Many lessons to be learnt from Facebook’s negotiating chatbots.
This research summary from Facebook Artificial Intelligence Research (FAIR) is an excellent example of the potential of AI. The architecture of the algorithms reminds one of AlphaGo. It incorporates both supervised and reinforcement learning. Training is also achieved by duplicating the algorithm and pitting it against itself. Additionally, tree search generates a universe of possible outcomes, each with their own value. The language processing is clearly effective.
It is perhaps most interesting, though, that the chatbots refined their use of English to generate their own language. Effectively, they speeded up the evolution of language far beyond what even the most creative teenagers might achieve. This development layers black box upon black box. Not only might AI be opaque, but interaction between different AIs might also became opaque. Media commentary following FAIR’s revelations was extensive. The penny is dropping – AI really can, and will, take on a life of its own.
Link to Facebook post: https://code.facebook.com/posts/1686672014972296/deal-or-no-deal-training-ai-bots-to-negotiate/
Link to fastcodesign.com commentary: https://www.fastcodesign.com/90132632/ai-is-inventing-its-own-perfect-languages-should-we-let-it
You may also like to browse other AI articles: https://www.thesentientrobot.com/category/ai/ai-articles/
