Google develops neural Turing machine learning computer
Neural Turing machine able to learn from past experiences
Google is getting into the business of computers that can learn after recently buying a secretive start-up called DeepMind that specializes in creating computers that mimic the human brain.
Matt Stambaugh, the Calgary Eyeopener's science and technology columnist, says the neural Turing machine is a computer that can program itself.
"It's really a very advanced kind of artificial intelligence," said Stambaugh. "A computer that if it hits a challenge that it doesn't have predefined code for it simply writes the code it needs to solve the problem."
Stambaugh said that in the past computers tended to be either fast or smart but this computer manages to be both.
Traditionally computers have only been able to do what humans program them to do but now they can attempt to solve a problem just like a human brain.
"It's writing more flexible code to say, 'OK go back into your memory banks, just like a human, go back to your experiences and see if you can find similar patterns or similar ways to solve a similar problem and if so start experimenting. Do some trial and error until you get what you think is a correct result' and when it gets that correct result, just like the neurons in our brain, it reinforces those electronic pathways and so in the future it knows this was a successful learning."
Stambaugh said that the technology still needs a lot of development but has lots of possible applications.
"Things like self-driving cars, or any other application computers, have to interface with humans because we are fairly unpredictable," he said. "We can do unpredictable things."
"How much code would you need to deal with every single possible situation on earth?" said Stambaugh. "It may be better to build a computer that can think for itself to deal with situations as they occur."