again to one of these issues that we're going to have to cope with societally because it really is important to find high-skilled, high-capability people to work on our hard problems because we actually have hard problems and we get better results out of very smart people. Than we get out of very stupid people. And so if we lose the ability to tell the difference, then, you know, things won't go very well. But on the other hand, to the extent that we are capable of using these technologies to augment our capacities, failing to do that is also going to be a big negative. Less well known, But these, another kind of use of these neural nets is a thing called reinforcement learning, which, or occasionally deep learning, which is capable of, for sort of known systems, creating very deep understandings. And some of the outcomes of deep learning technology have been quite extraordinary. For example, Google released a paper, maybe 4 or 5 months ago now, where they announced 220,000 new stable chemical crystalline structures. And in all of human history, we had developed somewhere between 50,000 and 70,000 known stable chemical crystalline structures. So they sort of 5x'd the number of known types of crystals. And something that's important to understand about crystals is that we have historical precedent for not being able to actually find these things experimentally until after we know that they're real. So there were a large number of crystals discovered during the 20th century as mathematical— going back to that lattice thing again. As the mathematics of lattices and also irregular lattices really advanced, we gained new abilities to understand new kinds of crystalline structure. And when we did that and then started doing the X-ray crystallography to actually look at the universe, we suddenly discovered that many well-known minerals actually had some of these obscure crystalline structures that we hadn't known about because we didn't know how to interpret the data we were looking at.

Wow. Okay, hold on for a second. A lot of that went over my head. I'll try to catch her. Wow. Okay, I do want to go into that you have a podcast and that you're a venture, but I have to ask one last final question on this because I know, like, I could talk to you for hours. I read somewhere that I first— I read 2050, they said that the AI systems will be smarter than we are. And then someone said, oh no, like you could say 2050, but really 2030 will probably be there where these systems will be actually smarter than a human brain. Is that yes, no, true?

Okay. Well, I would once— what I usually tell people under the circumstances is that you've got your tenses wrong. So another fairly impressive use of deep learning, and this is in combination with some of this generative AI stuff, There's a test, a math test given out around the world, and it's occasionally known as the hardest math test in the world. And it's given to kids for the most part. The scores go from 0 to 100, and the average winning score is 25.9. The average second place score in the world is 23. An approach combining deep learning and generative AI, um, along with a computer program called a proof, uh, a proof solver, uh, was able to get 25 on this test. Um, so strictly speaking, not smarter, you know, not smarter than the smartest human being of any given, you know, age, age cohort, um, but 94% as smart, 96% as smart as that, and smarter than the second smartest human being. Um, uh, so the, the what we think of as intelligence itself is one of the things that's on the chopping block for these systems.

Oh, that's interesting.

Chess, for example, has been regarded for centuries as a measure of intelligence. And as I said, in 1997, the then world champion lost in a 6-game series against Deep Blue. Now, he wasn't completely crushed. He won a couple and he lost a few more and there were some draws. The modern chess engines are something like 800 points stronger than modern top-level super grandmasters, world number 1, world champion, those guys. They're, they're 800-ish points. So put this in perspective, chess masters, basically, if you're in a, if you're in a town or a reasonable city and you've got like a chess club and they're not that serious, they're the best player at that chess club is about 800 points weaker than the best player on Earth. This is, this is another 800 points stronger than that. And these chess engines have been well beyond the capacity of human beings to play chess for decades now, and they're getting better at chess. So, so chess is enjoyed a big revival with COVID and everybody being online and getting to play chess as a hobby and so on. There's a lot more chess being played. The best chess players seem to be getting a bit better as a result of more competition, more play, and so on, although possibly also because of cheating. But the machines are also competing against one another, and the machines are getting better at chess faster than we are getting better at chess.

Wow. Okay. And I get that, all the moves. But one more question, and then we go into where your, your invention— where does wisdom fall in on this? Like, I always felt like wisdom was very important and something unique and different to each of us.

I think wisdom is very important. And I think that, again, I think hopefully we'll be able to call upon reserves of that in order to survive the transition of our societies to new kinds of societies that can cope with these things. Um, but again, it's, it's not outside of the realm of imagination, uh, which is the realm of imagination is the precise realm being accessed by computers that it would be possible to come up with a framework for wisdom and produce wise systems. That is absolutely not what anybody's currently focused on, but super-wise intelligences are a mythological and actually sociological point of human history. Things like the Oracle at Delphi were supposedly always right and always sort of put people on the paths that they belonged on. And it's not very difficult to sort of imagine a sci-fi scenario where the ChatGPT at Delphi could become a real thing somehow, although probably not with the current cast of creators.

Thank you. That's interesting. That— wow. All right, so again, we haven't met everyone, so just so you know, this is— I'm like blown away. Obviously Noah is like super, super duper intelligent. Like to put you up against a ChatGPT, Noah. You'll kick their ass. Okay, but seriously, tell me about the invention. And what is it called? The CDM?

Yes, uh, CDM, Coordinated Discovery Markets. So Basically, the marketplace exists to figure out the price that everyone that wants to sell and everybody wants to buy wants to buy and sell the same amount of stuff. So if you're just kind of like, you know, setting prices, you could just set prices wherever you want. You'd be like, wheat, cool, $100 a bushel. Well, people would be very interested in growing wheat at $100 a bushel. That's a lot more than you get right now. But bread makers are not gonna be very interested in buying wheat at $100 a bushel. They might not be able to sell those loaves of bread at $500 each. So, you know, suddenly we're transforming our, you know, cities into wheat farms. And at the same time, we're shutting down our bakeries. And like, you know, there's a major mismatch. Or you could say, well, wheat, let's— a nickel a bushel, that seems fair. Well, maybe it seems fair to you, the bread-buying public, but the farmers aren't going to go into that business. They'll grow rice or corn or something else instead. So there's a price point, um, and that price point is the result of the sort of co-negotiation of everybody in the marketplace. And, and we currently develop a marketplace where it's based on individuals negotiating private prices, which are then published to everybody. And the problem with that is that most of the information stream is actually robots trading with other robots. So going back to that, are they smarter than us? You know, how will this affect our societies? We have had quite literally AI systems based on many of these same techniques that OpenAI is using, trading as the majority of the volume in our marketplaces for decades. And that's led to a lot of the market instability that we see where crazy things happen, sometimes so crazy that they become national news or international disasters. The way my system works is by integrating in multiple viewpoints and providing a stable price to entire sectors of the marketplace to then see how much they want that price. So rather than being rewarded for being able to find somebody willing to take the worst deal you can come up with, you are instead rewarded for contributing towards the best deal that everybody can come up with. And the more you contribute to that, the greater your reward.

Wow, that sounds brilliant. So what— so if we go on to change.org, what exactly are we going to— how can we help you?

What, uh, well, so I've been working on this patent for actually around 9 years now, and the Patent Office is breaking its own rules, lying to Congress, and doing everything else it can to slow down and halt the adoption of this technology. And right now, it's been appealed. It was appealed last year, and it's been scheduled to be heard July 2025. And so between filing the appeal and waiting for it to be heard, there's not really much I can do. But about a month ago, somebody suggested, you know, I could do this. I could, you know, create a public record of people's being interested in having a better economy. And so that's there for people who would, who would like there to be better markets or, or indeed a patent system that wasn't behaving in a very, very untoward fashion.

Okay. All right. Thank you. Thank you very much. So go on and check it out, everyone. The link is there and I'll put it in the show notes. And you are a podcast host. So tell us about your show.

Uh, yeah, so it's called The Fourth Age, uh, and, and subtitled The AI Revolution. This goes to my thing is that the plow, the steam engine, and the computer are all technologies that change mankind's physical interaction with the environment, not just economically, but also politically and socially and even religiously. The, the religions of pre-agrarian cultures do not resemble tightly the religions of post-agrarian cultures. The religions of industrial nations do not resemble the religions of agrarian nations. So since I say that the computer is a bigger deal than the steam engine or the plow, I would expect our political, cultural, social, religious, and other institutions to also transform at least as much as, as they have during the previous revolutions. And that, and that if, if we're going to have civilization here in the Fourth Age, uh, we're going to need to grapple with these questions. So the podcast, um, it's so far it's about half and half discussions between Marty and I, and some guests, uh, are discussing these sorts of issues where we see AI having some of these impacts, uh, what kinds of outcomes are happening, and how, how we can use wisdom and also mathematics, uh, to try to address these issues.

Okay, and Marty is—

Marty is the, uh, retired CTO of Reddit.

I love Reddit too. I, I, I'm on it. I have the app, and I And but what I've noticed is a lot of questions because Google is now more like an answer site than a search site, right? So a lot of times when I ask a question, it'll put me on Reddit. But do you agree with that with Google?

Absolutely. So Google is another very interesting example of the production of an unstable superintelligence. So Google search back in the late '90s, early 2000s was fantastic. And it wasn't just because it was free and didn't have advertising on it. It was also because the internet hadn't reacted to the existence of Google search yet. And so Google's results were very much on point. Uh, but with the domination of search as how people interact with the internet, it became necessary to engage in search engine optimization. Rather than pursuing content as your core methodology. And what happened is Google search algorithm has becoming worse and less effective because it exists basically. And this is the fundamental problem that we're going to have across all of these types of deployments. If ChatGPT floods the zone with erudite, you know, text and speech, uh, and then sort of turns journalism and communication and sales into an undifferentiated mush of noise. Well, then society is going to disintegrate because nobody's going to actually know anything or be able to trust anyone anymore. Um, So it's only by creating systems that remain stable after they exist that we can, we can actually exploit these things on a durable basis. And so that's the challenge.

Wow, Noah, thank you so much for your time today. I feel, I feel bad we went over, but I really appreciate talking to you. It's incredible. And I'll put all the— can you send me an email with all of your information so I can put it in the show notes? Really appreciate your time. And even though a lot of it did go over my head— wow, you are, you are super intelligent— I learned a lot too. So thank you so much.

Thanks for having me here. This was great.

I appreciate it. I really appreciate your being bold and brave enough to come on with no prep needed. Thank you. Have a great rest of the day. Where are you located?

Charlottesville, Virginia.

Oh, beautiful. Okay. I'm in Daytona Beach, Florida. Awesome. All right, everyone. Thank you to Noah, my new friend. I don't know if he'd want me as a friend. I'm not that smart though. Okay. Thank you, Noah. Have a great rest of the day.

Bye-bye.

Thank you, Noah.