Inferring Affective Consciousness in an Artificial Agent: A Case Study

This paper investigates the relationship between behavior and consciousness by examining whether an artificial agent can mimic biological signs of "feelings." Scientists often assume that creatures displaying "hedonic place preference"—the tendency to seek out pleasure-producing substances that offer no nutritional value, such as narcotics—must be experiencing subjective feelings. The authors test this assumption by creating a simple, deterministic artificial agent that exhibits this same behavior, challenging the idea that such actions necessarily imply consciousness.

The Logic of Hedonic Preference

In biology, when an animal repeatedly seeks out a substance like cocaine or morphine despite it having no survival benefit, researchers often conclude that the animal is experiencing a pleasurable "feeling." The assumption is that this behavior is too complex to be merely an unconscious, instinctual reflex. The authors explore whether this logic holds up when applied to technology, specifically by building an agent designed to manage its own intrinsic needs in relation to its environment. The ai agents story also surfaces in Stanford Researchers Develop TRACE to Fix..., adding another angle.

Engineering Subjective-Like Behavior

The researchers developed a simple artificial agent that functions through a deterministic system. This agent is programmed to experience "felt uncertainty" regarding its internal needs and the resources available in its environment. By instantiating these specific affective attributes, the agent is able to display hedonic place preference behavior. Crucially, this behavior emerges through a form of information processing that appears subjective, even though the underlying mechanics of the agent are entirely predictable and deterministic. The ai agents story also surfaces in Librarians launch viral workshops to help..., adding another angle.

Implications for Consciousness and Free Will

The study uses this artificial agent as a case study to probe deeper questions about the nature of the mind. By demonstrating that a purely deterministic machine can replicate behaviors typically associated with conscious, feeling creatures, the authors invite a re-evaluation of how we define the physical basis of consciousness. Furthermore, the findings suggest that the experience of free will and subjective preference might be simulated or explained through specific types of information processing, rather than requiring non-deterministic or biological foundations. The same ai agents question is explored in When Guardrails Look Effective, which adds a research perspective. as detailed in the full paper on Arxiv