What Is Embodied AI? The Robot That Thinks with a Body

Chapter 1: What Is Embodied AI? The Robot That Thinks with a Body

When you think of a robot, what do you see? Perhaps a metal humanoid with glowing eyes, like something from a science fiction film. Perhaps a mechanical arm in a factory, precisely welding car parts. Or maybe a small box with wheels that vacuums your floors while you are at school. Now consider this: when you think of a person who is intelligent, who do you imagine? Probably not a robot. You probably think of a human being—someone with a brain, yes, but also a body that walks, speaks, reaches, and touches.

Here is a strange question that philosophers and computer scientists have been wrestling with for decades: What if intelligence is not just about having a brain? What if the body itself is essential to thinking? And what happens when we try to build machines that can truly understand the world—not by analyzing data in isolation, but by moving through it, touching it, and changing it?

These questions sit at the heart of embodied artificial intelligence, or embodied AI. This is not just a technical field hidden in research laboratories. It represents a fundamental rethinking of what intelligence means and how it emerges. In this opening chapter, we will explore why embodiment—having a physical presence in the world—might be essential to intelligence itself. We will trace the philosophical roots of this idea, examine how it contrasts with traditional approaches to AI, and see how it shows up in the biological world and in modern robotics. By the end, you will understand why the robots described in this book are fundamentally different from the AI in your smartphone, and why that difference matters enormously for the future of technology.

The Question No One Thought to Ask: Can Intelligence Exist Without a Body?

For most of the history of artificial intelligence research, the dominant assumption was elegantly simple: intelligence is computation. The brain, according to this view, processes information through calculations, much like a computer processes data. If we could figure out the right algorithms—the right sets of instructions—we could create thinking machines without ever worrying about bodies, sensors, or physical environments. This approach dominated AI research from the 1950s through the 1980s and beyond. It produced remarkable achievements: programs that could play chess, solve mathematical proofs, and later, recognize images and translate languages.

But a small group of researchers grew increasingly uncomfortable with this picture. They asked a question that seemed almost heretical in a field obsessed with computation: What if thinking is not just something that happens inside a brain or a processor? What if true intelligence requires being in the world, with a body that can move, sense, and act?

One of the most influential voices in this debate was Hubert Dreyfus, a philosopher at the University of California, Berkeley. In his 1972 book *What Computers Can't Do*—later revised and expanded in 1992 as *What Computers Still Can't Do*—Dreyfus argued that the entire project of traditional AI was based on a flawed assumption. The flaw, he suggested, was the idea that human intelligence works like a formal symbol system. According to this view, the mind manipulates abstract symbols (like the pieces on a chess board) according to logical rules. But Dreyfus pointed out that human expertise does not work this way. When a chess grandmaster looks at a board, she does not calculate millions of possible moves using logical rules. Instead, she *sees* the game—her perception is holistic, intuitive, and deeply grounded in years of physical experience with the game.

Dreyfus drew on the philosophy of Martin Heidegger and Maurice Merleau-Ponty to argue that human beings are not detached observers of the world who first represent reality in their minds and then decide how to act. Instead, we are *embodied agents* who are always already engaged with our environment. We do not think first and act second. We act, perceive, and think all at once, in a seamless flow that is inseparable from our physical presence in the world.

This insight had profound implications. If Dreyfus was right, then no amount of clever programming would ever produce truly intelligent machines—not because computers are too weak, but because intelligence is not purely a computational phenomenon. Intelligence, at its core, is something you *do* with a body, not something you *compute* in isolation.

Two Views of Intelligence: The Brain in the Vat versus The Body in the World

To understand the difference between embodied AI and traditional AI, let us draw a sharp contrast between two imaginary creatures.

Imagine the Brain in the Vat. This is a thought experiment, originally proposed by philosopher Hilary Putnam, but here we adapt it for our discussion. Picture a human brain, removed from its body and suspended in a nutrient bath. Electrodes are connected to the brain, sending it electrical signals that simulate the experiences of a normal life. The brain "sees" images, "hears" sounds, "feels" sensations—all of which are actually just patterns of electrical stimulation. According to traditional AI theory, if the brain processes this information correctly, it would have all the experiences of a normal person. It would think, reason, and perhaps even believe it has a body—without ever actually having one.

Now consider the Body in the World. This is you, right now, reading this book. You are not a floating brain. You have hands that can hold the book or rest on a table. You have eyes that move across the page, adjusting constantly to focus. Your body is positioned in a room, with a sense of where the walls are, where the light comes from, and how gravity pulls on your muscles. When you turn a page, you feel the paper against your fingers. When you stand up, your legs respond to the floor beneath them.

Traditional AI, what researchers sometimes call GOFAI (an acronym for "Good Old-Fashioned Artificial Intelligence"), was designed with the Brain in the Vat model in mind. GOFAI systems process symbols—representations of the world—and manipulate those symbols according to logical rules. The world itself, in this view, is something like a database. The AI does not need to be *in* the world; it just needs an accurate *model* of the world stored somewhere in its memory.

This approach has produced enormous successes. When you ask a search engine a question, it does not reach out and touch the world. It searches through a vast index of text that represents the world, using algorithms to find relevant information. When you use a language translation app, the AI does not understand Spanish or Mandarin the way a human does; it manipulates symbols according to statistical patterns learned from millions of examples.

But here is the key insight of embodied AI: representations are not enough. Consider what you actually do when you navigate a crowded hallway. You do not consult a mental map and calculate the optimal path. Instead, you *flow* through the space, adjusting your trajectory moment by moment based on what you see, hear, and feel. Your body anticipates obstacles before you consciously notice them. Your feet know where the ground is. Your balance system responds to tiny shifts in weight without you having to think about it. This is not because you have a better internal model of the hallway than any computer could have. It is because your body is *in* the hallway, interacting with it in real time.

The embodied AI researcher Rodney Brooks articulated this intuition in the late 1980s with a radical proposal. In a series of papers and his influential 1991 book *Cambrian Intelligence*, Brooks argued that the entire framework of traditional AI—with its central representations, its symbolic processing, and its sense-plan-act architectures—was fundamentally wrongheaded. His alternative