How to discover the natural kind of consciousness under population shifts
PsyArXiv April 18, 2026 preprint DOI: 10.31234/osf.io/hxcyf_v1 (opens in new tab)
Study at a glance
AI-extracted from the abstract| Characteristics | Theoretical or philosophical paper |
|---|---|
| Key points | Argues that framing artificial consciousness debates as verdicts obscures a deeper epistemic problem: consciousness science's human-based evidence may not extend to radically different systems, so the central question becomes discovery rather than detection. Proposes that proximity relations, understood as multidimensional similarities and differences, supply the missing structure for constraining evidential updates within a Bayesian latent variable discovery model, yielding a proximity-guided INK strategy. |
Abstract
Debates about artificial consciousness are often framed in terms of verdicts: whether a given system is or is not conscious, and which criteria or theories can deliver such judgments. This paper argues that this framing obscures a more fundamental epistemic challenge.Consciousness science has largely been developed within a narrow population – humans – under background assumptions that support evidential generalization within that domain but do not automatically extend to nonhuman animals or artificial systems. When our line of research expands to assess new populations that differ radically in embodiment, internal organization, and learning dynamics, the central problem should be not only detection but also discovery: whether there even exists a projectible cluster of properties that can support explanation and generalization across heterogeneous systems. Building on the iterative natural-kind (INK) strategy proposed by Bayne and Shea (2020) (see also (Bayne et al., 2024a, 2024b)), here I lay out a roadmap for a proximity-guided extension of this approach. The INK strategy treats consciousness as a candidate natural kind whosestructure is to be discovered through iterative revision of indicators, mappings, and theoretical commitments. However, in its abstract form, INK leaves open how evidential updates should be constrained under population shift. The present paper argues that proximity relations – to be understood as multidimensional similarities and differences between systems (Peters, n.d.) – provide the missing structure. By conditioning updates on explicit judgments about proximity within a Bayesian latent variable discovery model, a proximity-guided INK strategy renders several potential limitations of the INK strategy as visible, interrogable features rather than hidden liabilities. The result is a methodological roadmap for investigating consciousness under uncertainty, focused not on delivering verdicts but on clarifying what kinds of discovery about the natural kind of “consciousness” may be possible.