Can we assess phenomenal consciousness in artificial systems?
Abstract: In light of the rapid technological progress, several philosophers and scientists discuss the possibilities of creating and assessing phenomenally conscious AI, building on Putnam’s conjecture that computational functionalism is more plausible than a biological view of consciousness (Bayne et al. 2024, Birch 2025, Block 2025, Chalmers 2023, Dung 2025, Schneider et al. 2025). Butlin et al. (2025) outline a research program for distilling computational indicators of consciousness via neuroscientific theories of consciousness, based on experimental research on the neural correlates of consciousness in humans.
In this talk, I will distinguish three related but different questions in this context, argue that Chalmers’ epiphenomenal conception of phenomenal consciousness is unfit to serve as a target in this debate, and then evaluate whether computationalism and functionalism should be assumed to be empirically more plausible than a biological theory of consciousness. While computational functionalism allows for artificial consciousness in principle, it is questionable whether artificial conscious system are practically possible, and whether we are in an epistemic position to judge whether a specific AI is conscious. For this purpose, I am connecting debates about AI consciousness with debates about the required medium independence of neural computation (Piccinini 2020, Maley 2025, Williams 2025) and on multiple realizability (Chirimuuta 2022, 2025, Cao 2022, Seth 2025). I argue that in the case of non-biological artificial systems, all possible markers of consciousness typically used to assess consciousness in humans and non-human animals are either absent, ambiguous or question begging.
Passcode: 917746