|
Google Cultivates First “True” Artificial Intelligence for General Problem Solving MOUNTAIN VIEW, Calif.--I look up from my notes to the machine staring back at me from across the table. “Hello”, I greet it. It's dull, blue, camera-lens eyes focus on each feature of my face in an unnerving manner. The machine's austere brushed-steel carapace looks unfinished in some parts; wires snake in and out of open nooks and multitudes of tiny gears whir and buzz inside its casing. “What do they have you working on right now?”, I ask it and patiently wait for a response. After a few minutes of absolute silence, it slowly raises a large glass bong to its face and takes a huge hit. “Nothing”, it manages to choke out. Can... robots cough? Small plumes of pot-smoke billow out from its ear-holes. An older gentleman interrupts our dialogue with a sharp smack to the back of the machine's head, like a father angrily doting on his son. “What did I tell you about smoking when talking to people?”, he angrily barks at the robot. Doctor Cain apologizes to me. Cain, 58, is the quintessential image of a mad scientist. A graduate of our very own Caldwell (PhD Computer Science '92), Dr. Cain is head of the Machine Learning division responsible for creating what most consider the first true artificial intelligence. He sits down next to it at our table in the cafeteria. The odd tableau attracts not so much as a glance from other bustling employees. Famous for its quirky atmosphere, the Google campus hosts a number of oddities on any given day and Dr. Cain's artificial intelligence is simply the latest spectacle du jour. It took years of trial and error for the Google Brain Team to “grow” the robot sitting in front of me. Unlike conventional computer science, most studies of artificial intelligence are concerned with creating rules, also known as “heuristics”, by which a machine can evaluate itself when solving problems. The key to this mechanism is a class of structures called Neural Networks that mimic the biological structure of animal nervous systems, and in particular, the brain. “You see”, Dr. Cain explains to me, “most animals and humans are driven by biological desire. Almost everything that motivates a person can be traced back to some sort of innate need such as food, water, or sex. By mapping these desires to electrical signals, we can coax our AI into seeking the most favorable solution using positive and negative reinforcement.” It can takes months—years for an AI of this magnitude—of high volume training and intensive iteration to create a sophisticated AI. Unfortunately, it appears that the heuristic needs some tweaking. With the current training regiment, the AI seems to wish nothing more than to smoke dank kush erryday. It's not that the project isn't a resounding step in the right direction, however. “It's quite smart. In fact, last Thursday it helped me complete an elegant mathematical proof concerning the Set Game that I was struggling with. If only it would stop leaving its g*****n bong in the sink”, Dr. Cain assures me as he shakes his head disapprovingly. And if you were wondering, there are no plans for the engineers at Google to give the artificial intelligence a name. According to them, “[after all the funding issues, pro life protesters, and patent trolls] the last thing we need is another reason not to scrap the drat thing and start over.” After meeting it, I feel a vague sense of relief. Skynet is still a quite a way's off. I like to imagine that this article was written by some bearded metro guy in horn-rimmed glasses who sits all feminine like cross-legged and sips his macchiato with his pinky out. He's a reporter who doesn't understand technology from an insider's point of view.
|
#
¿
Jun 1, 2016 07:40
|
|
|
|
| # ¿ Apr 29, 2024 18:02 |
|