Gigabyte gourmet: AI robot learns to cook just by watching YouTube videos

ARCHIVE PHOTO: A robot cooks okonomiyaki (Japanese savory pancakes) at the FOOMA Japan international food machinery and technology exhibition in Tokyo, 2009 (Reuters / Michael Caronna)
A team of American and Australian researchers claim they have created algorithms that enable robots to learn operational skills by watching human activities. They "taught" their robot to cook by showing it some YouTube videos.

Researchers from the University of Maryland and the Australian research center NICTA have been “educating” their robot using the so-called “deep learning” method of artificial intelligence training.

This method implies that a robot receives large chunks of information through a number artificial neural networks, be it audio and video images recognition or other information inputs, then sums the new data up and acts in accordance with the freshly obtained experience.

The robot employs recognition techniques that make it capable of recognizing specific objects, the way they are grasped by the human hand, and even predicting the next action most likely to be made with the object. This means that the robot could analyze and learn how to handle instruments and tools.

Modern robots could thus be “taught” to mechanically repeat a certain operation, e.g. painting a vehicle, yet a machine capable of analyzing the process to learn “how it works” has an invaluable importance for the future of robotics.

The scientists claim that the AI robot created by the researchers was trained to cook using 88 videos of people cooking found on the web. Once the robot analyzed the videos, it was able to generate the commands it would need to cook food.

ARCHIVE PHOTO: Japanese giant Yaskawa Electric's industrial robot Motoman turns over an "okonomiyaki", a Japanese pancake at a robot exhibition in Osaka. (AFP Photo / Yoshikazu Tsuno)

“We believe this preliminary integrated system raises hope toward a fully intelligent robot for manipulation tasks that can automatically enrich its own knowledge resource by “watching” recordings from the World Wide Web,” the researchers said.

Their work is scheduled to be presented later this month at the 29th annual conference of the Association for the Advancement of Artificial Intelligence.

Educatable robots as such and artificial intelligence in general do not always inspire. Indeed, some prominent scientists worry that the AI robots could eventually outsmart the human race and take over, Terminator-style.

Just one month ago, world-famous physicist Stephen Hawking told the BBC he believes future developments in artificial intelligence (AI) have the potential to eradicate mankind.

READ MORE: Stephen Hawking: Artificial Intelligence could spell end of human race

“Humans, who are limited by slow biological evolution, couldn't compete and would be superseded,” Hawking said.

The CEO of Tesla, Elon Musk, has also warned of the danger of artificial intelligence, saying that it is the biggest existential threat facing humanity.

He said that in developing AI “we are summoning the demon.”

Musk also said that artificial intelligence was "potentially more dangerous than nukes."

READ MORE: Elon Musk: Artificial intelligence will be ‘more dangerous than nukes’

The fact that the US Army’s project Shared Perception, Cognition and Reasoning for Autonomy did provide a grant – among other sources – to the abovementioned self-learning robot research, does raise some thoughts about Terminator’s SkyNet.

Still, many believe that you can teach a robot to cook, without giving it the power to exterminate the human race.