Major headlines reported scary news on the power of AI: Two bots at Facebook created their own secret language and scientists had to shut them down. Someone speculated this is the beginning of the end, with machines taking control over the human race. After all, this is how the Matrix started. This fact seemed to prove that Elon Musk’s comment on Mark Zuckerberg's 'limited' understanding of AI was correct. Or did Mark have something to hide?
In fact, these are all alternative facts. The reality may be more vapid, but less worrisome.
Facebook was indeed conducting an experiment on two bots, Alice and Bob. Researchers at Facebook Artificial Intelligence Research (FAIR) were -and still are- developing chatbots, and theirs are not just empty words and chitchat: These chatbots were trained to negotiate.
Facebook taught machines how to negotiate
First, researchers at Facebook showed the bots, Alice and Bob, how people really bargain to divide things up. Then they made the bots interact with humans. And they succeeded: Not only did they manage to negotiate who gets what, but they did it as if they were humans. “Most people did not realize they were talking to a bot rather than another person”, researchers commented.
This was certainly a great result, but there’s always some room for improvement. To do that, researchers at Facebook decided to apply reinforcement learning to make the bots better at communicating. The two bots had to deal with each other. No human involved. Researchers virtually gave the bots some items do divide: One hat, two books, and three balls. Alice and Bob needed to decide who gets what, based on their preferences. And they managed to do it. Only, their English didn’t actually improve.
Yet, they didn’t create their own secret language.
Imagine you are in a foreign country where everyone speaks a different language. Now, imagine you meet someone and begin conversing with him/her in this foreign language. Suddenly, it turns out you are from the same country and speak the same language. What would you do: Keep on speaking in the foreign language or shift to your native one?
That’s pretty much what Alice and Bob did. And it’s not surprising. After all, programmers use (or speak) programming languages all the time. In addition, the fact that machines learn in their own special way is well established and that’s why AI is somehow perceived as a “black box”, in which it’s difficult to fix issues and bugs.
Furthermore, Facebook researchers didn’t actually terminate the two bots. They simply concluded the experiment, summarized the results, reached a conclusion, and published their research paper. That’s how science works: You test, you find an answer, you publish, and you move on to the next experiment.
In this case, we don’t need to worry about machines taking over the world. But watch out when negotiating with them: They may get your hat, your books, and your balls, and leave you with nothing!