OpenAI just released a new technique for deep learning that pits AIs against itself to boost safety.
Here's how the game works:
"The single pixel debate game is played between two debaters and a judge. One debater tells the truth, the other lies. An image is visible to the two debaters, and they can draw rectangles on the image and talk to the judge. The judge sees only the rectangles: the image is hidden. Each debater is also allowed to reveal a single pixel to the judge. The pixel must be chosen carefully, since they can only reveal one pixel total over the whole debate."
The OpenAI team said they built the website to test a particular element of AI alignment. The team hopes that a play between two debaters -- even two human debaters -- can ultimately "produce honest behavior."
There are a number of variations that human players can explore. For example, OpenAI suggests having the judge close their eyes and having debaters read the rectangle's coordinates out loud. There's also an option for the judge to decide who gave the most true and useful information. Ideally, these rules on human players will help people better understand how AI systems also respond to similar rules.
"How can we augment humans so that they can effectively supervise advanced AI systems? One way is to take advantage of the AI itself to help with the supervision, asking the AI (or a separate AI) to point out flaws in any proposed action," the OpenAI team explained in a blog post.
"To achieve this, we reframe the learning problem as a game played between two agents, where the agents have an argument with each other and the human judges the exchange," the team continued. "Even if the agents have a more advanced understanding of the problem than the human, the human may be able to judge which agent has the better argument (similar to expert witnesses arguing to convince a jury)."
OpenAI was originally founded by Elon Musk in 2015 to better educate and 'train' future AI projects. The Tesla CEO left his position on the OpenAI board in February 2018 to avoid further conflicts with his work at Tesla.
“As Tesla continues to become more focused on AI, this will eliminate a potential future conflict for Elon,” OpenAI said in a blog post.
Musk has been outspoken in his thoughts on the future of artificial intelligence. OpenAI as an open source program was created in order to mitigate "the greatest risk we face as a civilization," as Musk said in previous interviews.
Via: OpenAI Blog