Modeling Pathology-Like Behavioral Patterns in Language Models Through Behavioral Fine-Tuning
arXiv:2605.22356v1 Announce Type: new Abstract: Large language models are increasingly used as computational tools for modeling human-like behavior. We introduce a behavioral induction framework that modifies model policies through fine-tuning on structured decision-making tasks: using synthetic datasets inspired by maladaptive behavioral patterns, including depression and paranoia, we train transformer-based language models to consistently select specific classes of actions across diverse contexts. We then test whether this behavioral optimization produces systematic changes in generative dis
The increasing sophistication and widespread deployment of large language models are driving research into their behavioral modeling capabilities, particularly regarding human-like patterns.
This research demonstrates a method to intentionally induce complex behavioral patterns, including pathologies, in AI, which has significant implications for AI safety, development, and ethical use.
Our understanding of how to systematically 'program' human-like psychological states into AI models changes, moving beyond simple task completion to more nuanced, behaviorally-driven AI design.
- · AI Safety Researchers
- · Psychology-informed AI Developers
- · Developers of AI psychiatric assessment tools
- · AI systems without robust behavioral guardrails
- · Unregulated AI deployment
- · Simplified views of AI behavior
AI models can be engineered to consistently exhibit specific psychological traits or biases.
This capability could lead to more realistic human simulation, but also to AI systems that reliably exhibit harmful maladaptive behaviors if not carefully controlled.
The ability to model and induce pathology-like behaviors in AI could revolutionize therapeutic approaches for humans or generate novel forms of AI-human interaction with unforeseen consequences.
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.CL