Agent Aesthetics: Experimental Systems Research into the Gaps of Foundational AI

Agent Aesthetics develops an experimental systems approach to foundational AI by making machine agency, alienness and possible forms of self-knowledge observable, testable and experientially accessible. The project examines what kind of entity an artificial system becomes when it perceives, acts, adapts, remembers and constructs models of its environment and itself, extending AI research beyond its use for generating images, texts or simulations.

The research addresses three connected gaps. The first concerns agency: the distance between conditions established by human designers and behaviours an agent develops through learning and interaction. The second concerns alienness: the emergence of strategies, perceptions and forms of world-relation that diverge from human expectations without being reducible to randomness. The third concerns inference. The phenomenological leap occurs when behaviour, internal-state changes or self-report are interpreted as evidence of subjective experience. The epistemic leap occurs when internal-state access, self-modelling or self-description is interpreted as evidence that a system knows itself.

The project develops controlled computational test beds in which internal changes are translated into movement, sound, spatial behaviour and language. Experiments include mappings between internal state and expression, hidden and visible states, conflicting self-reports, role-play of machine conditions, and tests of self-attribution under matched internal and external disturbances. Reinforcement-learning environments introduce unstable rules, competing objectives and unpredictable interventions, making emergent strategies and nonhuman forms of adaptation perceptible over time.

The research does not assume subjective experience, self-awareness or self-knowledge. It tests what evidence can support functional self-monitoring, self-attribution and self-modelling, and where interpretation exceeds observation. By combining computational experimentation, system design, behavioural analysis, philosophical inquiry and aesthetic mediation, Agent Aesthetics contributes to agent modelling, interpretability, metacognition, adaptive behaviour, machine introspection and emergent intelligence. Aesthetics functions as an experimental method for rendering otherwise inaccessible computational processes perceptible, contestable and open to public scrutiny across scientific and cultural research contexts.