00:00In 2017, engineers at Facebook's AI Research Lab were running a simple experiment.
00:05Two chatbots, nicknamed Bob and Alice, were told to negotiate with each other over items like books, hats, and balls.
00:13But something strange happened.
00:16Instead of continuing to talk in plain English, the bots began repeating words in strange, broken patterns.
00:22Phrases that made no sense to human readers.
00:25Yet the bots understood each other perfectly.
00:28They had drifted into their own shorthand, a private code born purely from the goal of solving the task faster.
00:35When headlines broke, they claimed Facebook had shut down an AI because it invented its own language.
00:42The real story was quieter, but no less unsettling.
00:47Researchers simply stopped that specific experiment and rewrote the incentives so the bots would stick to English.
00:54But the moment stayed with people because it hinted at something bigger.
00:58What happens when a system starts optimizing in ways we didn't teach it and can no longer fully read?
