-->

For all our KMWorld friends - register NOW for London's KMWorld Europe 2026 & save £300 with the code HOLIDAYS. 
 Offer extended until 5 January - lock in this best rate today!

Why Embodied Knowledge Matters and Why AI Doesn't Understand

Article Featured Image

Potential Knowledge

A machine-learning large language model doesn’t have tacit knowledge. It consists of potential knowledge. That potential enables the system to generate surprising, useful, and usually true statements. Having to live within a body’s constraints on information and processing would only limit today’s AI. So, the old philosophers turn out to be right: Bodies are impediments to knowledge.

Of course, that obscures the fact that today’s chatty AIs’ potentiality ultimately does come from meat: the millions of situated bodies that thought things and wrote them down. AI’s statistical analysis of those texts positions words in a multidimensional coordinate system that lets it recombine them in ways that can produce new knowledge. It can even speak from the situation of a 17th-century French man. I tried that experiment with this very column, and it did a plausible job of expressing that man’s scorn and derision.

Yes, that French person was as generic as my prompt asked for. But that’s not entirely out of place, since being situated means being part of a specific culture and time, and cultures, by their nature, have some shared ideas and sensibilities. Our new machines are already good at simulating situated embodiment, although better at it for the classes of people who were allowed to speak in ways that could be preserved and eventually became part of the training materials for AI.

Bodiless AI Doesn’t Care

But there are some important aspects of embodiment that limit bodiless AI as well.

First, machine learning systems, unlike bodies, have nothing at stake. The creators of them do, but the machine itself doesn’t give a hoot if it gets something wrong or harmful, any more than your knife cares if you sharpen it. A model, of course, can be constrained by what it’s trained on and then by guardrails that apply rules—often over-vigorously—on the words that it’s allowed to utter. However, the model follows those rules without caring about them in the least. Why does caring matter for AI? Because caring is not a simple sensation or reasoning process. It’s a whole-body and whole-mind phenomenon. Getting it right, especially when it comes to specifics, seems like one of the most complex and wide-ranging functions of the embodied, situated individual that you are.

Second, AI agents will have lots of specific data from us, especially what we absorb or create with our computing devices and the data trails we leave behind, but our embodiment goes beyond what we type. What we write is a portrait of ourselves that we create. Bodies tell a different story. Often so softly that it’s tacit.

KMWorld Covers
Free
for qualified subscribers
Subscribe Now Current Issue Past Issues