Reports on OpenAI’s development of potentially humanity-threatening AI have led to widespread speculation, but further investigation uncovers a different story.
Reuters and The Information were among the first to report that OpenAI staff members had raised concerns about an internal research project called “Q*,” highlighting its “prowess” and “potential danger.” While this project could solve certain math problems at a basic level, there is debate as to whether the board even received the letter. Furthermore, experts are skeptical about the significance and novelty of “Q*,” suggesting it may not be as groundbreaking or threatening as initially portrayed.
AI researchers, including Yann LeCun, Meta’s chief AI scientist, questioned the novelty of “Q*,” pointing to an old lecture by OpenAI co-founder John Schulman, where a similar mathematical function was described. They believe that the “Q” in the name refers to “Q-learning,” an established AI technique, and the asterisk likely alludes to the A* algorithm. Both of these techniques have been used in AI research for years, undermining the notion of “Q*” as a revolutionary breakthrough.
Furthermore, historical examples of Q-learning and A* applications in AI, such as Google DeepMind’s algorithm for playing Atari 2600 games and the academic origins of the A* algorithm, raise doubts about the groundbreaking nature of “Q*.” Nathan Lambert of the Allen Institute for AI and Mark Riedl, a computer science professor at Georgia Tech, both dismiss the notion that “Q*” represents a leap toward artificial general intelligence (AGI). They argue that OpenAI’s work is not unique and is in line with existing trends in AI research being pursued by numerous other organizations.
While the specific details of “Q*” are unclear, there are suggestions that it could enhance the capabilities of language models, as outlined in a paper by OpenAI researchers. This approach could improve the logical reasoning of language models, promising better alignment and more efficient information consumption. However, the consensus among experts is that “Q*” and its associated mathematical equations are not indicative of an imminent threat to humanity.