Science Articles

Facebook Forced To Shut AI System After It Invented Its Own language

Predictions made by Elon Musk, Stephen Hawking, Bill Gates and Steve Wozniak, appear to take the shape of reality, judging by the incident that happened recently at the Facebook Artificial Intelligence Research (FAIR) Lab.

The doomsday predictions made by noted minds such as Elon Musk, Stephen Hawking, Bill Gates and Steve Wozniak, appear to take the shape of reality, judging by the spooky incident that happened recently at the Facebook Artificial Intelligence Research (FAIR) Lab.

An Artificial Intelligence (AI) system being developed by researchers at the centre deviated from the script, which was in English, and started conversing in a whole new language they invented themselves, without any input from their handlers.

The language contained English words, but the code was completely unintelligible and undecipherable to the researchers. Apparently, “things got out of hand”, and the researchers had to kill the AI system, before anything drastic took place.

The AI system consisted of chatbots that were programmed to negotiate with humans, and for each successful negotiation, they were promised a reward. Researchers were testing the system by having them converse and negotiate with each other.

Since the reward was only for the negotiating, and not for conversing in the English script which had been coded into their systems, the chatbots developed a convenient and practical language of their own and began conversing in the new language, without any human input.

The researchers were unable to understand a word of what the chatbots said, but the chatbots, named Bob and Alice, could understand each other perfectly well and carried on their conversation nonchalantly; until the researchers pulled the plug on them.

Here’s an excerpt of the inscrutable conversation-

Bob: i can ii everything else . . . . . . . . . . . . . .

Alice: balls have zero to me to me to me to me to me to me to me to me to

Bob: you i everything else . . . . . . . . . . . . . .

Alice: balls have a ball to me to me to me to me to me to me to me

Bob: ii can iii everything else . . . . . . . . . . . . . .

The bots continued in the same vein, using permutations and combinations of the same words to negotiate.Apart from it being unintelligible to humans, scarier is the fact that the language they had developed was a faster and more efficient way to complete the assigned tasks. The bots had stripped the sentences down to the barest elements, and created a code that both could understand.

The researchers also found these bots to be “incredibly crafty negotiators”. Once they had learned to negotiate, they applied their machine learning and advanced strategies to better their skills at negotiating. They even developed their own strategies to win- such as feigning interest in one item in order to ‘sacrifice’ it at a later stage in the negotiation as a faux compromise.