What is an intelligent machine? On Friday, Roger Brockett, a roboticist at Harvard University, gave a lecture at the McGill Centre for Intelligent Machines that attempted to answer this question.
In the 1950s, the Turing test was the standard for determining whether or not a machine was intelligent. In the test, a human interrogator engages with another human and a machine, all of whom are physically isolated from each other. The interrogator asks questions in the form of written texts to which the human and the machine respond anonymously. If the interrogator can’t distinguish between the answers, then the machine is judged to be intelligent.
But according to Brockett there are many flaws in this definition.
“Exams are not open book tests, but life is,” Brockett said. “In a Turing test you are in a sterile environment. You have no help in answering those questions. Moreover, that makes it more or less the standard aptitude test, where you either check the boxes or you do not. On the other hand, in real life, it is an open book test. You get whatever choices to solve the problem. The Turing test doesn’t reflect that.”
“[A] Turing test is really a test for interaction with humans as if it is a human. But there is more to intelligence than just interaction,” added Jeremy Macdonald, a PhD candidate in mathematics at McGill.
Brockett described three pillars that are essential for a machine to be intelligent: it has to extract information from, respond to, and make changes to the environment in which it exists.
Where does that put machines that Canadians use every day, such as search engines? To answer this, Brockett described the scenario of searching for an airplane ticket online. The search engine would extract all the information related to your flight and rank them in order of relevance. If the airline cancelled the flight, the results would reflect that. Search engines, then, are very intelligent.
Brockett admited however, that his criteria also has some flaws.
“When I was 10 years old, I read this Time magazine article where this psychologist would hang bananas on the ceiling and scatter around the room boxes and crates and place a monkey in the room and see if the money can stack the boxes together and jump on it and get the banana,” Brockett said. “He would observe the monkey’s behaviour through a keyhole. What he found when he peered through the keyhole was the monkey looking straight back at him”.
If intelligent beings sometimes fail to grasp the objective of tests, how can we expect machines to do it perfectly?
To answer this type of question, Brockett encouraged young people to do research. In particular, he said that inquiries into prosthetics, providing companionship and help to the elderly and impaired might be the most powerful ways to learn about intelligent machines.
At the McGill Centre for Intelligent Machines, much work has already been done by roboticists like Professor Frank Ferrie. This idea appealed Chris Warren, a U2 electrical engineering student who attended the talk.
“I want to build a utopia where human labour will be substituted with machine, where there is no scarcity, and no need for labour,” Warren said.
Others are more skeptical.
“It is not so much an engineering question as it is philosophical,” Macdonald said. “He needs to be more expansive.”
Ultimately, said Brockett, perhaps it’s not the definition of what it means to be intelligent that matters, but the contribution that machines make to society.
“We can build football players out of robots, but we devalue everything that machines can do unless it is intrinsically human,” Brockett said. “Poor old intelligent machines are going to need a powerful advocate if they are to get their due.”