Research Priorities for Robust and Beneficial Artificial Intelligence


How to maximize the benefits of AI while avoiding potential pitfalls. This article gives numerous examples of such worthwhile research aimed at ensuring that AI remains robust and beneficial.

Short-term research priorities:

  • Optimizing AI’s Economic Impact
    • Labour market forecasting
    • Identifying market disruptions
    • Policy for managing adverse effects
    • Economic measures
  • Law and Ethics Research
    • Liabilty and law for autonomous vehicles
    • Machine ethics
    • Autonomous weapons
    • Privacy
    • Professional ethics
    • Policy questions:
      • What is the space of policies worth studying, and how might they be enacted?
      • Which criteria should be used to determine the merits of a policy?
  • Computer Sciences Research for Robust AI
    • Verification (Did I build the system right?)
    • Validity (Did I build the right system?)
    • Security (Prevent intentional manipulation)
    • Control (Is it useable? Can I fix it?)

Long-term research priorities:

  • Verification, Validity, Security, Control redux
  • Intelligence explosion

(Their) Conclusion

In summary, success in the quest for artificial intelligence has the potential to bring unprecedented benefits to humanity, and it is therefore worthwhile to research how to maximize these benefits while avoiding potential pitfalls. The research agenda outlined n this paper, and the concerns that motivate it, have been called anti-AI, but we vigorously contest this characterization. It seems self-evident that the growing capabilities of AI are leading to an increased potential for impact on human society. It is the duty of AI researchers to ensure that the future impact is beneficial. We believe that this is possible, and hope that this research agenda provides a helpful step in the right direction.