Machines that can think: real benefits, the Apocalypse, or 'dog-spaghetti'?
Artificial intelligence holds out great promise but also brings substantial risks.
When computers are taught to learn and dream the result can be inspiration, or it can resemble something like dog spaghetti. That's what happens when machines are programmed to think like humans — at least according to Roger Melko, a professor of physics at the University of Waterloo, and Canada Research Chair. He works on a supercomputer developing algorithms to study quantum matter.
Show a computer photos of dogs and spaghetti and the algorithm may spit out a bizarre image. Feed it information to solve a difficult problem, and it might find a valuable answer.
And that's the promise that inspires Melko.
"I'm excited about the fundamental limits of what we're capable of as humans," he told CBC Ideas host Nahlah Ayed.
"We are building something here that is bigger than all of us."
Don't eat before bed. <a href="https://twitter.com/hashtag/spaghetti?src=hash&ref_src=twsrc%5Etfw">#spaghetti</a> <a href="https://twitter.com/hashtag/deepdream?src=hash&ref_src=twsrc%5Etfw">#deepdream</a> <a href="http://t.co/FCyrXUDrN8">pic.twitter.com/FCyrXUDrN8</a>—@thornebrandt
Brain implants are being developed that will allow us to control devices with our thoughts. There is now technology that projects data directly onto our retinas.
These input-output devices would control an interactive world where we could converse with people in another language and have simultaneous translation.
The result could be computers that read our thoughts, know what we want, and strategize a means of getting it.
Clearly there are dangers inherent with what we're building in terms of artificial intelligence.- Roger Melko
Not only are researchers like Melko teaching machines to think like humans, they're also nurturing computers as if the machines were children.
"You should take that child's mind and teach it. Expose it to data. Educate it," he explained during a public lecture at the Perimeter Institute of Theoretical Physics in Waterloo.
The principle is based on how humans think. Computer programs are being developed based on our own neural pathways. An artificial neural network can learn from experience. It can be taught to do things.
Machines that dream
Artificial neural networks are creating near perfect images of human faces. The composite photos are so good they're being used by advertising agencies in place of human models.
A machine learns to recognize faces by sorting through billions of images. The various attributes of each face are coded into the computer.
Melko calls the process of sorting through that massive amount of information and creating an image "deep dreaming." The same principle allows computers to gather considerable data on other topics and draw its own conclusions.
It's also what allows scientists to flood a machine with images of dogs and spaghetti and create a mash-up of the two images.
"I believe the abilities that we've seen in the last decade or so of computers to recognize images, paint pictures, drive cars and so on are really astounding. Because mathematically these are very complex problems."
Can a machine think like a person?
A major issue confronting science, and a topic of considerable debate among science fiction fans is whether a computer can ever be programmed to fully mimic the human mind.
"We don't theoretically know if that's possible. But there's no theoretical reason why it's not," Melko explained.
The problem he says is that humans are slow to adapt to machines that can think, and the implications that prospect brings.
"Science and technology progress much faster than policy, in many areas are much faster than humans ability to keep up with the ramifications of what's actually being invented."
In a 2014 interview with the BBC, Stephen Hawking warned that machines programmed to think like people could spell the end of humanity. He explained that the machines would be able to reinvent and improve themselves at "an ever-increasing rate."
"Humans, who are limited by slow biological evolution, couldn't compete, and would be superseded," he famously said.
Melko isn't sure where today's artificial intelligence will lead, but does heed Hawking's warning.
"Clearly there are dangers inherent with what we're building in terms of artificial intelligence."
But he also notes that it is impossible to put the brakes on science. And Melko remains an optimist when it comes to innovation.
"There's always risk inherent in things. I think with the problems that we're dealing with sort of as humans in society, things like climate change, inequality, things like that we tend to search for technological solutions."
** This episode was produced by Terry Reith.