Often when I give a public talk, somebody in the audience asks me about artificial intelligence and it's not a subject I feel particularly comfortable to talk about. I consider myself a roboticist, that is, somebody that builds machines, whereas artificial intelligence to me is a completely separate discipline, a sub-discipline within computer science.
Intelligence, of course, is also a very difficult concept to understand and I think it also has something to do with the historical context. Way back in lecture number one, we looked at this picture of the famous digesting duck. And at that time, people considered that this machine had life-like properties. Whether it was intelligent or not, I'm not sure, but people considered it to be very life-like.
Later on came these automata, incredibly intricate machines that could do seemingly intelligent things. And perhaps at that time, people would look at these machines and consider them to be intelligent because they did the sorts of things that intelligent beings or entities, such as humans, are able to do quite readily.
In the late 40's and early 50's, William Grey Walter built a number of cybernetic devices, most famously his cybernetic tortoise. And this was a machine that contained relays and vacuum tube technology.
It was able to exhibit light-seeking behavior. So if you put a light on the floor, this tortoise would move towards the light and stop when it got there. So again, this was a seemingly intelligent behavior. This light-seeking behavior is seen in many very simple organisms.
So at the time, late 40's, early 50's, something like this was perceived as being intelligent. Looking at it from our current viewpoint, it looks rather primitive and sad but at that time, I think it was a significant breakthrough.
The term artificial intelligence was probably first introduced in this research proposal from 1955 by a bunch of very imminent people, amongst them John McCarthy and Marvin Minsky, who are very very well known names in the artificial intelligence community.
By 1955, digital computers had become a reality and this group of scientists were interested in trying to explore what could they do beyond straightforward numerical calculation. So they were interested in things like neural networks, theories of computation, could we use computers to do natural language processing, could they perform acts of creativity?
What you'll notice from looking at the list of topics that they proposed to investigate in their summer project was there's no mention of robotics and there is no mention of computer vision. The reason for that is that those two things had not yet been invented.
The first industrial robot as we know it today was developed by Unimation Inc. and that company wasn't even formed until 1956, one year after the artificial intelligence project and Unimation delivered its first robot in 1961. But thereafter, a number other robots appeared in quite rapid succession.
They were all developed within artificial intelligence laboratories at Stanford and at MIT and chief amongst these robot designers was Victor Scheinman who designed the ORM and ORM I believe, is Norwegian for snake and he also designed the very famous Stanford arm in 1968.
Computer vision really got its start in 1964 with the PhD work of Larry Roberts at MIT. He was interested in understanding how computers could interpret a blocks world. So a camera attached to a computer would take a gray-scale image of a scene and with various sorts of early image processing algorithms, could find the edges and then fit a model to that. Larry Roberts went on to do really great things in the creation of the Internet.
So then, what is intelligence? I can look in the dictionary and get a definition of intelligence, but to me, this is not particularly helpful. I think we assume that human beings are intelligent so therefore, the behaviors manifest by human beings count as intelligence.
But we know that there are many animals that can exhibit intelligent behavior and it's been found in more and more simple animals. Once upon a time, we believed it was perhaps only the great apes that could exhibit some sort of intelligence but now scientists have found quite intelligent behaviors in all manner of species. Even birds are able to fashion simple tools to help them to achieve particular tasks.
A definition I particularly like comes from Rodney Brooks and that is that intelligence is in the eye of the beholder. So if you're looking at a machine doing some task and you think it's intelligent, then it is intelligent.
Perhaps the best known definition of intelligence is the Turing test and it was proposed by the English mathematician, Alan Turing, in a 1950 paper called Computing Machinery and Intelligence. In that particular paper, he proposed something called the imitation game, and of course that's the name of the recent movie about the life of Alan Turing.
The game has three players: A is a computer which is being tested for intelligence, B is a real human being, and C is the judge. The judge doesn't know which of A or B is the computer and so asks questions of A and B to try and determine which is which.
In order to mask the really obvious cues about which one of A or B is the computer, the game is generally played with keyboard and screen communication, so there's no verbal communications. The judge types commands on a keyboard, A or B respond with information displayed on a screen.
Now if the judge, after asking sufficient questions, is unable to determine which of A or B is the computer, then the computer is deemed to be intelligent because the judge, who is intelligent, is unable to tell the difference between it and a real human being.
Let’s do a quick recap of where humanity is with respect to machine intelligence, and how our concept of intelligence has changed over time.