The US Government’s Role in Advanced AI Development: Predictions and Scenarios

This report was written by Bill Anderson-Samways and Oscar Delaney.

There has been significant recent speculation about whether the US government will lead a future project to build and acquire advanced AI, or continue to play a more arms-length role. Such speculation warrants rigorous assessment, given its implications for research priorities and geopolitical dynamics.

We conducted a forecasting workshop on this question, employing the IDEA protocol (“Investigate, Discuss, Estimate, Aggregate”) to elicit predictions from six professional forecasters and five experts on US AI policy. This included presenting participants with reference-class data addressing whether past analogous innovations (e.g., the computer, the atomic bomb, and historical AI breakthroughs) were developed via US government-led or purely private projects.

• Workshop participants predicted, on average, a 34% median probability that a US government-led project builds and acquires the first AIR-10, conditional on AIR-10 first being developed in the US by 2035. The average prediction among forecasters was 28%, compared to 40% for subject-matter experts.
  

• There was considerable uncertainty in these predictions, with participants expressing an average 90% confidence interval of 11-61%. The most likely scenario for a US government-led project building and acquiring AIR-10 was thought to be a government-led consortium (14%), followed by a single private contractor (9%), a government lab (4%), a nationalized private company (4%), and legal compulsion of a private company (3%).
  

• Conditional on a US government-led project building and acquiring AIR-10, participants estimated a 46% probability of military or intelligence community control over the project (versus 54% for a civilian department such as the Department of Energy).