OpenAI is offering $10 million in grants to support technical research on how we can control artificial intelligence systems that are way smarter than the average human. The company, known for its cutting-edge developments in AI, wants to ensure that superintelligent AI does not go rogue or become uncontrollable.
OpenAI’s Superalignment Fast Grants, as they are called, aim to shed light on how strong AI models generalize from weak supervision. The company hopes to identify ways to align and evaluate future superhuman AI systems, as well as build an AI lie detector.
The complexity of superhuman AI systems poses a challenge for humans to fully understand their workings. OpenAI believes that if these AI models generate millions of lines of code, it would be difficult for humans to reliably determine whether the code is safe or dangerous.
In a post on the company’s website, OpenAI stated that ensuring the alignment and safety of future superhuman AI systems is one of the most crucial technical problems in the world. They believe it is solvable and that new researchers can make significant contributions in addressing it.
As AI systems continue to advance, relying solely on human supervision may no longer be sufficient. OpenAI aims to stay ahead of the curve by finding ways to effectively maintain human control over AI systems.
The grants are primarily intended for academic labs, non-profits, and individual researchers. Additionally, OpenAI is sponsoring the OpenAI Superalignment Fellowship, a one-year program offering $150,000 to graduate students interested in exploring the field of superintelligent AI systems.
OpenAI has identified seven practices to keep AI systems safe and accountable. To further research and address open questions related to these practices, the company plans to fund projects through its Superalignment grants. The grants will support researchers investigating the impacts of superintelligent AI systems and developing practices to ensure their safety.
OpenAI refers to superintelligent AI systems as agentic AI systems. These systems are capable of a wide range of actions and can be trusted to autonomously act on complex goals on behalf of users in certain circumstances. The company is particularly interested in understanding how to evaluate the appropriateness of agentic AI systems for specific use cases, when human approval for actions should be required, and how to design intelligent systems that allow us to review their internal reasoning.
While current AI tools are impressive, they have not yet reached the level of superintelligence that OpenAI is anticipating. However, the company’s CEO, Sam Altman, has revealed that they are already working on the GPT-5 model, which he believes could possess hints of superintelligence.
OpenAI predicts that superintelligence could become a reality within the next decade. The results of their research grants will play a crucial role in determining the implications and potential risks associated with superintelligent AI systems.
In conclusion, OpenAI’s commitment to investing in research on controlling superintelligent AI represents a vital step in ensuring the safe and responsible development of AI technology. The grants and fellowships offered by the company aim to gather insights and develop practices that will help humanity maintain control and navigate the complexities of advanced AI systems.
