In collaboration with Prof. Roman V. Yampolskiy at the University of Louisville, we proposed the modeling of AI safety problems as psychopathological disorders. This project is inspired by the many similarities of current and emerging AI algorithms with the learning mechanisms of natural cognition, and presents a tractable abstraction for investigating the deleterious behaviors that may arise in the current and future AI systems. The first publication of this proposal has gained widespread attention from the research community and the news media. We are currently developing a theoretical analysis of addictive and post-traumatic behavior that can result in DRL agents.
Current Team Members:
PI: Vahid Behzadan
Affiliate Research Groups:
Tools and Datasets:
Publications:
- Behzadan, V., Yampolskiy, R. V., & Munir, A. (2018). Emergence of Addictive Behaviors in Reinforcement Learning Agents. AAAI Workshop on Artificial Intelligence Safety (SafeAI) 2019. arXiv preprint arXiv:1811.05590.
- Behzadan, V., Munir, A., & Yampolskiy, R. V. (2018, September). A psychopathological approach to safety engineering in AI and AGI. In International Conference on Computer Safety, Reliability, and Security (pp. 513-520). Springer, Cham.