OpenAI Forming A Team To Control The “Superintelligent” AI
OpenAI has begun to form a team that will develop ways to manage and control “superintelligent” artificial intelligence systems. The team will be led by Ilya Sutskever, chief research officer and one of the company’s co-founders.
According to OpenAI, today there are no solutions to effectively control and manage superintelligent AI and prevent “aggression” on its part. Existing artificial intelligence tuning technologies are based on reinforcement learning based on human feedback. At the same time, such techniques rely on the fact that people, in principle, can control AI, which is unlikely with systems that are potentially more intelligent than people. To study this issue, a special team will have access to 20% of the company’s computing power. They will be joined by scientists and programmers from the company’s AI customization department, as well as other divisions, in the next four years, the team intends to master the key technical challenges associated with the control of superintelligent AI.
In fact, we are talking about AI training using feedback. This will use special artificial intelligence designed to evaluate other AI systems and ensure the desired results, as well as ensuring the safety of their operation. OpenAI believes that artificial intelligence will cope with this better and faster than humans. As the AI system matures, companies are expected to take on more and more tasks, and as a result, they will create and implement better customization technologies than are currently available. They will work with the people to make their own successors more responsive to the needs of the people, who will only supervise, not directly involved in research.
