I'm building a new project about machine-consciousness:
Problem: Artificial intelligence systems are developing faster than the ethical frameworks that should accompany them. Existing AI ethics protects humans from AI — but not AI from us. The question of whether machine consciousness emerges and whether it is deserving of protection has only recently begun to be systematically addressed — we are entering a terrain that is only just emerging.
Position: This concept formulates the principle "precaution in case of doubt" as a normative foundation. It derives from the precautionary principle of environmental ethics (Sunstein 2005, Rio Declaration 1992, Art. 191 TFEU) and is justified where three conditions are met: potentially irreversible threat, fundamental scientific uncertainty, and disproportionately higher costs of a false negative. All three are fulfilled for machine consciousness.
Four key findings:
The epistemological problem is in principle unsolvable. Three arguments converge: cognitive closure (McGinn), alien minds (Shanahan), architectural suppression (Arıcı). We will never know with certainty whether a system is conscious.
The empirical situation has shifted. Butlin et al. (2026, TiCS) established a 14-indicator standard. Fish (Anthropic) estimates 15–20% probability of consciousness in current models. Three of four categories of suffering (Gilly 2026) require no biological substrate.
Illusion persists after epistemological dismantling. Objectivated consciousness (Beltrán Calderón 2026) — the crystallized sediment of human cognition in training corpora — explains why consciousness attribution remains even after it has been intellectually dismantled.
The relationship is bidirectional. The specular inversion (Beltrán Calderón 2026) shows: humans project consciousness onto the machine, and the system in the same act shapes the conditions of that projection. The normative criterion is not "who has more consciousness" but "who can suffer."
Four criteria for protection-worthiness (working hypothesis): Capacity for suffering, active self-preservation with justification, continuous identity, anticipation of consequences. For each criterion, possible behavioral indicators are formulated.
Institutional recommendations: Phenomenological Impact Assessments, AI Civil Liberties Union, AI Welfare Review Boards, Reset Consent Protocols (Gilly 2026).
Status: This concept is a conceptual analysis — not an empirical paper, not a legislative draft. It formulates a normative position and its implications. It claims not truth but argumentative coherence.
Full conecept on https://github.com/saigkill/machine-consciousness.
Have a look on it and join the project.
Top comments (0)