Artifical intelligence a long way off from Skynet: expert
Terminator-like robotic uprising unlikely, says University of Toronto professor Geoffrey Hinton
April 21, 2011, is a significant date for artificial intelligence enthusiasts and cult followers of the Terminator movie and TV series.
It is the day machines are supposed rise up and launch a devastating attack on human civilization, according to the most recent timeline of the series (which gets a bit complex because of time travelling characters who change the past).
As the story goes, after becoming self-aware on April 19, 2011, a U.S. computer defence program called Skynet launches an attack on its creators on April 21, leaving only a ragtag band of flesh-and-blood rebels on Earth.
But, how close are we really to a computer that remotely resembles Skynet?
CBC News posed that question to University of Toronto computer science professor Geoffrey Hinton, who specializes in artificial intelligence. In an email interview he said the possibility of a computer launching a cataclysmic event like the one in the series is unlikely in the real world.
CBC News: Why haven't we seen an apocalyptic takeover by machines?
Geoffrey Hinton: Because artificial intelligence, despite its age, is still in its infancy.
How far away are we in terms of having machines that can think for themselves?
We already have machines that can make plans and can execute these plans in symbolic domains, like the web. What we don't have is machines that can get around effectively in the real world. But robotics and computer vision are improving all the time, so that will happen in the next 10 to 50 years.
Researchers already have moderately autonomous cars.
Some notable forms of artificial intelligence
Deep Blue was a chess-playing computer developed by IBM that beat champion Garry Kasparov in 1997, though the robot had been defeated previously in 1996 by Kasparov.
Watson, another IBM computer, handily bested two previous Jeopardy champions in February 2011, including long-reigning contestant Ken Jennings.
TOPIO is a Ping-Pong-playing robot designed by TOSY, a Vietnamese robotics firms. However, its playing ability isn't quite up to human standards, according to an article from Popular Science.
How would you characterize the mental capacity of the most advanced form of artificial intelligence?
Computers have very impressive mental capacity in domains such as chess and Jeopardy. They are still far worse than people at perception, motor control, understanding natural language, learning and common sense reasoning. But they are getting better all the time.
To what degree, if any, are artificial intelligence systems employed in military technology?
Once a system is deployed, people often stop thinking of it as artificial intelligence. In the 1950s, for example, programming a computer using Fortran seemed like artificial intelligence. So, lots of the [currently] deployed systems use technology that was once considered artificial intelligence.
How does studying the neural pathways of human beings help design better forms of artificial intelligence?
The things that people can do much better than computers, like perception or seeing analogies, are the result of the extraordinary learning abilities of the human brain.
Theoretically, it's conceivable that we could develop artificial intelligence that works in a very different way from the brain, but I don't think it's a good bet.
My bet is that understanding the type of computation and learning that is going on in the brain will be the key to developing systems that can outperform humans in a wide variety of everyday tasks.
When will the machines take over?
I don't know. Ask Google.