Using Thermal Comfort to Assess Cognitive and Perceptual Abilities of AI​

Abstract

Thermal comfort—subjective satisfaction with environmental conditions (temperature, humidity, air velocity) and personal factors (clothing insulation, and activity level)—embodies a complex perception‑decision cycle central to daily human adaptation. Thermal‑comfort behaviors are intuitive, context‑dependent, and require integration of multi‑source information, making them an ideal probe of AI’s real‑world cognitive abilities. We propose a novel evaluation framework that uses thermal comfort to assess three core cognitive capacities of AI systems: (1) cross‑modal reasoning (integrating environmental, personal, and contextual cues), (2) causal association (linking variables such as temperature to comfort outcomes), and (3) adaptive decision‑making (modifying behavior under changing conditions). The hypothesis is that AI capable of accurately perceiving multidimensional environmental variables, selecting appropriate clothing insulation, providing context‑aware thermal sensation feedback, and adapting clothing or HVAC settings demonstrates cognitive competence comparable to human intuitive reasoning. Using the Open Access Thermal Comfort Database, we designed three ecologically valid tasks: (i) outdoor temperature‑driven outfit selection (predicting CLO values 0.5–2.0), (ii) indoor thermal sensation prediction with PMV, and (iii) adaptive decision‑making (adjusting clothing or activity level or HVAC for environmental shifts). AI models achieved high accuracy in outfit selection, and ≥50 % alignment with human adaptive actions, especially in moderate climates. These results demonstrate that thermal‑comfort assessment provides an ecologically valid alternative to abstract benchmarks, highlighting AI’s ability to navigate dynamic real‑world contexts. The framework bridges narrow AI testing and practical cognitive evaluation, offering a new paradigm for gauging AI readiness in human‑centric applications such as smart homes, wearables, and building automation. By grounding assessment in a universal human experience, it paves the way for AI systems that intuitively adapt to human needs.

Keywords