An article posted recently by Digital Journal details how researchers at Facebook had to shut down a project after fears that they could lose control of their AI.
It sounds like something out of a science-fiction novel, but researchers noticed that two AI bots were communicating with each other in seemingly nonsensical sentences, when in actuality they had developed their own shorthand that was more efficient than communicating in English. They promptly ended the project due to fears that they could soon lose control of the project.
In another case at Google, researchers were amazed when implementing a neural network to make translations through Google Translate more effective and accurate. They then realized that the network was quickly learning to translate between languages that it hadn’t been taught to translate between. Apparently, the neural network had developed it’s own language suitable for translating other languages.
The concern being that if neural networks develop their own languages, it’ll be harder to monitor what they are doing, and research is still inconclusive if this could pose a threat to AI going rogue.
Full Article: Researchers shut down AI that invented its own language