Can Silicon Feel? Conditions for Artificial Valence and Their Implications for AI Governance

Abstract

Abstract Recent advances in artificial intelligence have renewed questions about whether artificial systems could eventually possess experiences that matter to their welfare, including pleasure and suffering. This paper asks what features an artificial system would need before it could reasonably become a serious candidate for phenomenal valence. It argues that behavioural sophistication, intelligence, reward signals, and verbal reports are insufficient evidence of valence because these features can be implemented without establishing welfare-relevant experience. The paper proposes four organizational conditions: internal causal integration, stable but changing internal states, valence–action coupling, and substrate-sensitive implementation. Internal regulation and embodiment are treated as cross-cutting features that may support these conditions rather than as independent requirements. The framework is then applied comparatively to contemporary AI architectures, including transformer-based systems, reinforcement-learning agents, agentic systems, neuromorphic architectures, and recurrent embodied systems. The paper does not claim that satisfying the conditions would establish consciousness or valence. Instead, the conditions identify when an artificial system would become a more serious candidate for scientific and ethical investigation. The analysis concludes with a precautionary governance framework focused on monitoring, independent assessment, and welfare-sensitive design as artificial systems become increasingly integrated, persistent, and autonomous.