Intelligence explained with math – MIT physicist

SS: Okay. Math is a very precise science. And so if everything in our universe is mathematics, does that mean that we can theoretically precisely calculate anything?

MT: Well, the famous physicist Niels Bohr once made this joke that in theory, theory and practice are the same; and in practice, they're not. My guess is that in theory, everything can be calculated. But in practice, we certainly cannot. There are many cases, even today, when we actually have already, we think, more or less, the right equations to study something. For example, there is something called quantum chromodynamics, which we believe really describes everything you need to know to compute all of chemistry and the whole periodic table that Mendeleev discovered and so on, just from pure math, but we can't do it yet, because the calculations are too hard. And there are huge supercomputer simulations being run right now to do it more accurately. So even if we were to discover the correct fundamental mathematical description, it's not like all physicists will be unemployed.

SS: You know, I talked to a lot of renowned theoretical physicists, and many of them told me that there's a great deal of chance in how nature works, like Paul Davies, I'm sure you know him. 

MT: Yeah. 

SS: How can that be reconciled with your mathematical, more determined view of the universe?

MT: Yeah, that's a really good question. We realised, of course, over 100 years ago that things appear to be random sometimes in physics because of what we call quantum mechanics. And 100 years later, top scientists around the world are still arguing bitterly about what this means. Some people think this means that something actually random happens from time to time when somebody looks at something and observes it. And others think that's just nonsense and that's nothing random is happening. I actually belong to the second camp but I'm not alone. I mean, even Einstein, famously said that he doesn't believe that God plays dice. If you look more carefully at this, the basic theory that is sort of master key of quantum mechanics, it’s called the Schrodinger equation, it's a mathematical equation, which describes how reality today predicts reality in the future. And the funny thing is it talks about a bigger reality, actually, than the one we can observe where maybe there is there's one branch, you know, where you're pregnant now, and another one where you already had a baby two years ago, and they can all be equally real. And if you take the point of view that the actual reality is bigger, and there are there all these different versions of life playing out in sort of in parallel, then there's actually nothing random happening at all. But if you want to get rid of all that extra reality, because you feel, ‘Oh, that's too much, that's too weird,’ then you can try to introduce this extra postulate that once in a while, something actually random happens that may pick out only one of the versions. What I love about this is it's not just philosophy, because right now we're investing around the world massive amounts of money to try to build quantum computers and they will ultimately only work at a large scale if these random things truly don't happen.

SS: Okay. I was going to say that in school, I was taught that two parallel lines never cross. And then I found out that when it suited mathematicians, they were allowed to cross. So if math is the language of the universe, how come its rules change when you guys see fit?

MT: Haha, that's a very interesting question. So mathematics is really the study of different kinds of abstract structures that you can describe with formulas very symbolically, and there are many, many different ones. And we thought we knew earlier, the only kind of space that we thought could exist, called Euclidean space, where two parallel lines never cross. It turns out by now mathematicians have discovered that there are all sorts of other different spaces, in some of them, all parallel lines cross twice, for instance. And that becomes an experimental question to figure out of all of these different structures that exist mathematically, which is the one that we live in. For example, it turns out that the kind of space we live in, is not that old kind that Euclid thought but a much more interesting kind.

SS: What does seeing the fabric of the universe as math give us?

MT: For me, it gives me optimism that we will actually be able to understand our universe, even better than we once thought. You know, in ancient times, we were almost resigned that we would never understand most of the world. And then we discovered that some parts of it were mathematical, you could predict how the things would move in a certain shape, you know, why it was x squared, a parabola, ellipses and then we got math that described light and magnetism and electricity and subatomic particles. If my guess is correct that it's all mathematical, then that means that we have the potential as humans to ultimately be able to understand, at least in principle, everything, which I think is inspiring.

SS: All right. Let's talk about more earthly matters because the latest book you wrote is actually concerned with more earthly matters – the future of artificial intelligence on Earth. You say the problem with AI is that it doesn't really comprehend what humans are and how our world is functioning. How do you explain to a robot what this world is about?

MT: Yeah, first of all, there’s a fascinating connection between what we just talked about with math, because gradually, more and more of the world around us has become explainable by science with mathematics. And sort of the last bastion is trying to understand what is intelligence and consciousness. And a core idea of artificial intelligence is that even intelligence itself can be described with mathematics if you think of it as information processing. And when we watch movies about AI, I think, they typically make us worry about the wrong thing. They make us worry about robots turning evil. But the real threat of advanced artificial intelligence is not that it turns evil, but just that it turns very competent that it has goals that are not aligned with our goals. For example, do you hate ants? Like if you're walking on the street, and you see an ant, you're going to go and, like, step on it on purpose?

SS: No.

MT: I didn't think so. You look like a nice person. But suppose you're in charge of building a new green energy hydroelectric plant, and just before you're going to flood this valley with water, you see that there's an anthill in the middle. What are you going to do?

SS: I’ll have to, I guess.

MT: Exactly. So it's not that you hated ants, but you had different goals that were not aligned with those of the ants, so too bad for the ants, right? So we want to make sure, if we ever create machines that are much smarter than us and more powerful than us, that their goals are aligned with our goals so that we don't become some– Because they're going to get their way, because they're just so competent, right?

SS: But it's really up to us to actually tell them what their goal is, right?

MT: That's right except that we're not doing such a good job with that at all. 

SS: Ok, how do we do that? Because if we’re to tell them what their goal is we need to explain– or not? Do we need to explain to them how the world works in order to explain to them what the goals are? Or these are two separate things, we don't need to maybe explain anything about how our world functions, just tell them their goals?

MT: We certainly need to explain a bit about– You know, you're going have a child, right? You're not going to be able to teach him to be a good person if you don't teach him anything about how the world works. But eventually, he'll grow up and become stronger than you probably, right?

SS: The child will be born with consciousness something that a robot will never have. I mean, consciousness is what really makes you human and you intuitionally understand how to function in the world and how the world functions.

MT: That's a very controversial question that people love to argue about, you know, whether consciousness is only possible in biological organisms, like humans, or whether it's something to do with information processing that could also happen in machines. But in either case, you know, there are also humans who are conscious, who can still do very bad things, right? So that's why I'm confident that your son is going to grow up and do good things because you as a mother are going to make sure that you don't just teach him about the world, but also instil in him good goals. And I think we're not doing a good job with that today, with our computers. For example, there was a German pilot who told his autopilot on his plane to just fly into the Alps. And it just said okay and it did, and this Germanwings aircraft crashed and over 100 people died. Nobody had bothered teaching this autopilot to never fly into fixed objects, right?

SS: Max, I want to talk about your concept of Life 3.0, okay? So let's say if life 1.0 is microorganisms, which are concerned with self-replication, and life 2.0 is basically humans, who can influence their environment, the programming, then life 3.0 is artificial intelligence-driven, able to overcome biological constraints. But can a clever robot really be seen as a life having no DNA, no cells and, like we mentioned, probably no consciousness?

MT: Well, we have to be humble and acknowledge we don't know exactly what consciousness is, what makes us experience things subjectively, but –

SS: Ok, let’s roll with DNA because we know for sure they're not going to have DNA. Let's roll with cells because we know we're not going to have cells, okay?

MT: But my guess is that both intelligence and consciousness actually are all about information processing. And I consider it carbon chauvinism, this idea that you can only be smart and conscious if you're made of carbon atoms. My suspicion is that what really matters is if the information is processed in certain, very complex ways that we don't particularly fully understand yet, and that it doesn't matter whether you're made of silicon or carbon or have cells or no. To me, the basic essence of life is that you have a process that can maintain its complexity and make copies of itself. And the basic essence of intelligence is simply that you can accomplish goals, the better you are at accomplishing goals, the more intelligent you are. And so I think that if we can make machines that are much more intelligent by that definition than us, it's either going to be the best thing ever happened to humanity or the worst. And I'm very, very excited about thinking what can we do now to steer things in a good direction.

SS: You know, in human society sometimes it's hard to push through even something as benevolent as free health care or equal rights, right? What makes you think that it will be easy for any society to make an even more drastic step towards a nanny state, and turn all decision making over to robot nannies?

MT: It's obviously hard, which is why we have so many problems on this planet. But we also have seen a lot of examples of how we can make progress. You know, if we were having this conversation 30,000 years ago, we would probably both be dead, because the life expectancy was very short back then. And now, because of technology, we figured out how to live longer, healthier and wealthier lives. And we've also gotten together as a society, particularly after World War II, and tried to use a lot of technology to actually help everybody get better off. In many countries, there's free healthcare, free university education, and so on. So the potential is there. If we can use artificial intelligence to ultimately figure out how to cure all diseases, and produce and take care of all, everything that everybody needs to be able to live happy and inspiring lives, that'd be wonderful. But if we instead use artificial intelligence like we very much do today, I think, in the media system around the world, to manipulate people into hating each other and hating each other not just between different countries but also within different countries, and to all these evermore fragmented groups, that, of course, is not a good thing. In fact, I'm working very much right now on this research project, where we're using artificial intelligence exactly to see how we can take some of the methods for figuring out the truth that we use in science, and better export them into the media ecosystem so that we as a species can just get along better by having a shared understanding of what's actually happening.  

SS: Okay, so let's talk about this, let's mention the Improve The News project that you have going, where with the help of AI, I understand, the newsfeed can be tweaked according to political bias. So it should help with the info bubble problem we have now, exposing people to other points of views. Why does it take an AI to do a simple thing like that? 

MT: Good question. So if you go to improvethenews.org, that little nonprofit website we set up, which has news on it right now, as you mentioned, if you need humans to read all of these many, many thousands of articles every day, and then classify them in different ways, you know, you would need a lot of money for it. So I couldn't do it, you know, as just an academic professor. If you have artificial intelligence, classifying all the articles automatically, then it becomes basically free so you can have it for free with no ads. Today, artificial intelligence is used already very heavily in media by all the social media companies, for example, but mainly not to classify the news, but to classify you, the reader, because their goal is to simply make you spend as much time as possible, watching advertising so they can maximise their profit. So they will look at you and everything you clicked on before and make a very detailed model of you and see how can we get you to click, click, click here. And these algorithms have discovered, of course, that one of the best ways to hook people is to make them angry, to trigger their emotions and to tell them things that feed their prejudices. And even though the original goal of this was just to make money, I think the result has been quite catastrophic, where we now have parallel universes appearing here on Earth, basically, in the media landscape, where you have all these different people, even different people I know here, who are living effectively in parallel universes and don't read the same news, they have completely different understandings of what's even happening, they can’t have a meaningful conversation. And I think if we're going to have a healthy democracy, it's very important that we are all living in the same universe and, you know, it's hard to figure out what the truth is, really hard. If it weren't so hard, we wouldn't need science, right? We could just appoint the fact-checking committee, run by some companies or governments telling us what the truth is. But we scientists, we fought for hundreds of years against that stuff. Like, imagine if today Galileo put out a tweet saying, ‘oh, Earth is actually orbiting the Sun’, and then Pope Urban VIII’s fact-checking committee said, ‘no, this is misinformation, it's not true.’ We fought against that sort of stuff so hard as scientists and I think, because of that, we really have something as a scientific community we can add to the society around the world that can help people find their way and figure out what's actually going on.

SS: Okay, but on a more practical note, can artificial intelligence be trusted to bring me a balanced info diet? Or is it too weak now to truly get the meaning of balance? I mean, you know, Fox calls themselves balanced. How would artificial intelligence know they're not?

MT: We should certainly not trust any algorithms just because someone tells us to trust them. One of my favourite quotes by Einstein is that the blind faith in authority is the greatest enemy of truth. And to me, the whole essence of being a scientist is that I would rather have questions I can't answer than the answers I can't question. So no, we should never trust anything, we should have a healthy scepticism towards everything. So the point of the improvethenews.org project is not to tell people, ‘Oh, don't trust those guys, trust this algorithm, this is the truth.’ No-no, this is rather simply a tool to try to empower ordinary people to see all the evidence more clearly so they can make up their own mind. If you're reading about some typical controversial issue, it's very nice if you can now just by moving a little slider, see, okay, what are people saying out on this side of the conflict and on that side of the conflict, and then make up your own mind. It’s kind of like, you know, we all have friends who got divorced or ended a relationship and we know that if you were to talk to only one of the two people about what went wrong, you tend to get a sort of biased version of the story. But if you can talk to both people about why they broke up, you can usually piece together a much more nuanced understanding of what actually happened. And it's the same, I believe, with very many geopolitical and national controversies as well, that if you give people the opportunity to hear both sides or all sides, they will themselves come up with a more trustworthy interpretation of what's happening. That's the idea. 

SS: Oh, Max, it's been such a pleasure talking to you. Thanks a lot for this wonderful insight. All the best of luck with your research. And I mean, maybe we should do this in one or two years’ time to see where we are at with the AI at that point.

MT: I would love that. Yes.

SS: Ok. Thanks a lot, Max. Have a great day and good luck with all your future endeavours. 

MT: Thank you so much.