Rethinking Intelligence as Movement–and why AGI won’t happen without embodiment

Intro - Framing the question of intelligence

Artificial intelligence is emerging all around us. The technology has been gradually making its way into the background of everyday products over the past 50 years, until suddenly ChatGPT opened everyone’s eyes to the immense power of the Law of Accelerated Returns. With Silicon Valley VCs and Big Tech alike pouring money into generative AI, in the next few years, we can expect to see the following changes in society:

Businesses using AI to make quality of life decisions for consumers, including who qualifies for a mortgage and life/death decisions (self-driving cars)
Jobs being displaced by new occupational roles at a pace we haven’t seen since the Industrial Revolution
Governments incorporating AI to make decisions in law enforcement, military warfare and the content that gets disseminated into classrooms

Beyond these immediate concerns, a once distant question has become imminent: will AI surpass human intelligence, and if so, what is our purpose in the world? Before we can discuss whether AI will surpass human intelligence, we need to have a common definition of intelligence.

What is intelligence? - The role our body plays

Ask today’s leading scientists and you’ll hear a variety of answers, from solving math problems to scoring highly on standardized tests. A common viewpoint is that the human brain is what gives rise to intelligence, and AI has been modeled off of this premise. The deep learning models that underpin GenAI are based on the concept of neural networks, inspired by how neurons in the brain form connections. As the thinking goes, the human brain is limited in physical capacity, whereas a digital neural network can scale infinitely, which will eventually result in a AI gaining superior intelligence. If we solely look at compute power, this makes for a compelling argument.

Exponential Growth of Computing visual chart — Source: singularity.com

But this assumes that modeling our brain, separate and apart from our body, is sufficient to replicate our intelligence. We will now challenge this assumption.

In order to do this argument justice, we would have to consider what it would be like to have a brain but to have never had a body; but the best we can do is imagine what it would be like to lose our current body, which unfortunately is not the same thing. We would still be considering intelligence with the benefit of having had a body for an entire lifetime. This conceptual gap makes questioning the role our body in intelligence deceptively challenging. It is also core to the problems underlying misconceptions about human vs artificial intelligence.

With this caveat aside, we will try to account for how the body influences the formation of our thoughts by considering the role the nervous system plays in our intelligence.

Intelligence as movement - the nervous system as an extension of the brain

One way to better appreciate the relationship between the body and brain is to reconsider the separation of the brain and nervous system.

At any instant, our bodies (nervous system + brain) are moving in relation to other entities in space and time, producing a somatic aggregation of sensory data. We aren’t merely looking at snapshots through our eyes like photographs. Nor are we hearing sounds through our ears as isolated recordings. Rather, our nervous system is simultaneously aggregating the experiences of 100 trillion cells that are all being moved in subtle ways (our eardrums vibrating in relation to the distance we are to an object, our retina firing from the movement patters of light, our lungs oscillating with the movements of our breath).

To ground this paradigm, let’s consider an example. Many of our conversations have to do with our experiences of moving around in the world. Let’s apply the concept from above and reconsider the definition of an everyday object: What is a coffee mug?

Scientifically, we can describe this object in similar ways as a machine might. We can objectively describe the atomic structure of the mug. Similarly, we can name the shape, color, weight and other attributes. These are all appropriate ways to measure the properties of the cup, but measurement is not what gives the coffee cup meaning, movement is.

What the cup means to a human being are the movements we might perform in context with the mug. It’s shape is defined more by the way it fits into our hands than geometrical properties. Similarly, we come to know a mug by the fact that they can hold liquids like coffee. Again, the word coffee has no inherent meaning, but to a human being who has had the experience of drinking coffee, the word represents the physical movements that happen inside our body as we feel warmth in our belly and as caffeine begins to dilate our blood cells.

Why does AI seem intelligent? Words, numbers and experiences

What makes AI seem intelligent is that it can manipulate the symbols that we use to represent real-world experiences. These words, numbers and pictures only approximate our experiences; we can communicate with each other using them because we can relate to the shared experience of having bodies that move in the world in similar ways.

Since unembodied AI can’t move in the world, it has to guess at the meaning. But seeing a picture of a coffee mug, or hearing the word “mug” means nothing without the experience of drinking coffee from a mug and the subtle cascade of movements it would cause inside our bodies.