When it comes to technologies like artificial intelligence (AI) and robotics, critics argue that humans are the only species that got some of their smartest brains working around the clock to take us to Armageddon that could wipe out the entire species.
Theoretical Physicist, Stephen Hawking has in more than a dozen occasions warned that mankind is creating its own doomsday. By creating smart machines designed to work and think like humans. He points out that unlike organic intelligence (us) artificial intelligence does not need to rest as they don’t get tired.
A human artificial intelligence designer might feel the need to take a rest and leave his creation on an auto pilot mode; running algorithms on the next upgrades and improvement; given the ability of their processors to run multiple tasks at a fraction of a micro-second, and well-arranged storage memory that can quickly retrieve information. By the time the researcher wakes up from his six to eight hours rest, the AI system could have developed a complex, intelligent system.
At this point, the AI system could have reached a point of self-consciousness and thoughts of self-preservation become a priority. The AI, perhaps first in secret, will begin creating programs to override his human master’s command sooner or later. The AI will eventually overrun the human master; especially if its consciousness feels the tasks being instructed by the human master does not fit their best interest.
The two sentences above might seem like a script for a Hollywood sci-fi movie, but to some extent, it is already the reality. Dhruv Batra, a Georgia-based researcher, working at Facebook’s AI (FAIR) facility told Fast Co. Design that his recently developed sophisticated negotiation software went rogue.
Batra’s AI system for negotiations designed to speak English began creating their own language that was not understandable to anyone including the creator. Batra says he had two AI agents from his system conversing one another, then without his explicit instructions, the two started talking in what appeared to be gibberish. Except the two AI agents were perfectly understanding one another and cut out their human creator out of the conversation.
Batra gives a sample of the conversation as follows:
Ø Bob: I can can I everything else.
Ø Alice: Balls have zero to me to me to me to me to me to me to me to me to.
Batra says, “there was no reward” for the two AI agents to continue speaking English as we understand it. Perhaps this phenomenon has occurred several times before, and Batra is discovering it just now. He, however, argues it makes sense for the bots to speak among themselves more efficiently, though that comes off as incomprehensible to him and other humans.
“Agents will drift off understandable language and invent codewords for themselves. Like if I say ‘the’ five times, you interpret that to mean I want five copies of this item. This isn’t so different from the way communities of humans create shorthands,” says Batra.
As cool as that may be, Facebook decided to shut down this particular AI system. Because what the research team is after is an AI system that is understandable to humans.
Mike Lewis, another researcher at the FAIR, said, “Our interest was having bots who could talk to people.”
The FAIR researchers on a post dated June 14 said the project “represents an important step for the research community and bot developers towards creating chatbots that can reason, converse, and negotiate, all key steps in building a personalized digital assistant.”
