AI Research Fragmentation Increases Aggregate Risk Despite Potential Individual Lab Share Decreases
Fragmentation in AI research, characterized by an increasing number of competing labs, leads to a rise in aggregate risk, even as individual labs' research shares might decrease. This phenomenon occurs because each additional frontier in AI development contributes to overall hazard. While more labs can raise each lab's racing hurdle, potentially causing research shares to fall, the cumulative risk across the industry still escalates. For instance, in calibrated scenarios with multiple labs, aggregate risk consistently increases. This dynamic is influenced by factors such as the contest for commercial leadership and the hazard externality, where each lab benefits from speed but bears only a portion of the associated risk. The White House Accord on Super Intelligence, signed by major AI and chip companies, focuses on safety audits but does not limit development speed, thus not slowing the competitive race among labs.
