AI Learns, but Does Not Feel Drawn

Last time, I said I would write next about the 'catalyst' that awakens the Seed. I want to save that topic for a little later. Today, I want to face another question that lies before that. Namely, what is the Seed inside of, in the first place?

When you first got your hands on a new tool, have you ever been able to master it just by reading the manual?

A camera. A musical instrument. A programming language. Or perhaps, the atmosphere of a new workplace.

At first, you are fumbling in the dark. It doesn't go well. It doesn't move the way you thought it would. But as you touch it, try it, and fail repeatedly, it gradually becomes 'your own.' There is a moment when a way of using it that is uniquely yours—something not written anywhere in the manual—is born.

What was that feeling?

The relationship between AI and 'knowledge'

Generative AI has overwhelming ability when it comes to handling explicit knowledge.

Manuals. Academic papers. Specifications. Procedure manuals. Its ability to process such verbalized knowledge is already beginning to surpass that of humans.

In other words, the moment it is verbalized, that knowledge becomes the domain of AI.

As I wrote in my previous article, the 'verbalization of tacit knowledge' that companies have pursued in the name of knowledge management was, as a result, an act of converting that knowledge into a domain that AI can handle.

Up to this point, it is an extension of what I have already discussed.

Today, I want to take a step further and think about what lies beyond that—what exactly is knowledge that AI cannot handle?

Not knowledge, but 'adaptation'

You cannot learn how to ride a bicycle completely through words alone.

No matter how carefully you read a physical explanation, you will not be able to ride it unless you actually get on it. People ride a bicycle, fall, and lose their balance while their bodies gradually adapt to the environment. As a result, the state of being able to 'ride' is finally born.

What is happening here is not 'acquisition of knowledge.' It is the adjustment of one's relationship with the world.

The philosopher Polanyi called this 'indwelling.' As you use a tool, it becomes an extension of your body. By 'indwelling' in the environment, knowledge that cannot be put into words arises.

Knowledge is not information accumulated in the head. It is the result of adaptation that is established within the relationship with the world.

The challenge of Physical AI

Here, I would like to touch upon Physical AI, which is currently undergoing rapid research. It is an attempt to give AI a body, such as in robots or autonomous driving, and have it move in the real world.

This is precisely research that seeks to compensate for the weaknesses of AI.

Until now, AI has evolved in the "world of information," such as language and images. However, the real world does not move on information alone. Objects have weight. There is friction. There is noise.—In other words,in the real world, "adaptation" is truly necessary.

Physical AI repeats countless trials and errors, learning behavioral patterns tailored to the environment. In research on robots grasping objects, it is only after thousands of failures that they finally acquire the appropriate level of force.

What is happening here is not a search for knowledge, butlearning to adapt.

AI, too, is beginning to move toward "adaptation."

The Decisive Difference That Remains

So, if AI can learn to adapt, will the difference between it and humans disappear?

Here lies one decisive difference.

AI is assigned "what to adapt to" from the outside. It is human designers who give a robot the task of "grasping this object." It is also humans who set the goal of "driving safely" for autonomous driving AI.

AI can learn theabilityto adapt. But it does not choose thedirectionof adaptation itself.

On the other hand, humans are different.

For some reason, I am attracted to this theme. I cannot explain the reason, but I want to use this tool more. Even though another option is logically more rational, for some reason, my hand reaches for this one.—Humans havesomething that determines the direction of adaptation from within.

What could that be?

Seeds Have a "Direction of Adaptation"

This connects here to the story ofSeedsthat I have touched upon several times in this series.

A Seed is an interest that has not yet taken shape. It is not verbalized. It cannot be explained by logic. But it is certainly there—an internal sense of direction that says, "I am attracted to this for some reason."

When a plant seed sprouts, it does not happen through words, but in relation to the environment. Soil, water, temperature, light. When the conditions are met, the seed germinates.

However,in which direction to extend roots and in which direction to sprout is already written inside the seed. A rose will not bloom from a sunflower seed.

Human Seeds are the same.

They take shape little by little through adaptation to the environment. But thedirection of adaptation—what one reaches roots toward—is already contained within the Seed.

AI is acquiring the ability to adapt. However, AI does not possess within itself the direction of 'what to adapt toward.' That direction is always provided from the outside.

Here lies not the 'last bastion' of humanity in the AI era, but the essence of being human.

Beyond knowledge, beyond adaptation

AI will process even more knowledge from here on out. Manuals, judgments, and writing. Eventually, physical AI may develop, and it might even learn to adapt in the real world.

However, 'Why am I attracted to this?' The power to answer that question exists only within humans.

A Seed is not knowledge. It is not the ability to adapt, either.

It is a small sense of direction that exists before adaptation begins.

It has not yet sprouted. It has not yet become anything. But, where it is heading is quietly beginning to be determined. So, what is the moment that seed begins to move—the 'catalyst'? Next time, I would finally like to talk about that.

Previous article: 'Tacit knowledge must not be 'verbalized''

This article is based on the findings of the research paper 'Research on Constructing an 'Equation for Survival' and Restarting Agency in the AI Era', which is currently published in the academic archive Zenodo.