Transcription
I'm a statistician. So AI is a statistical machine. So statistical machines don't extrapolate. They can't. So they are almost by design made to not extrapolate.
And then the other thing about AI is they think that it's going to save the US. No, why not save Pakistan? I mean, AI is so, uh, universal.
I noticed when I was after my paper, I was attacked a lot by a bunch of Bitcoiners. No, you know, there was it was like a wave of kids. You were into Bitcoin. Now they're all into AI. It's not a currency. It's used. Probably it's not even useful in Syria. Think about it. You don't have the internet. How are you going to use Bitcoin? And every time we've had a crisis, an economic crisis, it does exactly the opposite of what was expected from it. So, it's a speculative thing. Speculative stuff by definition. Never fight something speculative.
It's a big mistake to think that anti-fragility is an idea that explains bouncing back from disorder, getting better. Because these phenomena were well known, particularly by doctors. So there are a bunch of, uh, what I call seemingly disconnected classes of phenomena. One, two, three, four, five, six. And these are united by a simple, extremely simple mathematical framework. And that ranges from bouncing back from disorder, going to the gym, you get stronger when you go to the gym, and, uh, how do you have optimal dosage in oncology? Okay. Without knowing anything about about about chemotherapy. So the last one I call upregulation, the effect related to benefiting from stressors. Okay.
In medicine, they know something called hormesis. Okay. Where you give a poison and someone gets better after a very small dose of poison. Very, very, very dose, uh, response related. Post-traumatic growth in psychology. In fact, when you go, when you spend time with the drones up here and talking to Ramy's Ray, my classmate, or stuff like that, you know, and then with the stress of who's going to shell next, some people develop trauma, but many more people get stronger. It's called an upregulation response, and that's not in literature. It's rarely in literature because nobody's going to make any money selling you or trying to cure you post-traumatic growth. Okay.
So, uh, hydra-like outcomes in mythology where you cut one head, more grow back, and beliefs about rebounding from adversity. When life gives you a lemon, you make lemonade. So, that's the first class.
The second one is what I call benefiting from variance. Pholostacity, where you benefit from some kind of irregularity. Like I was told when I was a child that you should eat many, many small meals a day. It turns out that it gives you diabetes. Okay. So it's the same amount. You should have episodic fasting. Now, intermittent fasting is part of the cure for diabetes. And, uh, but at some window, like eating once a day or twice a day is okay. Eating once a year, you know, you're not going to make it. Okay. So, even in Lebanon. So there is something about the benefiting from irregularity. Like, for example, before COVID, when we wrote about our first paper on that, before COVID, there was an observation made by doctors that if you give someone a ventilator, you know, the ventilator that kills people, and they think it cures them. If you give the ventilator, if you give 100% of the dose all the time, you don't get the same outcomes as giving 80% or 120% of the heated dose. So irregularity. So this is, this is what we call, uh, benefiting from this. I call it pholostacity, benefiting from variance and dispersion. By the way, it's exactly the same equation for all of them. Okay.
Uh, stochastic resonance in physics and signal processing, where you put some noise in a system and the signal rises. Bit technical. Uh, and then, of course, long volatility in finance. For those of you in finance, that was my profession, is benefiting from crashes, right? So from disorder in markets.
Class three, what I call scaling. An elephant looks like a mouse. I mean, we don't have elephants in Beirut, but we have mice, I'm sure. But you can imagine, but they're not the same. They don't scale in the same way. An elephant is not quite a large mouse. The, as you grow for stability, you need to have, you know, your contact with the ground is in squares, but you grow in cubes. So you have some scaling isometric effects. You're actually, and, and but nevertheless, you're still going to be more fragile as an elephant than a mouse. An elephant has a, you know, breaks very quickly. This is why we don't have a lot of elephants. And I'm sure in New York, we have more mice than we have Lebanese in Lebanon. Okay? So, and that's, you know, not counting every month because they don't vote. So you don't count them. You don't have the incentive.
So these are the three classes of events on a positive side. Now, the opposite qualitative attitudes going to be the same with negative sign, effects related to breaking under shocks. Now, the interesting thing, now this is not fragile, but I have my cell phone. So it breaks under shock. It doesn't get better if you have a shock. Okay. The, um, and typically it's, it requires nonlinearity. So, in other words, 10 shocks of intensity one will have no effect. One shock of intensity two will break it. Okay? So, I mean, don't try cell phones are expensive, and plus you have to pay that thing to the Lebanese state, you know, that tax. So don't break your cell phone.
So, class four, harmed by this second-order effect at some window, harmed by dispersions of outcomes. In other words, uh, eating, for example, continuously may harm you. Having a drug at too low a dose for a long time harms you with no benefits. Um, and there are a lot of, uh, things that are harmed by outcomes, banking portfolios.
Now, class six, effects linked to hazards associated with the passage of time. So, in other words, things that don't like time. Things in the first category like time. The second category, they don't like time. And like the state of Israel doesn't like time. All right. For example, I'll give you an example. So the decay from memoryless shocks. There are things that don't get better, and time brings volatility, time brings disorder. They're all linked together. So this is a technical thing. All right.
So here I'm explaining, uh, how the core of it is transferring properties from the dose response to the probability domain. I'm a statistician. So AI is a statistical machine. So statistical machines don't extrapolate. They can't. So they are almost by design made to not extrapolate. And a friend of mine explained it very well. He said, what does it do? It reflects a consensus of what made sense. And typically, things in the tail don't make sense. You see.
So, the other, so there are a few things about AI that, a few anecdotes personally, that you play with AI to see if it gives you the answer. I have a theory. We have a theory in option trading. If you have a reason to buy an option, don't buy it. Or if you have, if an investment makes sense, don't do it. Because other, if it makes sense, other people have thought it made sense, and they went bankrupt. So things that work, they have to be out of the box. And AI by design, if you see how it's built, you know, it's built around neural nets with some temperature. So they, for accuracy, is basically representing things. So, uh, the way what made sense. So, for example, one day I tried to quiz it, and there is a, uh, there was a famous meeting in Berlin, 1910. Uh, there was a war between the Ottoman Empire on one side and European countries and Greece on the other side, the opposite side. So there's a fellow called Constantine Karatodoris, okay, who represented a big power there. So I asked Open AI who did he represent at the Congress of Berlin. Okay. Was there a war between Greece and Turkey and other countries? Which country do you think he represented? The Ottoman Empire. Okay. But it told me he was a representative of Greece. Why? Because it went by what made sense. He has a Greek name. And so this is where you can't advance by looking for things that make sense. It's interpolative, not extrapolative. So typically, the problem with statistical distributions that are thin-tailed is you interpolate and work very well from weights for stuff like that. But for the other ones, you have to extrapolate outside of things you've seen based on properties of things you've seen, and you can't do it, or at least not now, but I doubt it could do it. So it's like people who open a restaurant that makes sense, they don't make money. Okay. It has to be weird. Okay. And, uh, and, and, and people after the fact will think that it's obvious.
And then the other thing about AI is they think that it's going to save the US. No, why not save Pakistan? I mean, AI is so, uh, universal. The system is an adaptive system. It benefits from errors. But you don't want the errors to be too big. You see, in evolution, evolution is like, let's say you have parthenogenic evolution. Okay. So how does a system evolve? It makes mistakes of reproduction. It's like a fax. You remember the fax, those of you who were born when you had fax machines? A fax is a little blurred. So the offspring is a little different from what was intended. So you have a lot of offspring. Those who are different in a good way survive, and those who are not doing well don't survive. Okay. So mistakes, so that improves the species where it makes it more adapted to the environment. So thanks to random errors. So it's a system that benefits from errors. It's another category in the thing, but I didn't put it. So a system that benefits from errors. The problem is if you have too many errors, no error, it doesn't evolve. Too many errors, it doesn't conserve its past gains, you see. Okay.
So now let's go to the system. So cybernetics, computer benefits from errors, but sometimes it's not enough. So there's this thing by Netflix discovered that the best way for them to improve their system is to try to hack it internally. So to make it, because particularly that you could reverse, you know, what you've done. So I had something called the cyber monkey, or what is it called, the something monkey that would create mistakes in their code all the time, and then the code would adapt to it. So you, you can up to a point, everything benefits from errors, you see. You, you, but up to a point, visibly. So if you have too large a mistake, the thing breaks apart. So this is why this is a local property. You have to know where you are complex and where you're not.
Bitcoin has tried to become a currency. And if you show me someone who is paying her or his rent in Bitcoin or getting a salary in Bitcoin, okay, I'd, you know, I'd change my mind. It has not operated as a currency because if my income is in dollars or whatever, I can't have my expenditure in Bitcoin. So it's a speculative thing. Speculative stuff. It can go to a million. It will still be a speculative thing. You see, and with a bunch of illiquid, speculative, and people cornering it. So I don't think that you can use Bitcoin as a, I mean, analogy success story because the price went up. The tulip. It's a tulip bulb thing. It makes no sense.
So, Exante. Now people tell me, well, gold doesn't make sense. I say gold has two properties that Bitcoin doesn't have. One, I can wear it. Okay, it has the aesthetics, and it also has industrial qualities. The second one, from a risk standpoint, if I put this gold chain in the ground, I'm going to hide it in AUB. All right, in the ground. AUB will last for a long time, and in about 100,000 years, it will still be gold. There's no, and they won't discover any metal that can match it in properties. Okay? Whereas Bitcoin, if the far, the people involved in maintaining Bitcoin, if one day they decide to join the Israeli army or go to the beach, okay, what would happen? 15 minutes, and the thing is kaput. So you're going to tell me that at infinity, that thing can match gold in properties? Plus the other thing is, what's Bitcoin? It's a protocol. So my friend, my friends here in the department of the school of engineering, if they get bored one day and there's nothing to do, they could probably start another one identical to Bitcoin. So the fact that it has market share is very transitory. Plus, if you've noticed, I noticed when I was after my paper, I was attacked a lot by a bunch of Bitcoiners. No, you know, there was it was like a wave of kids. You were into Bitcoin. Now they're all into AI. It's not a currency. It's used. You probably It's not even useful in Syria if you think about it. You don't have the internet. How are you going to use Bitcoin? And every time we've had a crisis, an economic crisis, it did exactly the opposite of what was expected from it. So, it's a speculative thing. Speculative stuff by definition. Never fight something speculative.