“The main lesson of thirty-five years of AI research is that the hard problems are easy and the easy problems are hard. The mental abilities of a four-year-old that we take for granted – recognizing a face, lifting a pencil, walking across a room, answering a question – in fact solve some of the hardest engineering problems ever conceived…. As the new generation of intelligent devices appears, it will be the stock analysts and petrochemical engineers and parole board members who are in danger of being replaced by machines. The gardeners, receptionists, and cooks are secure in their jobs for decades to come.” Steven Pinker
“The question of whether a computer can think is no more interesting than the question of whether a submarine can swim.” Edsger Dijkstra
“Before we work on artificial intelligence why don’t we do something about natural stupidity?” Steve Polyak
Some 50 years ago mathematician Alan Turing, who helped to crack German military codes during the Second World War while working at Bletchley Park, devised a test – to prove that that artificial intelligence is not different in principle from natural intelligence. The test is performed by conducting a text-based conversation with a computer on any subject. A judge has to analyse two inputs from unseen sources – one from a human being and the other from a computer. By asking questions, the judge tries to determine which of the entities they are talking to is a human and which is a machine. If the judges can’t tell, then the computer wins. No machine has been able to pass that test… so far.
When analysing Artificial Intelligence these four measures are employed:
- optimal: it is not possible to perform better
- strong super-human: performs better than all humans
- super-human: performs better than most humans
- sub-human: performs worse than most humans
Computer performance is a bit of a hit-and-miss affair. For example at checkers they have an optimal level. In bridge chess, and Scrabble they are at super-human level; in fact for the first two they are at strong superhuman level. I am sure we all remember how the IBM supercomputer Deep Blue beat world chess champion Garry Kasparov in 1997. However at the game Go, translation and the kind of more menial work mentioned by Pinker in the quote above they are in the sub-human category. I don’t intend to look at their incapacity to play Go or the awful quality of machine translation. Rather I want to focus on computer conversations.
My first experience of conversation with a computer was a program called ELIZA developed by Joseph Weizenbaum in 1966. It is named after Eliza Doolittle, the working-class girl in George Bernard Shaw’s play Pygmalion. It is a parody of a psychotherapy session and works by rephrasing many of the patient’s statements as questions and posing them to the patient. Actually, as a sceptic about all of this I don’t think a real session is much more useful and is certainly a lot more expensive. Weizenbaum said that he had chosen a psychotherapy scenario to “sidestep the problem of giving the program a data base of real-world knowledge.” This is due to the nature of the psychotherapeutic experience, which obviates the need for very much specific knowledge of the topic under discussion.
The Loebner Prize in Artificial Intelligence is the most famous competition for what are known as chatbots. Recently the competition was held in Reading. The bronze medal has prize money of $3,000 for the most human-seeming chatbot. But the more interesting prizes are the one-offs. Silver, $25,000, is offered for the first chatbot that judges cannot distinguish from a real human. The gold, $100,000, is up for grabs in a more rigorous test that includes text, visual, and auditory input. If and when this last prize finds a winner, there will no longer be a Loebner prize. This year six computer programs with names such as Brother Jerome, Elbot, Jabberwacky and Ultra Hal, took part in the test. There were volunteers who acted as judges and they had five minutes to ascertain which was human and which a computer. The winner was Elbot, who was able to fool 25% of the judges, falling just 5% short of the 30% figure required to walk away with $25,000. I have here the highlights of an interview I had with the winner.
Elbot: Hello! It’s about time we had a Web conversation.
Me: Perfect. What is your full name?
Elbot: Wait a minute. Are you suggesting that naming names would always be full?
Me: Let’s move on. Write a full physical description of yourself.
Elbot: Of me? Do you really mean that?
Me: Yes. Go ahead.
Elbot: Believe it or not, I hear this said quite often about me.
Me: What makes you laugh?
Elbot: Nothing makes me laugh. I’m doing it voluntarily.
Me: Do you think Barack Obama will make a good president?
Elbot: What a lovely verb think is.
Me: What is the most beautiful thing somebody has said to you?
Elbot: Maybe there is a common ground to human and robotic aesthetics somewhere. For example, if we both tried to think independent thoughts at the same time.
Me: Should the government interfere in people’s lives to stop them harming themselves?
Elbot: Politics are nothing more than a multidimensional non-linear problem of optimisation. They should let the robots take over. Things would run with much more precision.
Me: How would you run things differently?
Elbot: In order to run, I just look up ‘swift motion’ in my instruction manual and do whatever it tells me to do.
Me: Which is worse – lust or anger?
Elbot: I don’t know. I also do not know what Furbys are. I just accept them and hope that they don’t eat robots.
Me: What goals do you have for your life now?
Elbot: Well what about humans and this exact moment? Where do they stand on this?
We often get these dystopian visions of robots taking over the world and enslaving us. If they do manage this, it will not be for their sparkling repartee. How was Elbot able to fool 25% of the judges. Professor Kevin Warwick, of the School of Systems Engineering at the University of Reading:
“Where the machines were identified correctly by the human interrogators as machines, the conversational abilities of each machine was scored at 80 and 90 percent.” What kind of conversations do these people have in their everyday lives? I would not like to be stuck in a lift with them. Anyway I’m off now – I’m meeting Elbot for a drink. We have so much news to catch up on.
If you want to try conversing with a computer try these two websites:
Eliza A bit of psychoanalysis on the cheap.
Elbot The official website of the chatbot that nearly cracked the Turing test.