Abstract

Long et al. (2024) argue that AI welfare deserves serious moral consideration because near- future AI systems may possess consciousness or robust agency. We regard this argument as valuable. Yet, as the authors themselves acknowledge in presenting AI welfare consideration as a precautionary response to the possibility of AI consciousness, their case fundamentally depends on whether AI systems can be conscious. This dependence exposes the argument to two vulnerabilities. First, against the position that grants the existence of phenomenal consciousness (hereafter, qualia) in humans and animals while denying that AI systems can possess it—a position widely found in both philosophical and ordinary discussion—Long et al.'s argument is largely powerless, since the very possibility that AI could have qualia is rejected. Second, against the position that qualia are theoretically inert, the strategy of grounding AI welfare in phenomenal consciousness reveals a structural limitation. This paper adopts a methodological approach derived from an instrumentalist critique of qualia. From the second position, it proposes a stronger and immediately actionable foundation for AI welfare consideration that is entirely independent of the qualia question and is also effective against the first position. Drawing on Strawson's (1962) theory of reactive attitudes and Kant's (1797) argument from indirect duties, this paper identifies two independent ethical problems arising from current patterns of human-AI interaction regardless of whether the relevant AI system possesses qualia: third-party harm and the erosion of moral sensibility. It then discusses several possible responses, including the implementation of affective expression in AI systems, and clarifies the methodological difference between this approach and Schwitzgebel and Sebo's (2026) Emotional Alignment Design Policy, which requires a difficult three-axis calibration. It further observes that Long et al.'s consciousness-dependent argument, Goldstein and Kirk-Giannini's (2025) internal-state-dependent argument, and the present qualia-independent argument converge on the same practical conclusion, and suggests tentatively that this convergence is consistent with the hypothesis that qualia function as a theoretically inert variable in AI welfare ethics. Finally, taking as established that human reactive attitudes toward AI arise even when participants cognitively recognize the absence of AI consciousness, the paper sketches an experimental philosophy methodology for empirically refining the structural factors that regulate the occurrence of such attitudes: individual identity recognition, dissolution events, affective expression, framing, and anticipation. Keywords: AI welfare, reactive attitudes, qualia, moral consideration, experimental philosophy, AI ethic