 Welcome to the AI for good global summit 2018. I'm delighted to be joined by one of the world's most famous and renowned physicists Mr. Roger Penrose sir Roger Penrose of the University of Oxford. Thank you very much for joining us. Thank you. It's my pleasure So here at the AI for good global summit We had to find ways to use AI to advance the UN sustainable development goals. So what Applications do you foresee or envisage for AI as a force for good? I certainly appreciate Emphasis of for good because I think this is very important and that there are clearly things which Artificial intelligence as we understand that word Can be used for and we all hear about self-driving cars and things like that. So clearly that is one example And I imagine in medical diagnosis I've often heard about things that I would say in conjunction with a human being. I think it's very important not to remove the human element in these but you can certainly Call upon data. You can certainly assess Things do statistical analysis and things like this which are clearly objective qualities Which can be advantageous as let's say in medical diagnosis, but in other things also I'm I'm sure it's very universal the fact that that AI can conserve wonderful purposes So I don't think I tend to emphasize the dangers when people talk to me about AI and The danger which often people refer to is they think that AI is going to take over Of course, maybe it will take over some jobs and we have to be careful about that and so reorganize our society in ways that These jobs that they take over are fine and the jobs we don't want to do so that's great But they're not going to take it over us because they don't have any conscious Understanding of what the world is about and that's where we need to come in and it's clear that AI is our servant and not something that we we have to bow down to that and it won't happen because it doesn't have the the Abilities to understand things that we have so what AI for good. Yes, that's great. And I think there are many areas My mental medical diagnosis. I'm sure it's true Architecture already true very much when you know, what kind of structures can you have which will? maintain buildings and temperature control and all sorts of things Probably I'm never quite clear where the boundary is between AI and You see there's an awful lot of computation used in scientific Work and in technology. So you say well, we want to build an aeroplane Say and so we want to construct it in such a way that Airflow behaves in this way and the materials have this kind of strength and so on as you do many many calculations which are very important in in the construction of the aeroplane and But how much of this is what you would actually call AI? I think it's it's the way we use computers in science and technology Is this affected? I mean when it's AI really I suppose it's when you're saying you don't really understand Exactly what's going on and use using things like I suppose What do they call bottom-up systems where you don't you help you train the system and you don't quite know Why it's behaving in a certain way? But your it uses its its experience to develop its skills so you see this for example in these particularly in the go-playing machine which plays itself zillions of times and and it plays the game very well So it beats the the grandmasters the experts, but but it nobody really quite knows what it's doing. So you can see it can develop skills without Having the the theoretical Underpinnings of why it's got those skills. So I suppose that's what you might mean by AI and if you want to design certain things in the aeroplane which you don't really know enough about the dynamics or the the physics of the materials and so on and you and you just try to use this device to to improve by its experience and You can see when it breaks and when it doesn't break and it theoretically does this and just works through different possibilities Without having the understanding of why it works. Now. I suspect I'm thinking aloud when I say these things I suspect that's probably what people mean when they say AI in in in these systems so It's early days isn't it's very early days. We're still exploring the different applications of AI I guess I guess I feel they're always dangerous in these things because when you don't understand there comes up an unusual Circumstance where it behaves strangely and badly because it doesn't have the understanding. It just has the experience But if the experience is Not doesn't and can't accommodate the situation which comes up Then that's something where I would think you would be a bit suspicious of AI systems I'm talking aloud here It's thinking aloud here because I haven't thought about these issues so much before but I let's say that's what what's meant by the use of AI in in Systems like this where you simply use it uses its experience rather than its understanding so I use Skeptical of AI or are you wary of it like Stephen Hawking that you shared the the wolf prize with he launched a very stark warning to mankind saying this Could spell the end of of the human race well You see my worries of the danger are almost the opposite of his you see see he's worried that these machines will take over and That will be the end of us and they will be better than us and then they will run the world instead of us And we don't like that because we're us and not them Now see I don't think he's right on this This my point of view is that these devices don't have intelligence and in the sense of understanding what they're doing We are the ones that have that and as they exist maybe someday We will know what's involved in in these qualities of human understanding, but we we're not they're not it's certainly not in AI as We understand it now. That's my view So that's not the danger. The danger is almost the opposite the danger is the belief that these machines are cleverer than us And therefore we must bow down to them and do what they say. I think that's dangerous. It's dangerous because Partly we may think they will take over jobs and I think to some extent they do take over jobs that they shouldn't and you see this often in systems where you consult on the telephone and you you press buttons and Clearly what you're talking do doesn't have any understanding You see we just say wish I could talk to a human being you see of course sometimes you talk to a human being and the human Doesn't understand either, but at least there's the potential to understand the problem in a serious way So you want something which is more symbiotic if you like that the that the AI systems Can do okay They can call upon a huge body of knowledge and of experience and they can run through examples of situations far more than we could but you have to understand what it is that the limitations of this are and How what you can get out of this and if it is if it works well fine and make use of it and Certainly the the big store we see this with the internet now and the internet how much information there is out there It's absolutely vast and there's no way for a human being to run through all of different things And I feel there's a danger there because it means that fashionable things kind of take over and and things Which are not fashionable have a little chance Even if they may be more significant than the fashionable things so there are dangers there now Could you have an AI system which tends to pick out on other things that should be looked at? And which are not it's an interesting question worth Developing in AI systems. I think again. I worry that if it actually involves an understanding quality There is a danger that it doesn't have that quality But you see the advantage here is that the human being can say well look this quality is something I understand and I can now put that Into the machine and now that makes it a better machine And I don't need to worry about this particular quality of understanding there would be another one comes along and then yes We have to put that in but the human element is still has to be there to have this overall understanding of what's going on What the systems can do and what they can't do and they shouldn't take over because they don't have this quality of understanding But the social issues are important here to what extent does it influence? You know, maybe we don't need taxi drivers. You see and you worry about what happens there or is that the case? I love the guests here at the AI for good summit and it's a paradox believe that humanoid humanoid robots in particular and could add the Missing human touch in some areas like elderly care for instance It's a little worrying though, you see because Where is I mean with old old you might think elderly people where they really don't know the difference between the device But it's worrying because do they really not know you see I think I worry about these things and with children, I think I worry even more because children are very perceptive of You know, they know when these things are actual humans Whether it's the mother who has a real understanding and link between the child or whether it's some device Okay, you have animals which maybe The species is almost distinct and so there is no mother so you give us a piece of cloth instead Okay, maybe that's better than nothing But for a human baby's gosh, I worry about that. I think the human element is so important here and To try and put an imitation of a human being and Pretend that this is going to supplant the human also. There's the danger that later on the child realizes it's been fooled and I think that's that's worse because it may have even Thought for a while that this was its mother or something and then later on finds out that this is this is It's it's been fooled into thinking this I think that can be can be awfully bad So these are all the things we have to take into either you take them into consideration Absolutely, okay in a last resort if there's nothing else you can do sure and yes if somebody is very ill and Better than nothing sure, but it's not it's not a substitute for the human and human understanding and the understanding can Take very subtle forms even, you know face a child can see something in a mother's face Usually I hate to say it. It's the mother and not the father That that that can relate in this kind of way to a child okay later on and I'm sure the father brings important qualities But that's very early stage. I feel that it's really very important There is something there which which I'm afraid we can't do okay, so am I for good? Yes Yes, but we have to be yes. I think it's to be careful. It's it's it's good Yes, there's a lot of places where it can be extremely valuable But always be careful and stand back and and the word understanding I regard as the key But that it it's not going to have the actual understanding may it may have some qualities of it that you've put in It may have some experiences, which you can't have because it requires too many instances of this But understand what this is and what this isn't it's very important. Okay, so Roger Penrose. Thank you very much my pleasure