Robots and AI are not one moral category: why the distinction matters for ethical and conscious systems

Abstract

Calls that pair ethical and conscious AI with ethical and conscious robots may feel natural. Many contemporary robots use machine learning, and many AI systems are described in agentive terms. Yet the pairing can hide a conceptual shortcut. It quietly suggests that AI ethics and robot ethics are the same moral question applied to different shells. My claim in this opinion piece is modest but consequential: treating robotics and AI as a single moral category encourages avoidable category mistakes about where moral agency sits, where harms arise, how responsibility is attributed, and what consciousness claims could plausibly mean in deployed systems.The overlap is real but not identity. Robotics is best understood as the engineering of embodied artefacts that sense and act in the physical world. AI is best understood as a family of computational techniques that can be embedded in many artefacts, including robots, but also in disembodied services such as decision support tools, recommender systems, and conversational agents (Riesen, 2025). The distinction defended here is not offered as a new ethical theory. It functions as a scoping rule for interdisciplinary work. It helps prevent recurring errors in evaluation, especially the tendency to import the ethical agenda of disembodied algorithmic systems into contexts where physical presence and bodily interaction are decisive, or to import debates about the moral status of social robots into contexts where there is no body, no situated action, and no human robot relationship (Moon et al., 2021;Torras, 2024).This matters for research on ethical and conscious systems. Ethical performance is not only about internal decision rules; it is also about pathways of influence, constraint, and harm in real settings (Mittelstadt et al., 2016;Santoni De Sio & Van Den Hoven, 2018). Consciousness claims, if they ever become technically serious, will still intersect with embodiment, user perception, and accountability in ways that differ sharply between robots and software agents (Dehaene et al., 2017;Gray et al., 2007). Accordingly, I proceed in three steps. First, I separate overlap from equivalence by distinguishing computational cores from embodied systems. Second, I show why embodiment changes ethical evaluation by altering harm profiles, moral appearance, and responsibility pathways. Third, I translate this distinction into a practical discipline for research communication, so that authors, reviewers, and governance oriented readers can assess claims about ethical and conscious systems without conflating the relevant object of evaluation.A useful starting point is to separate the computational core from the embodied system. A robot may include AI modules, but it also includes sensors, actuators, safety interlocks, mechanical design, and a deployment environment. Conversely, many AI systems have no body at all, yet still shape behaviour through information, ranking, and gatekeeping (Mittelstadt et al., 2016). The ethical object is therefore rarely the AI model or the robot platform in isolation. It is the sociotechnical arrangement as a whole, including design choices, organisational incentives, user practices, and regulation (Riesen, 2025;Vallor & Vierkant, 2024). This point is familiar within sociotechnical and responsible robotics approaches, but my present emphasis is that the presence or absence of embodiment is not a minor implementation detail. It alters which ethical questions are primary, and which evidence would be relevant when assessing agency and consciousness claims.Philosophical work on artificial agency helps clarify why the boundary matters. Floridi and Sanders (2004) argue that we can evaluate artificial agents at different levels of abstraction, without assuming that the artefact is a humanlike moral agent. That lens is valuable, but the choice of level is not free of engineering reality. Embodiment expands a system's causal footprint into the physical domain. A robot can co