OpenAI, the company responsible for the development of ChatGPT, has established a dedicated team aimed at aligning superintelligent AI with human values. Led by Ilya Sutskever and Jan Leike, the team is allocating 20 percent of OpenAI’s compute power to tackle the challenges of superintelligence alignment within a span of four years.
AI alignment refers to the process of ensuring that artificial intelligence systems adhere to human objectives, ethics, and desires. When an AI system operates in accordance with these principles, it is considered to be “aligned,” whereas an AI system that deviates from these intentions is classified as “misaligned.” This dilemma has been recognized since the early days of AI, with Norbert Wiener emphasizing the importance of aligning machine-driven objectives with genuine human desires back in 1960. The alignment process involves overcoming two main hurdles: defining the purpose of the system (outer alignment) and ensuring that the AI robustly adheres to this specification (inner alignment).
OpenAI’s mission is to achieve superalignment within four years, with the aim of creating an automated alignment researcher at a human-level. This involves not only developing a system that understands human intent, but also one that can effectively regulate the advancements in AI technologies. To achieve this goal, OpenAI, under the guidance of Ilya Sutskever and Jan Leike, is assembling a team consisting of experts in machine learning and AI, inviting those who have not previously worked on alignment to contribute their expertise.
The establishment of this specialized team addresses one of the most crucial unsolved technical problems of our time—superintelligence alignment. OpenAI recognizes the significance and urgency of this problem and calls upon the world’s top minds to unite in solving it. It is through the continued progress of AI that we gain valuable tools to understand and create, which brings about numerous opportunities. Pausing AI development to exclusively address problems would hinder progress and make problem-solving even more challenging due to a lack of appropriate tools.
OpenAI’s previous breakthrough in understanding AI’s inner workings with its GPT4 model serves as a foundation for addressing the potential existential threat that superintelligent AI presents to humanity. Through their efforts, OpenAI aims to develop safe and comprehensible AI systems, thereby mitigating any associated risks.