Ray Dalio: Artificial Intelligence Principles | AI Podcast Clips
00:00:00.000 | Let me talk about, if we could, about AI a little bit.
00:00:05.560 | So we've, Bridgewater Associates, manage about $160 billion in assets.
00:00:15.340 | And our artificial intelligence systems algorithms are pretty good with data.
00:00:21.120 | What role in the future do you see AI play in analysis and decision making in this kind
00:00:28.120 | of data rich and impactful area of investment?
00:00:34.680 | I'm going to answer that not only in investment, but I give a more all-encompassing rule for
00:00:41.960 | AI.
00:00:44.100 | As I think you know, for the last 25 years, we have taken our thinking and put them in
00:00:51.720 | algorithms.
00:00:53.040 | And so we make decisions, the computer takes those criteria, algorithms, and they put them,
00:01:00.920 | they're in there and it takes data and they operate as an independent decision maker in
00:01:06.800 | parallel with our decision making.
00:01:09.000 | So for me, it's like there's a chess game playing and I'm a person with my chess game
00:01:15.720 | and I'm saying it made that move and I'm making the move and how do I compare those two moves?
00:01:20.080 | So we've done a lot.
00:01:21.240 | But let me give you a rule.
00:01:23.440 | If the future can be different from the past and you don't have deep understanding, you
00:01:34.280 | should not rely on AI.
00:01:39.520 | Those two things.
00:01:40.680 | Deep understanding of?
00:01:42.480 | The cause effect relationships that are leading you to place that bet in anything.
00:01:48.840 | Anything important.
00:01:50.360 | Let's say if it was do surgeries and you would say, how do I do surgeries?
00:01:54.720 | I think it's totally fine to watch all the doctors do the surgeries.
00:01:58.920 | You can put it on, take a digital camera and do that, convert that into AI algorithms that
00:02:09.080 | go to robots and have them do surgeries and I'd be comfortable with that.
00:02:14.560 | Because if it keeps doing the same thing over and over again and you have enough of that,
00:02:19.960 | that would be fine even though you may not understand the algorithms because if the thing's
00:02:25.320 | happening over and over again and you're not asking, the future would be the same.
00:02:29.400 | That appendicitis or whatever it is will be handled the same way the surgery, that's fine.
00:02:35.440 | However, what happens with AI is for the most part, is it takes a lot of data with a high
00:02:45.160 | enough sample size and then it puts together its own algorithms.
00:02:51.000 | There are two ways you can come up with algorithms.
00:02:53.620 | You can either take your thinking and express them in algorithms or you can say, put the
00:03:00.000 | data in and say, what is the algorithm?
00:03:03.840 | That's machine learning.
00:03:05.480 | When you have machine learning, it'll give you equations which quite often are not understandable.
00:03:12.480 | If you would try to say, okay, now describe what it's telling you, it's very difficult
00:03:16.720 | to describe and so they can escape understanding.
00:03:20.760 | It's very good for doing those things that could be done over and over again if you're
00:03:25.440 | watching and you're not taking that.
00:03:27.360 | But if the future is different from the past and you have that, then if the future is different
00:03:33.560 | from the past and you don't have deep understanding, you're going to get in trouble.
00:03:38.960 | That's the main thing.
00:03:40.560 | As far as AI is concerned, AI and let's say computer replications of thinking in various
00:03:47.000 | ways, I think it's particularly good for processing.
00:03:51.340 | But the notion of what you want to do is better most of the time determined by the human mind.
00:04:00.680 | What are the principles?
00:04:01.680 | Like, okay, how should I raise my children?
00:04:05.040 | It's going to be a long time before AI, you're going to say it has a good enough judgment
00:04:09.840 | to do that.
00:04:10.840 | Who should I marry?
00:04:12.120 | All of those things.
00:04:13.160 | Maybe you can get the computer to help you, but if you just took data and do machine learning,
00:04:17.480 | it's not going to find it.
00:04:18.760 | If you were to then take what are my criteria for any of those questions and then say, put
00:04:25.800 | them into an algorithm and you'd be a lot better off than if you took AI to do it.
00:04:30.600 | But by and large, the mind should be used for inventing and those creative things.
00:04:38.040 | And then the computer should be used for processing because it could process a lot more information,
00:04:44.200 | a lot faster, a lot more accurately, and a lot less emotionally.
00:04:49.640 | So any notion of thinking in the form of processing type thinking should be done by a computer.
00:04:57.080 | And anything that is in the notion of doing that other type of thinking should be operating
00:05:02.560 | with the brain, operating in a way where you can say, ah, that makes sense.
00:05:09.720 | You know, the process of reducing your understanding down to principles is kind of like the process,
00:05:17.040 | the first one you mentioned, type of AI algorithm where you're encoding your expertise.
00:05:23.040 | You're trying to program, write a program.
00:05:25.320 | The human is trying to write a program.
00:05:27.880 | How do you think that's attainable?
00:05:30.960 | The process of reducing principles to a computer program.
00:05:38.280 | Or when you say, when you write about, when you think about principles, is there still
00:05:43.760 | a human element that's not reducible to an algorithm?
00:05:49.600 | My experience has been that almost all things, including those things that I thought were
00:05:56.960 | pretty much impossible to express, I've been able to express in algorithms.
00:06:05.080 | But that doesn't constitute all things.
00:06:08.560 | So you can, whew, you can express far more than you can imagine you'll be able to express.
00:06:16.800 | So I use the example of, okay, it's not, how do you raise your children?
00:06:22.400 | You will be able to take it one piece by piece.
00:06:25.520 | Okay, at what age, what school?
00:06:29.080 | And the way to do that, in my experience, is to take that and when you're in the moment
00:06:37.260 | of making a decision or just past making a decision, to take the time and to write down
00:06:45.080 | your criteria for making that decision in words.
00:06:49.960 | That way you'll get your principles down on paper.
00:06:53.600 | I created an app, online call, it's right now just on the iPhone, it'll be on Android.
00:07:00.320 | Yeah, I tried getting it on Android.
00:07:01.920 | Come on, now.
00:07:02.920 | It'll be- Let's get it on Android.
00:07:03.920 | It'll be, in a few months it'll be on Android.
00:07:06.120 | Awesome.
00:07:07.120 | But it has an app in there that helps people write down their own principles.
00:07:12.560 | Because this is very powerful.
00:07:14.560 | So when you're in that moment where you've just, you're thinking about it and you're
00:07:19.000 | thinking your criteria for choosing the school for your child or whatever that might be,
00:07:25.240 | and you write down your criteria or whatever they are, those principles, you write down
00:07:30.280 | and that will, at that moment, make you articulate your principles in a very valuable way.
00:07:39.560 | And if you have the way that we operate, that you have easy access, so then the next time
00:07:44.800 | that comes along, you can go to that or you can show those principles to others to see
00:07:49.480 | if they're the right principles.
00:07:51.360 | You will get a clarity of that principle that's really invaluable in words and that'll help
00:07:56.880 | you a lot.
00:07:58.520 | But then you start to think, "How do I express that in data?"
00:08:03.460 | And it'll shock you about how you can do that.
00:08:07.000 | You'll form an equation that will show the relationship between these particular parts
00:08:12.280 | and then the, essentially the variables that are going to go into that particular equation
00:08:18.880 | and you will be able to do that.
00:08:20.520 | And you take that little piece and you put it into the computer.
00:08:25.460 | And then take the next little piece and you put that into the computer.
00:08:29.440 | And before you know it, you will have a decision-making system that's of the sort that I'm describing.
00:08:35.520 | - So you're almost making an argument against an earlier statement you've made.
00:08:40.560 | It convinced me, at first you said, "There's no way a computer could raise a child," essentially.
00:08:47.440 | But now you've described making me think of it.
00:08:49.480 | If you have that kind of idea, meritocracy, you have this rigorous approach to bridge
00:08:55.080 | water takes and investment and apply it to raising a child.
00:08:59.320 | It feels like through the process you just described, we could, as a society, arrive
00:09:05.080 | at a set of principles for raising a child and encode it into a computer.
00:09:11.520 | - That originality will not come from machine learning.
00:09:16.640 | - The first time you do it, so that the original, yes.
00:09:19.640 | - That's what I'm referring to.
00:09:20.680 | - But eventually, as we together develop it and then we can automate it.
00:09:24.800 | - That's why I'm saying the processing can be done by the computer.
00:09:29.200 | So we're saying the same thing.
00:09:30.640 | We're not inconsistent.
00:09:32.320 | We're saying the same thing, that the processing of that information and those algorithms can
00:09:37.200 | be done by the computer in a very, very effective way.
00:09:40.960 | You don't need to sit there and process and try to weigh all those things in your equation
00:09:44.900 | and all those things.
00:09:46.440 | But that notion of, "Okay, how do I get at that principle?"
00:09:50.680 | - And you're saying you'd be surprised how much you can express.
00:09:56.400 | - That's right.
00:09:57.400 | You can do that.
00:09:59.520 | So this is where I think you're going to see the future.
00:10:04.680 | Right now, we go to our devices and we get information to a large extent.
00:10:13.040 | And then we get some guidance.
00:10:14.840 | We have our GPS and the like.
00:10:17.000 | In my opinion, principles, principles, principles, principles, I want to emphasize that.
00:10:22.280 | You write them down.
00:10:23.920 | You've got those principles.
00:10:25.720 | They will be converted into algorithms for decision-making.
00:10:30.400 | And they're going to also have the benefit of collective decision-making.
00:10:34.960 | Because right now, individuals, based on what's stuck in their heads, are making their decisions
00:10:40.560 | in very ignorant ways.
00:10:42.560 | They're not the best decision-makers.
00:10:44.320 | They're not the best criteria.
00:10:46.280 | And they're operating.
00:10:47.480 | When those principles are written down and converted into algorithms, it's almost like
00:10:52.860 | you'll look at that and follow the instructions.
00:10:55.680 | And it'll give you better results.
00:10:58.480 | Medicine will be much more like this.
00:11:01.140 | You can go to your local doctor and you could ask his point of view and whatever.
00:11:05.160 | And he's rushed.
00:11:06.360 | And he may not be the best doctor around.
00:11:08.960 | And you're going to go to this thing and get that same information or just automatically
00:11:13.720 | have an input in that.
00:11:15.320 | And it's going to tell you, okay, here's what you should go do.
00:11:18.840 | And it's going to be much better than your local doctor.
00:11:21.760 | And that, the converting of information into intelligence, okay, intelligence is the thing.
00:11:30.080 | We're coming out with, again, I'm 70 and I want to pass all these things along.
00:11:35.320 | So all these tools that I've found need to develop all over these periods of time, all
00:11:42.520 | those things, I want to make them available.
00:11:44.880 | And what's going to happen as they're going to see this, they're going to see these tools
00:11:50.360 | operating much more that way.
00:11:52.400 | The idea of converting data into intelligence, intelligence, for example, on what they are
00:12:00.000 | like, on what are your strengths and weaknesses, intelligence on who do I work well with under
00:12:05.760 | what circumstances.
00:12:06.760 | Personalized.
00:12:07.760 | Intelligent.
00:12:08.760 | We're going to go from what are called systems of record, which are a lot of, okay, information
00:12:14.920 | organized in the right way to intelligence.
00:12:18.960 | And we're going to, that'll be the next big move in my opinion.
00:12:24.000 | And so you will get intelligence back.
00:12:27.320 | And that intelligence comes from reducing things down to principles and to...
00:12:31.680 | That's how it happens.
00:12:32.520 | [silence]
00:12:42.520 | [silence]