Richard Feynman: Can Machines Think?

This is an excerpt of a question and answer on the topic of artificial intelligence from a lecture by Richard Fineman on September 26, 1985. I actually record the audience's questions because they are barely audible in the original question. Do you

think

there will ever be a machine that

think

s like that? human beings and being more intelligent than human beings, first of all, they think like human beings, I would say no and in a minute I will explain why I say no and second, being more intelligent in human beings is a question of intelligence to be defined. If I You will ask if they are better chess players than any human being can be.

Yes, I'll tell you one day. They are better chess players than most humans at this point. One of the things we always do. is that we want the dawn machine to be better than anyone, not just better than us, if we find a machine, it can play chess better than us, we are not very impressed, we keep saying it and what happens when it faces the Masters. Let's imagine that human beings are equivalent to the Masters and everything well the machine has to be better than a person in everything that the best person does at the best level is fine but it is hard for the machine but with respect to the question of whether to do it is thinking like a machine my opinion is based on the following idea that we try to make these things work as efficiently as possible with the materials we have, materials are different than nerves, etc., if we wanted to make something that runs quickly on the ground then we could see a cheetah running.

We could try to make a machine that runs like a cheetah, but it's easier to make a wheeled machine with fast wheels or something that flies just above the ground in the air when we do. a bird, planes don't fly like a bird, they fly but they don't fly like a bird, okay, so they don't exactly flap their wings, they have another type of contraption in front of them that spins or the most modern plane has a tube in which you heat the air and expel it. at the back a jet engine a jet engine like internal rotating fans and so on and it uses gasoline, it's different, so there's no doubt that later

machines

won't think like people think in that sense with respect to intelligence I think which is exactly the same, for example they are not going to do arithmetic in the same way that we do arithmetic but they will do it better let's take an ematic bath very elementary mathematics arithmetic they do arithmetic better than anyone else, much faster and different, but it is fundamentally the same because at In the end the numbers are equivalent, so that's a good example that we're never going to change the way they do arithmetic to make them more like humans, which would be going backwards because human-made arithmetic is slow, cumbersome and confusing in the end. a place full of bugs where these guys are fast, if one compares what computers can do with human beings, we find the following comparisons quite interesting, the first of all if I give you a human being. a problem like this I am going to ask you to return these numbers to me every other in reverse order please right now I am fine a series of numbers and I want them you didn't return them to me in reverse order every two I will tell you I will make it easy for you just give me back the numbers just as I gave them to you are you ready one seven three nine two six five eight three one seven two six three any one will have to be I can do that now and that's more than no more than twenty or thirty numbers, but you can give it to a computer 50,000 numbers like that in s confirm any reverse order the sum of all of them do different things with them and so on and don't forget them for too long instead, there are some things that a computer does much better than a human and you will better remember that if you compare

machines

with humans, but what a human has to do for themselves always, they always do this, they always try to find one thing, but they can do it better than the computer, so now we know many things that humans can do better than a computer.

She is walking down a street. She has a certain type of movement. and you know Jane, right, or the sky is sinking and you see her hair move a little bit, it's hard to see from a distance, but that particular funny way she did it behind her head looks, that's good for recognizing things and recognize patterns. It seems to be something we haven't been able to put into a defined procedure. You would say I have a good procedure for recognizing Jack. Just take lots of pictures of Jack because of the way you can put a picture on the computer.

In fact, through this. method here, if this was much finer, I could tell if it's black and white at different points. You know, in fact, you get photographs in a newspaper with black and white dots, and if you do it with Stuart well enough, you can't see the dots. With enough information I can upload images for you to put all the images of Jack in the different circumstances and that is the machine to compare them. The problem is that the new actual circumstances are different, the lighting is different, the distance is different, the tilt of the head is different and you have to figure out how to allow all that and it is so complicated and elaborate that even with the big machines with the amount of storage available and the speed at which they go, we can't figure out how to make a defined procedure that works at all or at least works anywhere within a reasonable speed, so recognizing things is difficult for machines today and some of those things are done by a person in an instant, so there are things that humans can do that we don't know how to do in a file system, so it's recognition and that brings me back to something that I left, which is what kind of filer cannot be imitated by the machine, a filer that has some special skill that represents that requires recognition of a complicated type, for example, a firefighter Kirk in the fingerprint department looking at the prints fingerprints and then makes a careful comparison to see if these fingerprints match.

It has not been. He's almost ready to be. It's hard to do. It is almost possible to do it with a computer. He said there is nothing. I look at the two fingerprints and see if all the blood spots are the same but of course that is not the case the finger was dirty the impression was made at a different angle the pressure was different the ridges are not in exactly the same place if you were trying to match the exact same image it would be easy but where is the center of the print which direction do you turn your finger where it has been crushed a little more a little less where there is some dirt on the finger if in the meantime has a wart on his thumb, etc., complications, these small complications make the comparison much more difficult for the machine, for the blind filing system, which is too, too much, too. slow to be certainly completely impractical almost at this point I don't know where they are, but they are going fast trying to do it where can a human get through all that somehow just like they do in the game of chess? be able to grasp patterns quickly and we don't know how to do it quickly and automatically audience question: Can computers discover new ideas and relationships on their own?

Well, it depends on what they mean by themselves and it's hard to discover new relationships, our computers can Have there been computers that do things like problems that are getting better at geometry or something where they've turned the problem of finding a proof of a theorem in a defined procedure? Well, once you do it, although it's an elaborate and silly way of doing tests, they can do it right now a computer can't do all the different things that a person can do, you know, it's very difficult to find any way to define with quite precisely something that we can do and that we can say that a computer will never be able to do there are some things that people invent that say that while you do it you will feel good or while you do it you will understand what you are doing or some other abstraction.

I'd rather feel like these are things like while doing this, we'll be able to get the lice out of your hair. No, he doesn't have hair. Lice can be scratched. Okay, so you have to be careful when you say what the human does if you add. to the actual result of his effort some other things that you like appreciation of aesthetics but you didn't do that. I'm not saying you did it, but a lot of people do that when they ask questions and if we add things that we think we're doing on top of what we actually do and we just look at not just the result of what we're doing, but a lot of additional things, then It becomes harder and harder for the computer to do it because humans have a tendency. to try to make sure that they can do something that no machine can do somehow it no longer bothers them, it must have bothered them in earlier times that machines are physically stronger than them, they can lift heavier weights than people who can.

He moves things faster than people who can run fast, if you can fly you can do it, his strengths are terrible, etc., and we don't sit around worrying that there is some way the guy can turn his hand and that some machine cannot do it. We can easily make machines that are better than us at predicting the weather, for example, because what you do to predict the weather is look at old records and see when circumstances were similar and assume that the results will be similar, added to that. certain amount of analysis of the movement of the wind according to the laws of physics and its current amount of tricks together.

Well, now the speed will be higher and the prediction effectiveness is higher if you could analyze more cases to get a better chance to get closer and make a longer and more elaborate calculation that includes more variables, which is too difficult for us in time to make the prediction now we have to make the weather prediction, let's say the weather for three days from now we have to predict it in three days or that damn thing is useless, right, and we worked at a certain speed, but the computers worked faster fast and they can do more and therefore, for example, for weather forecasting, in the end maybe not today, but someday it will not be at all.

It is inconceivable that the machine can make faster, more effective and more accurate weather predictions than we can. However, we have given you the procedure. Now the question is what happens if we don't do the procedure right. People have tried it. This game of giving instead of a direct procedure a kind of what has been called heuristics, tests an analogy to get a new idea of ​​how to do something, compares this with that, tests an extreme case, etc. and a man named Lynette has I got further with this Do I have time? How much time do I have because we want to see these slides?

What I know is interesting information but how much time do I have? It's time for the slideshow now since I don't have any more time. Don't say more about the subject, no, but it takes a few minutes, that's why I asked you for your time, okay, you make this machine, which again was a filing cabinet. You understand what it does. It seems like you're trying to find the answer to something by looking at the different possible possibilities, but which ones are you trying? they're kind of like patterns on the chest instead of everything it says, try moves near the center of the board first, you know, and don't mind the ones in the corner or something or some kind of principles and applied it first maybe a kind of naval game it's a game that people play in California that is all organized according to rules it's fun someone makes all the rules battleships cost this much armor cost this amount weapons cost this much and so on and so on and you have so much budget for your Navy and you are going to make several types of ships with different types of weaponry and then this type of ship, the weaponry of a certain thickness that costs a certain amount, can only withstand projectiles of a certain strength that you know, etc., so You tried with the money to fix buy different to design different types of ships to make your Navy better than the next and when they got together there are ways to calculate what there are no real navies, it's a game. which is the best and all the rules are set but a large volume is fine at the cost of everything in the power of everything, armor piercing possibilities etc. and it is a good game and mr lumic tried his program in this game. and he put in his program the extreme case of his wrist and things like that and won the championship in California, of course, he tried a lot in different cases, but he didn't try all the cases, not like the game of chess, there were also those. many things, but it was guided by its own things, now within that was this: if you got a better bite from the navy, your own calculation and used one of the heuristics, mark that eristic at a higher level as more valuable, now use the most valuable heuristics.

First see that the Machine's ability depended on its learning, so to speak, which of its tricks work most effectively most of the time and then are used the most, so it isexactly what you would like to look smart, well, you won. And how did she win? It turned out that he won that year by making a big battleship with all the armor, which was very stupid, but when you calculate it, it's better than any of the normal things that no one thought of, but his machine thought that next year he would enter again and this time he won by cheating all his money to win a hundred thousand because they changed the rules so the big battleship wouldn't win, you know, but it will change a hundred thousand small pots. narrow carrying each of the weapons that were very likely to be knocked out is fine, but there were 100,000 of them it didn't cost much each and they couldn't knock them all out, so some disgusting mosquitoes came and it turned out that when you calculated again he won the third year and he was not allowed to play on it and he is proud of this machine and this tourist business for a number of other problems, as he tests it a lot and tries new heuristics, etc., and it has become very interesting.

He complained that there were a number of errors in it and when he gave a talk about it I said that I thought I would say my comment later, one of the errors, for example, was that the Machine got a heuristic and invented his doll, except his doll in Excel, so the damn machine, what? He, since he had it, it was because it's hard to get time on the computer. He needed a lot of time on the computer. He had 50 machines a night from a package company or something. He always worked at night and arrived in the morning.

Damn things, he would try all my stuff and get the results. I used it to do math and various other things. One day comes and a heuristic is made. You develop the heuristic when he poses the problem or a new idea. She writes. He says if it's his or the machine and I was that student or a machine, okay, and this was every year, or every question he didn't win, don't pay attention to it, that saves a lot of time, I could do a lot better. He didn't have any problems with this boy toy, so he didn't pay any more attention to it than that bad night, okay, sorry, he'd fix that book the next time he had a mistake.

It was he who entered and looked and found. That heuristic number 693 had scored 999 out of a thousand is a very useful heuristic all night long. This thing kept using the 9 693 heuristic more and more and it was a novelty where the wonderful edge could solve all the problems. It was fantastic. Well, when you figured out what the heuristic was, the next thing you see to make this work is you change the numbers in the heuristics every time something works, just to assign credit, so to speak, to the heuristics that were used. Okay, so this heuristic was by assigning credit always the credit sign to the 693 heuristic and that's how that thing came about.

I say they both show intelligence. If you want to make a smart machine, you'll get all kinds of crazy ways to avoid work. I say don't pay attention to the problem of sneakily developing some kind of psychological distortion, we always do the same thing and don't worry about anything else, etc., so I think we are getting closer to intelligent machines, but you are showing the necessary weaknesses of an intelligent person.