A contest of artificially intelligent wits
Researchers are pitting an AI program against itself to judge its intentions.
Why do this? Nonprofit has previously shown that AI systems that train themselves can develop unexpected and unwanted habits. If it is doing something impossibly complex, supervising the training process might not be feasible.
The solution: Having two systems in natural language while a person observes lets humans understand the AI program’s reasoning behind making certain choices.
Progress so far: Two AIs arguing with one another requires more sophisticated technology than exists currently. However, OpenAI has created demos using extremely simple examples to show where the technology could go as its capabilities expand .
Why do this? Nonprofit has previously shown that AI systems that train themselves can develop unexpected and unwanted habits. If it is doing something impossibly complex, supervising the training process might not be feasible.
The solution: Having two systems in natural language while a person observes lets humans understand the AI program’s reasoning behind making certain choices.
Progress so far: Two AIs arguing with one another requires more sophisticated technology than exists currently. However, OpenAI has created demos using extremely simple examples to show where the technology could go as its capabilities expand .
No comments:
Post a Comment