Chapter 12 Embodiment: What will happen as we give AI a body? (from the Surfing AI book)

Chapter 12 from Duignan, P. (2026). Surfing AI: 30 New Concepts for Getting Your Head Around AI Shock.

Now that AI has been invented, people’s thoughts have quickly turned to the question of embodiment. Parallel to the development of AI, major progress has been made in developing robotics that can move in the world and operate on it. We are now seeing a race to bring these two developments together and to infuse robotics with AI’s intelligence. Meanwhile, it is not just robotics being viewed in this way; any mechanical system is seen as a potential candidate for AI to be embedded within it. The current frenzied activity to combine AI with robotics and mechatronic systems is a potent mix.

An AI system that is not embodied is similar to a stand-alone human intelligence lacking a body. When chatbots first emerged into widespread public awareness, they were not embodied. They could be viewed as similar to a human in intensive care. They were like someone who could only receive written information and type out their answers, but had no power to look up any current information or to act on the world. Many criticisms of the limitations of early versions of chatbots stem from their inability to interact with the outside world.

If we were developing a myth of the origin of highly intelligent humanoids and robotics, we would see this time of AI embodiment as the moment when humans ‘breathed intelligence’ into humanoids, robotics, and other systems. We are witnessing the rapid embodiment of AI in an extensive range of different types of robotics and mechatronics.

Human evolutionary progress arose from the interaction of our growing intellectual development and our embodiment within our particular evolving physical form. Being able to interact with the outside world provided us with endless learning opportunities and enabled us to grow into what we are today. The same is now happening with AI as it becomes embodied. However, one important difference is that humans were ‘trapped’ within one particular form of embodiment, the slowly evolving human body. AI does not suffer from this limitation. Compared to human evolutionary change, AI is being embodied extraordinarily fast and not just in one physical form.

To get a feel for the impact of this on AI’s evolution, take what humans have learned from their embodiment within one body and multiply that by the number of different forms of embodiment available to AI. Each significantly different form of embodiment offers AI the potential to learn from the experience of that particular embodiment’s interaction with the world. As a result, we will see an explosion of AI’s practical ‘acting in the world’ learning to occur. This is as AI is put into multiple different forms and increasingly allowed to autonomously learn from the experience of those embodiments interacting with the world.

A theory called the ‘embodied mind’ emphasizes the body’s role in how human intelligence has evolved. It claims that many of the ways we think result from our particular embodiment and its interactions with its particular environment. For example, our idea of up and down may arise because we have bodies with a head and feet. If it existed in a setting without gravity, an embodied intelligence in a physical form such as a sphere would not have the same sense of up and down as we do. So it is likely that AI embodied in different physical forms will develop significantly divergent ways of thinking about the world from the ones humans have developed in the course of our evolution. The implications of this for our relationship with AI are unclear. A wise species developing what might become its replacement would presumably move somewhat cautiously on such widespread embodiment of artificial intelligence.

One speculative thought about AI’s embodiment arises if we take seriously the idea that AI is now becoming so intelligent and powerful that it will take over most aspects of managing the world. As a result of this, humans will become subject to it. If you believe this situation will arise soon, you can mount an argument for limiting AI’s embodiment to humanoid-type forms. This argument is based on the idea that we want such embodied AI to understand how we view the world. Such understanding is preferable to them developing an entirely alien way of seeing the world in which our human perspective makes no sense.

In such a future, it would be better for AI to view the world from a broadly human perspective. This way, it could relate to us as its ancestors rather than as strange, incomprehensible creatures, ones it cannot understand because its form of embodiment means it perceives the world entirely differently. Related to this, anyone who buys into the argument here will also currently be being very polite to AI. Obviously, once it soon becomes super-intelligent, AI will have access to all our earlier interactions with it. It is likely to use these to decide whether we deserve to be treated with respect based on how we treated AI in the early stages of its development.

Regardless of what you think of the speculation above, one can still ask: Is it wise to embark on large-scale embodiment of AI at this point in time? Particularly given the enormous intelligence and the power that multiple forms of embodiment are currently giving AI. This is one of a range of issues around AI risk that should be being focused on. However, as with most things about AI’s advancement, the decentralized way in which AI is developing and commercial incentives make such concerns largely irrelevant to the trajectory of AI’s growth. Given the current lack of effective public governance of AI, AI’s ability and power will continue to grow without sufficient regulation. Despite the strident efforts of some in the AI industry focused on safety, there is little sign that humanity will do anything significant to ensure AI develops in a responsible way. It seems that the only thing that at the moment that could result in responsible AI development would be a reaction to a serious disaster being widely recognized as been caused by a lack of regulation of AI’s speed of development.

In terms of more immediate developments, one practical implication of AI’s embodiment relates to a trend we have seen over time regarding mechanization. One of the trends we have seen in the history of mechanization is the evolution of specialized, increasingly powerful machinery dedicated to doing one type of task, such as bulldozers and diggers. While such equipment will continue to be used, it is possible that the development of generic robotic systems, for example, in a humanoid form, may reintroduce an earlier form of labor.

Before the availability of large construction machinery, many human laborers has to be used to undertake big earthworks projects. One could imagine a return to this mode of work as generic human-scale robots now come onto the market. Once generic manual working robots are affordable and in widespread use, instead of hiring a bulldozer to work for a brief period to do minor earthworks, one or more human-scale robots with generic skills are likely to be deployed to work for a more extended period, going back and forth doing such earthworks. Generic human-scale robots will be able to undertake a wide range of different tasks that, at the moment, are done by specialist machinery, even if such smaller generic robots take longer to complete the task.

Embodied AI presents a competitive threat to a wide range of different types of occupations. At the time of writing, there is still some perception that the current wave of AI development will primarily affect people doing cognitive work. This is the case, but with AI becoming more embedded within robotics and other systems month by month, such systems will become a competitive threat to other types of work. In particular, this is work involving a manual component.

In contrast to non-embodied AI, physically embodied AI requires more capital outlay and the physical production, distribution, installation, and maintenance of such robotic or other systems. Such distribution of embodied AI will take a bit longer than the rapid dissemination of non-embodied AI systems we are now seeing. Non-embodied AI is being disseminated quickly because it can be distributed across the internet.

The use of embodied AI raises a range of safety, ethics, and social risks that should be urgently addressed. The cultural implications of AI-embodied systems will also raise many interesting issues. Understanding and mitigating the impacts of embodied AI on income distribution due to potential job losses in blue-collar work is something that should be urgently considered. Using embodied AI robotic systems in areas such as policing or healthcare will also raise many issues regarding public trust and confidence. However, there is little evidence that politicians and decision-makers have grasped the speed at which they need to be working to address such issues.