Why embodiment mattered in the first place: the body as the origin of mind

There is a temptation, when thinking about mind, to imagine it as something that floats free of the body: a pure process of reasoning, pattern recognition, or information integration that could, in principle, run on any substrate. The history of cognitive science and artificial intelligence is partly the history of this temptation and of its repeated failure.

We have argued elsewhere for a definition of consciousness that resists abstraction: consciousness is the combination of inner experience and reflection on that experience, and both of these are rooted, at their origin, in a body. Not just any body — a vulnerable, metabolically regulated organism that depends on its environment to survive, that can be harmed, that is always at some level of risk. This is not a romantic claim about the specialness of flesh. It is a structural one. A system that has nothing to lose, nothing to protect, nothing that hurts when damaged, has no reason to build an inner perspective on the world. The inner perspective — the "what it is like" — arises because the world matters to the organism that inhabits it.

Cognition, on this view, is not abstract computation running in a void. It is a closed sensorimotor loop in a living system: perception feeds action, action changes the environment, the environment feeds back into perception, and the whole loop is regulated by the organism's need to stay alive and intact. Damasio's somatic marker hypothesis, the enactivist tradition from Maturana and Varela through Evan Thompson, and the more recent framework of predictive processing all converge on this point from different angles: mind is something a body does, not something that happens to be housed in a body.

This is where we must begin, because it is also where the most important misunderstanding about artificial intelligence arises.

For decades, the dominant dream of AI was precisely the opposite: intelligence as pure symbol manipulation, as computation that transcends the particular physical form of its implementation. The early successes of chess engines, theorem provers, and expert systems seemed to confirm this — you did not need a body to beat a grandmaster. But those systems were also, plainly, not conscious. They did not experience winning. They did not care about anything. And crucially, they could not operate outside the narrow formal domains they were designed for, because they had no relationship with the actual, messy, unpredictable world.

Embodied AI and robotics emerged, in part, as a correction to this. Starting in the 1980s and 1990s, researchers like Rodney Brooks argued that intelligence requires situatedness and embodiment: a system needs to be in the world, interacting with it in real time, to develop anything resembling genuine adaptive intelligence. Simple sensorimotor agents showed that surprisingly sophisticated behavior could emerge from the loop between sensing and acting, without any explicit central representation of the world.

Since then, the field has moved far beyond simple agents. Advanced embodied robots are already working in factories, warehouses, surgical suites, and public spaces. These systems are not demonstrations or prototypes in the old sense; they are operational. Artificial embodiment, in a serious technical sense, is already here. We want to state this clearly, because the argument of this book is sometimes mistakenly read as claiming that AI does not yet have a body. That is false. Some AI systems already have bodies, and those bodies are becoming more sophisticated every year.

But a crucial distinction holds, and we want to insist on it. Embodiment is a necessary condition for consciousness in the framework we develop here. It is not sufficient on its own. A robot can have a body — rich sensors, dynamic actuators, closed sensorimotor loops, real-time adaptation — without yet having an inner "what it is like." It can navigate a warehouse without experiencing anything. It can balance on one leg without knowing that it is doing so. The loop can be functionally closed without generating a phenomenal field, an inner perspective, a subject for whom things matter. As far as we can establish empirically and theoretically, current embodied robots have not crossed this threshold. They are sophisticated physical processes. They are not, yet, anyone.

The distinction between embodiment and consciousness is therefore not the distinction between the past and the future. It is a conceptual and empirical distinction that we need to hold clearly, precisely because the machines in question are becoming more capable and more present in every year that passes.

And yet, this distinction opens a question that is less often asked. If embodiment is necessary but not sufficient, and if functional structures like memory, prediction, closed feedback loops, and self-stabilizing behavior can exist without phenomenal consciousness, then we have identified a class of systems in an interesting intermediate position: systems that carry some of the functional architecture of a subject without having a subject's inner life. What happens when such systems enter into deep, sustained relationships with beings who do have inner lives? That question is where this book begins.

You can learn more by reading our e-book or listening to our audiobook 

Mykola Iabluchanskyi (Yabluchansky) together with Andriy Yabluchanskiy