What is AI?

It's a slippery definition...
08 October 2019





What is AI?


The University of Cambridge's Beth Singler took on this question for us...

Beth - It's quite a slippery term, and some people use it when actually what they're talking about is not necessarily Artificial Intelligence. And one of the ways to understand the term and where it comes from and what it means is to think about the history.

And the term was first sort of coined and used at a particular conference in 1956 at Dartmouth College. About 8 very intelligent people have been working computer sciences got together and decided there was this thing that they could create called artificial intelligence where they could create machines that could do all the things that humans could do basically. Perceive things, understand things, make decisions, be broadly intelligent by their understanding of what intelligence could and should be.

And since then the term has really been applied wherever some machines seem to be doing something that’s smart. Under that umbrella term you do get things like machine vision systems, expert systems, robotics gets lumped in with A.I. as well. And a lot of the time when people point to something and say “that's artificial intelligence” it might not actually be in the same category as those original thinkers were talking about. It's become something that's very applicable to lots of different places.

At the very simplest level, it's the use of algorithmic systems to make deductions that then become part of the data set, that then go back into the algorithmic systems of that kind of iterative process and it can do some very spectacular things. So if you've been watching the news about a particular A.I. game systems like Alpha Star and Alpha Go from Google Deep Mind, it can be very very super intelligent and very narrow field. So playing computer games particularly well because it's iteratively learned how to play computer games very well on this algorithmic process.

What it can't necessarily do is decide it's going to stop playing the computer game. So it has that very narrow form of intelligence that isn't directly mappable onto human intelligence in the way that the original founders, those eight gentlemen in 1956, thought. And by the way they thought they'd fix this whole problem of artificial intelligence in about two months, over summer with a 10 man team. “We'll just get it we'll get it done”.

So now A.I. is something that has been applied to lots of different things, partly because of our conception of it from science fiction as well. So our ideas of what it is and could be are very much influenced by sci fi tales and you know a lot of companies now say that they're using artificial intelligence and they might not necessarily be using exactly the same thing. They might be doing advanced statistics or as one chief technology officer said “I'd just rather say we did maths” and leave it at that.

But A.I. has this kind of glamour around it. So as I say it's a very slippery term. It's hard to kind of pin down to some people what it actually is.

Adam - So your computer might be very good at doing video games but it's not going to be able to, say, diagnose patients with a heart condition?

Beth - Exactly. So there are A.I. systems that can do diagnostics in that way, look at images, machine vision imagery, and they can check that and make a diagnosis. But no it can't be doing a broad range of things at this stage, although that is the ultimate goal - Artificial General Intelligence - for a lot of people.

Phil - If the criteria for an A.I. is that it can't stop playing a video game then a lot of people out there might be included.

Beth - True yes. So that's one of the interesting things seeing the comparison between A.I. systems playing computer games and humans playing computer games. And some of the world's best games players are exactly those people who have played hours and hours of games against other players.

But what A.I. can do exponentially better is play against itself. Millions millions of times to improve. And again that's a very narrow usage of A.I. Alpha Star can play Starcraft 2 fantastically well. But they discovered you know initially it was beating humans but that was when it was capable of seeing the whole map. Once they could see the map as a human sees a map, bit by bit through the fog of war, it lost to a human player. So these situations are also set up for A.I. to show off what they can do as well.


Add a comment