OpenAI has unveiled Superalignment, a new programme to make new scientific and technological advances that would allow the direction and control of artificial superintelligence (ASI) systems within four years.

ASI, a possible advanced AI, has the potential to tremendously benefit mankind, but it also offers concerns if not aligned with human ideals. OpenAI is aware of these risks and pushes for preventative actions. As a result, Ilya Sutskever, co-founder and chief scientist of OpenAI, and Jan Leike, head of alignment at OpenAI, are headlining the Superalignment effort. To assure ASI’s alignment with human ideals, both will bring together renowned academics and engineers from OpenAI and other universities.

The major aims of the Superalignment team are to improve AI systems’ ability to supervise and evaluate the behavior of other AI systems, to ensure they can manage unexpected inputs and situations, and to create approaches to comprehend how AI systems function. This will help in detecting and correcting any flaws or misalignment before they cause difficulties. The team also believes that the main technological hurdles of ASI alignment can be solved in four years. This is a lofty objective, but the team is convinced that it can be met with a focused, collaborative effort.

Aside from the Superalignment programme, OpenAI is also focusing on the safety of current AI models, as well as understanding and mitigating other AI dangers.

The sources for this piece include an article in OpenAI.