The West Of Ireland & AI With Murray Shanahan

Murray Shanahan is a professor of cognitive robotics at Imperial College London and a senior research scientist at DeepMind. With a PhD from Cambridge University he has carried out work in Artificial Intelligence, Robotics, and Cognitive Science.  Murray was a scientific advisor for the 2014 film Ex Machina. In 2017 he joined DeepMind, retaining his professorship at Imperial College on a part-time basis.

In the podcast we spoke about John McCarthy, the son of an Irish immigrant also called John from Valentia Island in Kerry  and maybe the most important Irish American computer scientist pioneer and inventor, known as the father of Artificial Intelligence (AI).

--------------------------------------

FOLLOW MURRAY SHANAHAN, Professor
➡️ Twitter:  https://twitter.com/mpshanahan

BOOKS BY MURRAY SHANAHAN
https://amzn.to/38bJPZI

--------------------------------------

Oriel - A Magnesium For Sleep 😴
Affiliated With Oriel Magnesium Store: Get deep sleep💤, boost your energy 💪 and immune system.

Disclaimer: This post contains affiliate links. If you make a purchase, I may receive a commission at no extra cost to you.

Chapters

• 0:03 Introduction
• 1:39 Murray Shanahan Introduction
• 2:32 Artificial General Intelligence
• 3:56 DeepMind
• 5:20 AlphaGo System
• 7:56 John McCarthy & AI
• 11:08 The West Of Ireland & AI
• 12:06 Deep Learning
• 14:05 Machine Learning Techniques
• 15:53 Close

Transcript

Ina O'Murchu : 0:03

Hi, and you're very welcome to my podcast show. In this episode, I got the chance to speak with Murray Shanahan. Murray is the Professor of Cognitive Robotics at Imperial College London and a senior research scientist at DeepMind. With a PhD from Cambridge University, he's carried out work in Artificial Intelligence, Robotics, and Cognitive Science. Murray was a scientific advisor for the 2014 film - Ex Machina. In 2017, he joined DeepMind, retaining his professorship at Imperial College on a part time basis. In the podcast we spoke about John McCarthy, the son of an Irish immigrant, also called John McCarthy, from Valencia Island, in Kerry, on the west coast of Ireland, and maybe the most important Irish American computer scientist, pioneer and inventor, known today as the father of Artificial Intelligence. In the mid 1950s, John McCarthy coined the term Artificial Intelligence which you would define as the science and engineering of making intelligent machines. We also spoke on the podcast about George Boole, a largely self-taught English mathematician, philosopher and logician. Most of his short career was spent as the first professor of Mathematics at Queens College Cork, in Ireland, now known today as University College Cork. He worked in the fields of Differential Equations, of Algebraic Logic, and is best known as the author of the book "The Laws Of Thought", which was first published in 1854, which contains Boolean Algebra. Boolean Logic is credited with laying the foundations for today's Information Age.

Murray Shanahan : 1:39

So my name is Murray Shanahan and I am Professor of Cognitive Robotics, Imperial College London, or at least I am for one day week - that used to be my permanent position, my full time position. And the rest of the time I work at DeepMind, which is this London based AI research lab owned by Alphabet bought by Google. So that's where I spend most of my time. DeepMind you know, were a startup, they started - back in 2010 and their, the idea of the founders, Demis Hassabis and Shane Legg, was to facilitate, was to try to create Artificial General Intelligence and Artificial General Intelligence, that really is what we think of as AI. When we think about AI in science fiction, it's really the kind of thing that we don't really know how to build yet. But AI that has the same kinds of abilities as human beings. So today, we only see these things in science fiction films but there's every reason to think that we might be able to build, build into computers, the kinds of, the kind of level of intelligence human beings have. And that little word general in the middle of Artificial General Intelligence is there because AI today is actually is a very important thriving field and we can do amazing things with with AI and Machine Learning in particular, that have huge commercial application but all of those things are narrow applications for Artifical Intelligence, usually, your Artificial Intelligence system can only carry out one task. Like, for example, labeling images in your photo album, or maybe answering a few questions, you know, from Wikipedia or something through Alexa or Siri, or something like that, or self driving cars or playing chess. So those are all different systems. It's not like one system that's doing all those different things. It's a different system doing each of those tasks. And the way we human beings are able to learn to do any of those things. And that's why the key word General is there, we want to try and build Artificial General Intelligence so something that can do a whole range of tasks, the whole range of intellectual tasks that a human being can do.

Ina O'Murchu : 2:47

It's not tomorrow, might take a little bit of time to get there.

Murray Shanahan : 3:56

It's not tomorrow, but it's the mission. The mission of DeepMind was to try to fulfill that kind of dream and eventually build this Artificial General Intelligence. And they had, so DeepMind had a great success back in 2014, where they produced this little system that was used in technical, deep reinforcement learning. So that's deep learning with neural networks, and reinforcement learning, which is kind of trial and error learning. And they used this deep reinforcement learning system to learn how to play these Atari video games, those retro video games like Space Invaders and Breakout, and so on. And the great thing about this little system was that all it saw was just the pixels on the screen it didn't know what you know what the objects were there it didn't know.... You know, it didn't know anything at all, except that just the raw pixels on the screen and the score. And just based on trying to, you know, play the game by trying random actions and figuring out what you know, gave it a positive score. It gradually learns to get and can get very, very, very good at these games and you could give it any one of these sort of suites of, of Atari video games it would learn to get really, really good at those those games. So that was enough to really, to impress Google, who then bought DeepMind and then funded, funded us, you know, in a very, very, generous case. So now we're in a position to pursue just pure research, really, the majority of people here are doing pure, pure science, pure research, and 2016 I think it was - was to the production of this AlphaGo system, which learned to play the Chinese game of Go. A very, very difficult game, more difficult than chess in many respects, mathematically speaking, it's more complex than chess. And although chess was something that we could solve with computers, you know being better than humans, way back in 1997, the first computer was able to beat the world champion at chess, Deep Blue, beat Garry Kasparov in 1997. Nobody thought that Go was the sort of thing, where humans were going to be beaten by computers for a long time yet, because it's just mathematically just far too huge a space of possibilities to explore using just brute force search, which is the way Deep Blue worked. So it came as quite a surprise when, when DeepMind produced this AlphaGo system, which was then able to beat the world's very, very best player. So Lee Sedol, very famous top, absolutely top of the world player was beaten by the the AlphaGo system. And then Ke Jie was beaten and now I think all great Go players recognize that the latest incarnations of AlphaGo are really way better than even the best human.

Ina O'Murchu : 6:46

So we are getting there

Murray Shanahan : 6:47

Well, I'm not I'm still not sure that we're getting there because those are still kind of, it's very impressive that you can actually get to superhuman performance, or something like Go but it's still a very, so AlphaGo, can't you know, go into into my kitchen and make a cup of tea or, or or learn how to make a cup of tea or figure out how to make a cup of tea in say, a strange kitchen that it's never seen before. So humans have this, this ability to generalise from past experience to completely, to new kinds, kinds of settings, drawing on a common sense understanding of the world, but we still don't know how to get computers to do so that's very much the kind of thing that I'm interested in...

Ina O'Murchu : 7:28

It's how to transfer that General Intelligence?

Murray Shanahan : 7:29

It's how to transfer or how to build into it the details

Ina O'Murchu : 7:35

We shouldn't be too overly concerned not yet anyways, it's not this year.

Murray Shanahan : 7:39

You could be concerned or you could be excited on how the implications are going to be at developing AGI but you don't have to be either just yet because it's not quite around the corner but if it were around the corner, then, you know, if and when it is, we do need to think quite hard about how we use such a very powerful technology. My family history, so my father's family come from County Kerry and my father's family all came from Valentia Island off the coast of County Kerry. So my father's father's family and my father's mother's family all come from that one little island. Valentia. And about 15 miles off the coast from Valencia Island, is a little fishing village called Cromane, the founding figure the founding father of Artificial Intelligence is a man called John McCarthy and John McCarthy, he coined the term Artifical Intelligence and his father was born and raised in Cromane, in County Kerry, so John's family were really very much a Kerry family, so he absolutely, you know, that's where his pedigree was from County Kerry and and so he John really is, the person who is, you know, more than anyone else, credited with founding the whole field and it dates back to 1956. So in 1956, there was a sort of inaugural conference, that John organised in Dartmouth. So this is known as the Dartmouth conference. And in this conference, many of the thinkers who then became the leaders of the whole field, for many decades to come, they sort of got together for the first time and, and discussed the possibility of making computers that will carry out these kinds of, you know, intellectual, intellectual tasks like understanding language and proving mathematical theorems and things like that. And so John actually coined the phrase Artificial Intelligence in the little proposal document that he wrote for that Dartmouth conference. And then of course, he also made many you know fundamental contributions to the whole field from from early on, and I knew him quite well so I was very lucky to meet him and I got to know him quite well because we worked in the kind of, well I was working at that time, there, in this sort of corner of Artificial Intelligence very much based on Logic, which is the corner of AI that he pioneered more than anything else that he made a lot of contributions to. And so I started, you know bumping into him in conferences and found myself presenting to audiences where this very famous, you know, elderly statesman from the field was in the audience and pretty soon I got to got to know him quite well and then he even invited me, so he was at Stanford at that point in his in the States, in California, invited me to go and work with him a couple of times in Stanford. So I had a couple of visits with him, for just short visits for a few weeks at Stanford and generally I knew him pretty, pretty well. He died in 2011, I think, sadly, but very, very important in the field.

Ina O'Murchu : 10:46

Wow. That wasn't that was not so long ago. How fantastic is that? That's, that's pretty cool. Yeah.

Murray Shanahan : 10:56

So I would like credit, you know, the whole field to, I'd like to make County Kerry (attribute it) for the field of Artificial Intelligence but unfortunately, I think we have to give a bit of credit to Cork. Because there's another really interesting connection with the west of Ireland and which is which is through George Boole. Now George Boole was a professor at the University of Cork, University College Cork, in the 19th century. So this is long before computers and Boole was a pioneer of mathematical logic when you when you think about And Or Not, so the kinds of things that are absolutely at the foundation of computer hardware, and are built into all kind of computer languages, and an absolute foundation of Logic. That was all started by George Boole and much of the pioneering work was done in, in Cork, in the city of Cork, you know University College Cork, and every undergraduate of Computer Science students, every undergraduate student had to learn Boolean Algebra, you know in their first year and it's absolutely the foundation of computer hardware and, you know, and reasoning about how computer programs work and so on. So Boolean was some sort of foundational, but another interesting fact is that one of Booles, great great grandsons is Geoffrey Hinton. And Geoffrey Hinton is the person who perhaps arguably more than anyone else is responsible for the current boom in Artificial Intelligence. So Geoff is uh, you know, Geoff is very much a living person who was a professor at Toronto University, and he really sort of started inaugurated the whole field of Deep Learning those who think we're in Deep Learning based on neural networks is at the heart of you know everything that's making AI such a successful field, and he is George Booles great, great, grandson. So it might be a little bit tenuous, but there's definitely this Irish, western Irish connection there so much as I'd like you know to award the whole field to County Kerry I have to give County Cork a little bit of credit.

Ina O'Murchu : 13:00

Good, very good. And there is a difference. That's for sure that people of Cork will let you know! Yeah, so John McCarthy's background and George Booles, I mean, they're, you hear the names but you don't make the connection. So it's a really nice story to be able to tell. The way with everything has gone and the way most of the funding goes to Harvard, MIT, you know, it's okay, a lot of the money goes that direction to Stanford, you don't really tend to hear about the sort of foundations you know where these professors come out of. And also there is I suppose the tradition of things being in the family? It gets into the genetics somehow, I think?

Murray Shanahan : 13:04

Yes. Yeah. I mean so clearly, Geoff, comes from a very academic family. So I looked it up on Wikipedia earlier on because I knew you were going to ask me about it. And in fact, it says it seems like his whole family of Booles, the many, Booles descendants were academics as well.

Ina O'Murchu : 13:35

Yeah, so I really think a lot of things do take generations to, I mean, in Artificial Intelligence has been a slow, a slow burner. Oh, boy. Oh yeah. So we just didn't have the Data did we?

Murray Shanahan : 14:05

Yeah, well Data has been the critical thing, to really enable, just checking for Deep Learning for it to work. It kind of worked in the lab back in the 1980s and 90's. But it wasn't really, you know it sort of reached limits where it wasn't really very, successful on, you know, complex images with lots of with high resolution. There's only really with availability of very large quantities of labeled data, that it's been possible to really make the best use of these Machine Learning techniques.

Ina O'Murchu : 14:39

Okay, very interesting Murray. Thanks very much for that. I mean, with your surname as well. I mean, there's no escaping the Irish connection, for sure!

Murray Shanahan : 14:50

My daughter is called Kerry and my son is called Liam so...

Ina O'Murchu : 14:53

My God. That's really strong! Yeah, there's so many things that you only find out about and you don't tend to find out about and you don't tend to know about until you go away and people tell you about things. You know, it's it. There's so much - so much of history I don't know much about, you know, in Ireland, and everything just kind of falls in and you figure out things as you go. This is a really nice story.

Murray Shanahan : 15:15

So John told me that he'd been over to County Kerry, I remember telling him that my family came from Valencia, and he said, oh, that he'd been out there and he thought he had some cousins, you know on Valentia. And I thought this was, I thought this was a kind of, you know, how sometimes, you know, Americans, think the tiny country everyone knows everybody and I had no idea actually until after he died, that he meant it really, he really meant it. Really he came from really just a few miles from where, where my family came from, but I didn't actually know that at the time.

Ina O'Murchu : 15:47

Yeah in Ireland it's just one degrees of separation, one and a half degrees. Yeah, definitely. It's not six degrees here. Thanks for listening and check out my podcast. You can head over to Spotify and find my podcast there or on Apple iTunes. Subscribe to my podcast on my website, you can head over to www.inaom.io for further details.