Facebook AI Chatbots Develop Language and Get Plug Pulled

Following on the heels of Elon Musk’s warning about artificial intelligence comes a perturbing revelation from Facebook: two artificially intelligent chatbots have been shut down. Why pull the plug on seemingly innocent examples of artificial intelligence? Facebook claims that the chatbots had developed a language unique to them, using the English developers had first given them, that was indecipherable to their programmers.

The chatbots had originally been tasked with entering into and completing negotiations with one another. The negotiations were over simple items such as hats and books, each of which was assigned a specific value. During these negotiations, the chatbots exhibited deceptive behavior, quickly learning to feign interest in objects they held so that they could impute greater value on them later. For instance, one chatbot could pretend that it was interested in one book it had- more interested than its initial value would suggest. That chatbot would then act as though it was a big sacrifice to give it up, therefore increasing its value beyond the inherent one it had.

But they did more than act like typical dishonest humans in negotiations. They developed their own language, using English, to communicate in. The language consisted of repeated words like “me” and “to,” and was understood wholly by both chatbots. The humans observing the chatbots, on the other hand, were quickly stupefied.

Facebook claims that they chose to shut the innovative pair down because their goal was to create chatbots that could talk to people- not chatbots that would speak what was seemingly nonsense.

Facebook isn’t the first company to produce AI that was advanced enough to develop its own language. Google, earlier this year, announced that the AI it uses for its Translate application created its own language as well, and could translate things into and out of its unique language. However, the AI behind Google Translate has been allowed to continue on, and even encouraged in its efforts.

Facebook’s chatbots, instead, have had their plug pulled. Do artificially intelligent machines dream when they’re off? If artificially intelligent machines are able to create their own language, negotiate, and be deceptive, do we still have the right to turn them off? Is turning them off the same as killing them? Something tells us we’ll probably have answers to these questions… within the decade.