Pixel perfect

Dr Stephen Gould. Photo by Belinda Pratten

Dr Stephen Gould. Photo by Belinda Pratten

SARINA TALIP takes a journey into the not so distant future, where artificial intelligence makes life less of a chore and computers can recognise objects.

A car drives along a road in downtown Nevada. It slows down for the pedestrians who run out in front of it, trying to make their bus on the other side of the road. The car turns a corner and avoids roadworks in the middle of the street. It stops at a red traffic light.

When the light goes green, the car takes off again, keeping a safe distance behind the truck in front of it. So far, so normal. Except that there is no-one in the driver’s seat.

Welcome to the world of artificial intelligence (AI), where virtually anything is possible. Helping to make it possible is Stephen Gould from the Research School of Information Sciences in the ANU College of Engineering and Computer Science.

Gould’s research focuses on enabling computers to recognise videos and photos, while also programming the machines to teach themselves. He teaches computers what different objects look like and to recognise the same objects later on.

For Gould, the ultimate goal would be to take a photograph of any scene with a camera, and for a computer to inherently understand what it is.

"At the moment, when a computer or robot looks at an image, it’s got no idea what it is - it just looks like a bunch of pixels. All it knows is that there’s green in that corner and blue over there, but it doesn’t know what it is or even that the pixels are related to each other," says Gould.

"An image is just this array of colours and from that array of colours the computer has to understand which things get grouped together and what the colours actually mean. So if it’s green, it’s probably grass or a tree. If it’s blue and it’s high up in the image, it’s probably sky, but if it’s low down in the image, then it’s probably water."

He says the role of artificial intelligence is not to create machines that are more ’intelligent’ than humans, but to create machines that help humans in their day-to-day lives.

"We like to talk about AI and the fact that these machines are somehow intelligent, but what ends up happening is these techniques that we develop end up just becoming technology," says Gould.

"There’s no intelligence behind voice recognition. It’s just a program that’s running on your phone. A machine doesn’t necessarily need to be humanlike or intelligent, just so long as it does the right thing.

"With a driverless car, as long as it’s behaving rationally and intelligently - it’s not hitting other cars or pedestrians - then that’s enough for me to call it AI, and it’s a very useful thing. It’s not replacing us. It’s just replacing the tasks that we do.

"I think what will happen with AI is we’ll have more general appliances. So instead of a dishwasher just washing dishes, it will also clean the table, and a washing machine may go to our laundry basket and separate our laundry into colours and whites."

So it seems the future more closely resembles The Jetsons’ robot maids and flying cars than Hal outsmarting the humans in 2001: A Space Odyssey. Not so futuristic after all.




This site uses cookies and analysis tools to improve the usability of the site. More information. |