Our lives in the age of Artificial Intelligence

Transcript

(Music plays)

Chris Neff: Welcome to Open for Discussion. The University of Sydney podcast looking at research through a personal and critical lens. I'm your host, Chris Neff.

Joining me today is Dr Michael Harre from the University's Faculty of Engineering and Information Technologies. Michael is an artificial intelligence aficionado, try saying that 3 times fast, and lecturer in complex systems. Thanks for joining us today Mike, on Open for Discussion.

Dr Michael Harre: Great, thanks for having me here.

Chris Neff: Can I ask first, what is artificial intelligence and secondly, what got you into artificial intelligence? What was your background, how did you come to this?

Dr Michael Harre: Well so, artificial intelligence, that's a,...that's a..(laughs) big definition. Ahh.. There's no straight forward answer in the sort of...general definition but usually it's something that runs on a computer that;s trying to do something kinda human like but perhaps not being very..very good at it.

Chris Neff: Ok good. Because my understanding is mostly a star trek, star wars aahhhh... Will Smith movies and other kind of..

Dr Michael Harre: Ah yeah there's some great movies out there and the movies out there umm..they portray a combination of robot and ah..robots with some form of human like intelligence but umm..what you see in the robots is a bit more like um..what we think of as human behaviour and sort of expressing our human mind ah..but that's not what we are able to do at all with artificial intelligence yet. The real question is..is..I guess is when we worry about artificial intelligence, do we worry about being super human or do we worry about them being super computational things like ahh.. we can have really intelligent humans, we've got massively intelligent humans. In fact there's an Aussie..

Chris Neff: Yes..yes..yes Mike, I ahh...I'm often accused of being a super intelligent human and I know the feeling.

Dr Michael Harre: It's..it's..it's confronting, it's isolating not to be understood isn't it?

Chris Neff: Mmm..mmm

Dr Michael Harre: Ahh..so (laughs) ahh.. the actual..the people with these.. is an Australian. There is a very famous, in mathematical circles, Australian called Terence Tao. Ah..he's got an IQ they think over 200 um..he's recognised as being one of the great IQ scores of all time. Ahh.. massively intelligent character. Terence isn't going to take over the world.

Chris Neff: No, that's what I meant by it, that's what I..

Dr Michael Harre: (laughs) Well so we have these...these..these human qualities that we think of and ah..then we have what we have in computers and in computers it's really about fast tick over ahh..processing, relatively simple algorithms, those sorts of things, and then what to we do with those.

Chris Neff: So there's a difference between thinking about artificial intelligence relative to complex computations and thinking about it in terms of human behaviour.

Dr Michael Harre: Absolutely. So this actually gets us to how I got into artificial intelligence. Umm..I got a grant very early out of my PHD to work for the US ahh..military on how to put real psychology into a neural network. So how do you take an artificial neural network that is really common in computer science and how do we do things which look much more like what humans do. It turns out that;s a spectacularly hard thing to do. And of course we can only do it with very very small problems. So you can do that, but only in very, very narrow ways. It's not, it's a very specific thing in humans that we are trying to model there. What we don't know how to model is the whole brain and how those sorts of components interact.

Chris Neff: So whats a good example of that? I mean is it flying a jet, is it...driving a truck? You know...I mean..you know..is it that kind of stuff?

Dr Michael Harre: Well what we wanted to know was how humans do those kinds of things right. So what the US military was really worried about, there was a couple of incidents that really worried the US military. One was called the Vincenzo incident. Ahh..which is where there was an enormous failure of communication on the command bridge of this..of this..um..missile ship and it shot down a civilian aircraft. And there was so much going on, there was a great deal of confusion and there wasn't any understanding about the decision making processes that were going on on the bridge that led to this..this failure. So they wanted to..to know how..how we made decisions and how we might be actually be able to..to..to make it more effectively using artificial intelligence. So that was the goal. Um..it turns out to be incredibly difficult. So for about 3 or 4 years I worked on this particular problem of how do we ahh..start to understand the unconscious ahh..processes. How does your intuition inform the way you make decisions, and so you can do that ahh..but only that very narrow thing. So..so..so that's possibly the most interesting thing that artificial intelligence can do for us long term. It gives us a way to look at the way in which we see ourselves to understand what we don't understand about ourselves.

Chris Neff: And what would you say to people who say that AI or artificial intelligence is a threat?

Dr Michael Harre: I would say that w..what people are worried about are probably not the things that we should be worried about. So when we..when we think about artificial intelligence what we usually think about is a..is a super computing brain sitting in the back that wants to take over the world. Waaoo..that's a big call right? Umm...so...we can we can play with that idea right? What are they going to try and do when they get online? They are going to try and hack systems? They are going to try and take down..ahh..I don't know..the CIA..they try and get a hold of the nuclear codes and those sorts of things? OK. That's what people do anyway. Like there are..are black hackers out there that try and do this all the time. So..what are they going to do that we don't already do to ourselves? So the real question is, not what would happen if we had these super intelligent umm..AI's roaming around trying to take over the world..our super intelligent humans don't seem to want to take over the world.Terence Tao’s not going to take over the world as far as I can..

Chris Neff: No in fact, some of the least intelligent humans appear to want to take over the world.

(laughing)

Dr Michael Harre:The..the..point about AI..ahh..is that they just run algorithms, simple computer algorithms right. If you put them into um..a position where they do have access..and you get them to move really fast..so fast that we can't stop them, then you're in trouble. I mean that's where things can go wrong, right. But of course we already do that. We have algorithms that run our trading algorithms on ahh..ahh..on the share markets. The hypothesis at the moment is that they of course flash crashes. There was one in 2010, there was one in 2013 and two in 2014 on the same day.. and we don't really know what caused them.

Chris Neff: There are some algorithms that are setup on the stock market for buying and selling. So they're essentially it's like bots..

Dr Michael Harre: Absolutely..

Chris Neff: ..that are buying and selling based on you know, large bits of data...

Dr Michael Harre: So..

Chris Neff: ..but you're saying that that can lead to problems.

Dr Michael Harre: That..that leads to massive problems. Now these aren't..these aren't..we wouldn't call them intelligent right. These are just little bits of code ticking over, really simple codes that do if, then, else type statements right. Umm..that's not what we think of as intelligence but it's like a really primitive intelligence. But it's those statements, when they're miscoded by humans, that cause these flash crashes. And they run autonomously, we don't have direct control over them. They run so fast, we trade so fast these days that umm..they can run amuck. The algorithms are incorrect in some small way, they hook up with a whole bunch of other algorithms and they trigger a cascading failure in the market. There you go, you have four and a half billion dollars lost in the space of 15 minutes in one instance. That's what should worry you. The idea that AI's is this..is this huge super computing, problematic thing that's going to take over the world.. It is important that they point out any potential dangers of any engineering technology

Chris Neff: Mmmhmm.

Dr Michael Harre:....and scientists should take responsibility for that stuff. Um..it's just unfortunate that I think their angle is very focused when I think it could be much broader and could actually be perhapsmore on target.

Chris Neff: So we've got problems, we are just looking at the wrong ones.

Dr Michael Harre: Absolutely..and you know..we need to manage these problems. Umm..it's the bips we don't recognise. It's the unknown, unknowns right that we really need to be far more umm.. actively trying to seek out.

(Music plays)

Chris Neff: You can subscribe to this Podcast on iTunes or Soundcloud. You can find me on twitter @christopherneff.

So Mike, how does artificial intelligence design and development relate to the research work you do in complex systems?

Dr Michael Harre: So, I'm interested in the way we make decisions. Ahh..so what artificial intelligence gives us is a really clear way to start thinking about how we make decisions. So..how do we put that into a computer? That sort of forces us to think really carefully about what we actually do. So, there are some really clear things that psychology tells us about how we actually think and make decisions in complex situations. When we make decisions, what we do is we pick up on not just the overt signals that we see in our environment, but we pick up on the queues as well in the environment and that changes the way we make decisions. So sometimes we pick up on the wrong queues..ahh...and we make incorrect decisions. So my job umm..and a very large chunk of my research is to do that for economics. So for me ahh..artificial intelligence informs economics about how real people make real decisions in real economic situations.

Chris Neff: So your point is about using artificial intelligence to make humans more efficient, not necessarily to make..robots more efficient.

Dr Michael Harre: Absolutely, yeah..we want to know how we work right.

Chris Neff: Mmmhmm.

Dr Michael Harre: ..and so..ahh...if we can find that out, we can put it into an AI, then we can put that AI into situations that are kind of like ours economically speaking ..um.. in really complex situations we can see how they behave. So that's the idea.

Chris Neff: Where would you locate this conversation about artificial intelligence and research and design in the broader conversation about...you know..whether it's the digital revolution or..you know ..is this the next the industrial revolution? What are we looking at and how do we place this with what else is going on around us?

Dr Michael Harre: It's really hard to pick those technologies when you are in the middle of it. So if you were in the middle of the printing press revolution would you have been able to say well the printing press is the thing. Well there ..there were other things going on at the same time. AI is definitely one of them. It's one of the front running contenders but there is other stuff going on at the moment. There is quantum computing for example. Is it all about networks? Is it about the information networks or is it about the software running over those networks? One of these is going to win this..this...this battle and so that's going to be the revolution in there but we are right in the middle of it. I mean..we are in the heart of this battle and it's really hard to see..what..even what the potential outcomes are, let alone which ones are going to win.

Chris Neff: So where do you see the greatest advantage for the application of AI?

Dr Michael Harre: The most immediate advantage to AI and it's happening right now, is what it's doing for our ability to dig deep into data and our own mental frameworks are relatively limited and what we can use AI to do is help us reformulate um..all of that really intricate data in useful ways to try and understand our health, our economic well being, our engineering directions... all of those things that are so hard for us to picture ourselves, that's where it's going to have the most significant impact. And that's really dumb stuff in terms of human intelligence. This is really data sorting, data aggregating, data filtering, all those sorts of things. So here's a great computer example. There's a computer called Watson. So Watson won Jeopardy, the American game show host..

Chris Neff: Mmm..mmm

Dr Michael Harre: Ah..game show competition. It won against 3 champions and what do you have to be able to do in order to win a Jeopardy against humans. You have to have this massive database of um..information that you have to infer the original question from the answer. That's actually a really non trivial thing to try and do right. So you have to have something really fast that sorts through staggering amounts of detail in really clever ways and comes up with a probability of whether it's likely to be true or not. Now, it's not what humans actually do. Those 3 competitors that Watson was playing against were doing themselves something vastly different from what Watson was doing. Ahh..so we..we..this is not human like computation in some sense. So that sort of artificial intelligence, we don't see any dark side to..to that sort of analysis...ahh..sorry. There's privacy issues and all those sorts of things around..around..um sort of legal issues and those are all really relevant and we need to be very careful of those ahh..but in terms of the benefits that come out of the potential to do it, there is a lot of positivity there.

Chris Neff: Ahh..I understand you ah..worked on developing an artificial intelligence related to the Chinese game Goh.

Dr Michael Harre: Yeah, so..one of the things about Goh and chess and any of those really complex situations..

Chris Neff: Can you tell us what go is?

Dr Michael Harre: Ahh..so Go is a ancient Chinese or oriental um..game. So it's a board game and the idea is to..to earn territory to..to beat your opponent in a more complex game than chess. So that's where..that's where..where part of the modern interest in artificial intelligence comes from. They are playing our most complex games far better than we are. So.. the question that we were asked was..um..how do we play go when our ability to hold information in our conscious memory is so small? And we know how small it is, we can actually measure it. Umm..and it's tiny and in computer science it's called..its about 4 bits..it's..it's nothing..umm..but we make really complex decisions using much much more information than what we actually are aware of. So how do we..how do we do that? How do we actually put that into an AI? Well so you can do it umm..by having a neural network which picks up on all the queues, not the entire board configuration but just the..the statistically regular queues on the board and then you combine that with local configurations, small chunks of..of things that we are actually consciously aware of. Once those 2 systems in our..in our brain come together, um.. that's how we make our decisions. So that's how we actually build those sorts of AI's or that's how we would if we would actually build a human brain.

Chris Neff: Can I ask if there's you know..,how does AI contribute to the inequality gap because whenever we think about making more efficient umm..calculations about budgeting or economics that usually translates into rich people getting richer and poor people getting poorer..you know when..

Dr Michael Harre: We would hope that it opens it up to being far more egalitarian and..and..a perhaps I can illustrate that with an example. Ok so in..in Australia, you've got a..you've got a mineral extraction base. There's a limited number of spaces you can move into. If you've got a technology umm..economy like Japan, there are other spaces you can move into. And the idea is that you want to move your economy in a direction that really grows your economy. And in principle that should lift the whole base of the economy right. So GDP correlates with overall health and those sorts of things right. So..with an AI, you could literally sort through all of the opportunities you've got to move your economy in the right direction. So if you're a policy maker um..in Central Africa and you wanted to start thinking about how to move your economy, you could start with an AI that could look at all the potential ways in which you could reconfigure your economy over the next 20 years. That's actually something you could do. And so that sort of poverty gap can help to be mediated by these..by these technologies. And you know you can run..because it's in the cloud you can run them off desktop PC.

Chris Neff: Wow..so it's like a package of software?

Dr Michael Harre: Yeah, absolutely. Literally, you just need someone who knows how to run the software and that's not terribly difficult. You need the political will..don't get me wrong..ahh..but that's the human side. There's no reason why you couldn't do that in any economy.

Chris Neff: So where do you think, Mike, the.. focus of attention should be on AI in the future?

Dr Michael Harre: I think ahh..the..the biggest benefits are already in play. Umm..so the..the engineering and the health side of those things, that's already been done and that'll..that'll roll out over the next 20 to 50 years and that'll have massive benefits for hopefully everyone. Umm..cheaper resources, cheaper medicines, umm..cheaper access to all sorts of technological umm..results and those sorts of things but..but that's already happening so that'll just play it's own course. The real umm..power of artificial intelligence is to try and understand how we can actually build these in ways that are in some sense..a little bit more interestingly like us. Ok..how can we..how can we understand ourselves in a sort of reciprocal way. How can we understand AI's to understand us.

Chris Neff: Ahh..Mike each week I ask our guest what the relationship is between their research and sharks and I've been waiting for this week. What's the relationship between sharks and artificial intelligence?

Dr Michael Harre: Umm.. I think the..I think the relationship between sharks and AI's is that what we can do is manage fish stocks and in particular we know that sharks are um..top level predators and we know that ahh..they are enormously endangered. Like we..we wipe out hundreds of millions of sharks each year umm..so if we want to understand how ecosystems are stabilised, how we can manage them better, the best thing we can do for sharks is apply artificial intelligences in the same way we talk about economics to the problem of managing of food and fish stocks to the benefit of sharks.

Chris Neff: That was a very polite answer to a potentially controversial questions. So there's a whole industry within Hollywood that teaches us to fear artificial intelligence and ah..the string of movies is not stopping, it's expanding. Why do you think that is?

Dr Michael Harre: Because people love existential doubt and..we're all in fear and we are looking for things to..so it's the same thing with nuclear weapons, the same thing with biological ageing...meteorite strikes. We love to play with the idea that we are all going to be wiped out by something. As curious as that is, and so that's what Hollywood feeds into. Whether it's real or not...for artificial intelligence..

Chris Neff: This is..this is an asteroid movie basically starring AI is what you're saying.

Dr Michael Harre: (laughs) Exactly, umm..but in terms of AI I think what people are really worried about is not actually umm.. so much what the practical reality of AI is..what they are probably most worried about is this vision of a world dominating umm..thing which can somehow ahh..control our lives. Ahh..it's hard to imagine the actual practicalities of that. Like how..how would that actually roll out in practice in terms of what they have to get hold of, how they have to break security codes, how they have to control things, it's not clear. Humans aren't able to do that and we are not clear..if it's even theoretically possible to do that so..I don't know, I have my doubts.

Chris Neff: Well in the movies..they..they do it in 5 minutes but I do appreciate your point that AI's is about humans, it's not about robots and if you were going to be concerned about the technology and the research and design of AI's, it is around these, you know the way that..the way that some of it can be used by..by humans, not by robots.

Dr Michael Harre: Absolutely, and on that point..umm..they are meant to be efficient. They are there to make us efficient at doing something right. And umm..in doing so they are faster than us. And so the problem is that these very small pieces of code are so much faster than us that they run away from us very quickly. And so that's the real danger..that it's the speed in which these things ahh..ahh..react. They don't have to be intelligent, they just have to be faster and more connected.

(Music plays)

Chris Neff: Thank you very much, thanks for joining us today Mike.

Dr Michael Harre: My pleasure, thank you for having me.