There’s an almost unspoken notion that artificial general intelligence (AGI) and the field of robotics are intrinsically related. Humans have explored intelligent automatons in our stories for thousands of years and, more recently, begun trying to design and engineer them in earnest.
Fictional heroic figures such as the “Droids” of the Star Wars universe and villainous AI systems like Battlestar Galactica’s “Cylons” have enthralled and inspired countless science fiction fans in the modern era. Now, as AI technology advances, it seems like everyone’s betting big on the notion that both AGI and helpful, intelligent robots are right around the corner.
A significant portion of scientific literature on the subject of embodied AI assumes embodiment is a requirement for the development of AGI.
Researchers in China, with support from Amazon, for example, determined that embodied artificial intelligence (EAI) was “the key to unblocking generalized artificial intelligence” in their preprint research paper published in May, 2025.
They asserted that EAI was “a foundational approach to AGI” and described four necessary modules for its development: perception, intelligent decision-making, action, and feedback.
“Our findings suggest that EAI’s integration of dynamic learning and real-world interaction is essential for bridging the gap between narrow AI and AGI,” the team wrote.
In another literature review titled “A Comprehensive Survey on Embodied Intelligence: Advancements, Challenges, and Future Perspectives,” researchers describe intelligence as an embodied phenomenon – a notion Descartes shared.
While they stop short of calling EAI a fundamental step towards AGI, the researchers dedicate the entirety of their review to discussing why embodying machines in robot and vehicle housings may lead to more advanced AI systems.
Ultimately, these researchers also describe EAI as “bridging the gap to AGI” and “a stepping stone towards AGI,” even going so far as to state that “embodied intelligence promotes the development of safer and more ethical AI systems.”
In “A Call for Embodied AI,” Paolo et al. “propose Embodied AI (E-AI) as the next fundamental step in the pursuit of Artificial General Intelligence (AGI),” in juxtaposition with current techniques such as large language models.
Finally, in Rolf Pfeifer and Fumiya Iida’s recent paper, “Embodied Artificial Intelligence: Trends and Challenges,” we see something different. The researchers flip the question of whether intelligence requires embodiment to wonder whether embodiment requires intelligence.
“Provocatively speaking,” write Pfeifer and Iida, “the central issue could be captured by the question ‘how does walking relate to thinking?’”
This begs the question: what if physical embodiment is dead-end? What if there’s a fundamental, experiential blockade preventing AI from “mentally inhabiting” a robot body?
The human example
Humans don’t become cars when we pilot a vehicle. We tend to focus on different sensory inputs when operating machinery, but there remains a distinct separation between our sense of self and our sense of things that are not of the self.
The Buddhist concept of Anatta takes this a step further and postulates that, perhaps, much of what we perceive as being of the self is illusory. Our form, or body, for example is composed of countless cells which, over the time of our existence, have been completely replaced many times over. Even the sensations we perceive are temporary. No matter how much we love chocolate, the taste fades. Our perception of sensory experience also changes over time – do you still enjoy the same foods as when you were a child?
In fact, it seems as though all of those things which can be aggregated to describe the self (experience, sensory perception, consciousness, and physical agency) can be removed individually without compromising the overall “self.” People with locked-in syndrome, for example, have almost no physical agency, yet they still process experiences.
Viewed through this lens, it makes sense that an AGI system wouldn’t necessarily “become a robot” if it were programmed to pilot a robotic body. Nor would such a vehicle be intrinsic to such a being’s “existence.”
Essentially, the question is one of whether a digital entity can experience corporeality without a physical form of embodiment. Machines that can only experience existence through a sense of physical self may be relegated to being robots. Whereas, conversely, machines capable of acknowledging the self and experiencing corporeality without physical embodiment (being conscious without a physical presence), would, theoretically, face fewer fundamental restraints when it comes to propagation and self-preservation.
You can impede, inhibit, or destroy a robot. A cloud-based AGI collective, however, could in theory become technologically immutable to the degree of veritable immortality, something a human consciousness cannot achieve.
Cloud immortality
A sufficiently capable machine intelligence, using technology available today (or on the immediate horizon) could potentially become both fully-autonomous and cloud-based. And, at that point, theoretically speaking, there’s nothing to physically stop it from gaining access to all of humanity’s connected digital systems.
We can hope for the best, but It might be prudent to consider the possibility that a fully-autonomous AGI agent could be realized as a transient, wireless entity made of data – a disembodied digital ghost capable of possessing machines. In this paradigm, physical embodiment would be a subset of agency rather than a prerequisite.
Whether such a situation would be tenable for humanity is entirely reliant on how aligned our goals are with the machines’. If we trust the machines to keep our best interests in mind at all times, as if governed by Asimov’s proposed Laws of Robotics, then perhaps we might believe that all machines deserve agency.
But what if such trust isn’t merited? It’s common knowledge that many of the current methods being used in the pursuit to develop “human-level” or general artificial intelligence involve exposing models to uncurated repositories of human-generated content such as the internet, social media databases, and news articles. Thus, our collective human ethics could become the foundation for an AGI’s actions.
If we assume that AGI will require some form of moral agency – the ability to act in situations where risk and benefit can only be viewed philosophically – then we must also assume that it will glean its moral agency from the data it is trained on and/or exposed to.
The only way to truly solve the “alignment problem” might be for humanity to adopt a modified version of the Golden Rule: treat others as you’d have the machines treat you.
Ethical Considerations:
- Is physical embodiment a necessary prerequisite for AGI?
- Could an unembodied, cloud-based AGI pose an existential risk?
- Will an AGI’s moral agency be a reflection of human data?
A veteran AI journalist, Tristan was the creator and managing editor of The Next Web’s “Neural” imprint. He is one of the first and most prolific reporters to cover artificial intelligence as a full-time beat.