14 DAY TRIAL // As if the fiction about murderous robots and other types of artificial intelligence weren't scary enough on their own, not to mention all the books and comics, researchers from the US and Australia have taken to teaching robots how to use sharp and pointy things.
That is literally what they have been doing. “They” being researchers from the University of Maryland Institute for Advanced Computer Studies (UMIACS) and one scientist at the National Information Communications Technology Research Center of Excellence in Australia (NICTA).
They have developed a robotic system that can teach itself to perform various robotic actions, though so far only one skill has been gained: knife wielding.
You read it right, ladies and gentlemen, the people who created the robot, and who are undoubtedly some of the brightest minds on the planet, decided that wielding knives was an essential skill to have.
Fortunately, the robot didn't learn to stab people or anything. Rather, it watched YouTube videos enough to pick up a rudimentary cooking skill, which naturally involved chopping things.
Why using kitchen knives is a big deal
Using knives to chop things in preparation of, say, an omelet actually requires a fair bit of dexterity, involving every joint and muscle in the arms and hands.
Thus, it was a good starting skill for the robot to gain, one that is more than enough to illustrate how advanced the breakthroughs in artificial intelligence, computer design and natural language processing are.
Those were the three research areas which had to combine in order to successfully create the new robots capable of learning for themselves.
Although, as we've seen, it's not true artificial intelligence we're witnessing, but a form of machine learning that is just a step towards that ultimate goal.
The key breakthrough was that the robot was able to decide for itself which motions from the ones observed would allow it to accomplish the task efficiently.
In addition to knives, the robot was able to manipulate a glass and pour water in it from a bottle. Other tools should follow soon.
Practical applications
Complex robotics don't really have many yet, and that's not likely to change while the proof of concept machines still look like mechanical arms sticking out of an immobile, central hub. But we're slowly creating the pieces needed to eventually build a truly dexterous and reasonably intelligent robot.
This particular piece will be presented at the Association for the Advancement of Artificial Intelligence Conference in Austin, Texas, on January 29, 2015.