Approach

We aim to gain a better insight into the dynamics of races toward transformative AI and potential opportunities for effective altruists (EAs) to increase safety among race participants. Furthermore, we want to take steps toward turning computational AI race modeling into an academic field that has safety at its core. Lastly, we aim to create technical and conceptual foundations that enable future researchers to build computational models with different assumptions about the important features. By making our work possible to build upon we can make future work in this area faster and more comparable.

We believe that computational modeling can reduce the risks of bad outcomes of AI races by:

  1. Assisting in systematically discovering possible scenarios, even ones not found through qualitative reasoning
  2. Allowing researchers to interactively and visually explore different scenarios and their counterfactuals
  3. Predicting possible outcomes of AI races or parts of races
  4. Suggesting actions EAs can take to improve safety during a race
  5. Generating data for transformative AI scenarios that haven't happened yet
  6. Helping to prevent an AI race from happening in the first place
  7. Preparing EAs for how to behave if a race happens
  8. Positioning the AI governance community to be a first mover and important player in a future AI race modeling field and thereby being able to influence an AI race if it happens in the future