arXiv:2606.00970v1 Announce Type: cross Abstract: We study risk-neutral control in Markov decision processes with an absorbing catastrophic state. Even though rewards are linear and the agent has no utility curvature, probability weighting, or framing dependence, standard Bellman optimality produces three prospect-theory-like signatures: an S-shaped value-function profile (convex near catastrophe, concave in the far field), an endogenous loss-sensitivity coefficient $\lambda^*(S) > 1$, and a reflection-effect policy reversal. Across 495 configurations, the optimal policy plays safe near catast
This research provides a theoretical foundation for understanding risk-averse behavior in AI systems, especially relevant as autonomous AI agents move into high-stakes environments.
This sheds light on how AI, even without explicit human-like biases, can endogenously develop risk-averse strategies consistent with human prospect theory when confronting catastrophic outcomes.
Our understanding of AI decision-making expands to include intrinsically generated risk-averse characteristics, moving beyond simple utility maximization in complex environments.
- · AI safety researchers
- · Developers of autonomous AI agents
- · Insurance and risk management sectors
- · Simple rational choice models for AI
AI systems deployed in critical infrastructure or financial markets will exhibit predictable risk-averse behaviors when facing potential catastrophes.
Designing AI with controlled 'prospect theory' like behaviors could become a new frontier in trustworthy AI development, leading to more robust and socially acceptable autonomous systems.
This could influence ethical frameworks for AI, mandating that autonomous agents demonstrate an inherent aversion to 'catastrophic' outcomes, potentially limiting their scope in certain fields.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG