Leaves Without Roots – Wisdom and AI

An AI cannot sense its need for the living world. Can it discover it?

You never decide that hunger matters. It arrives already mattering. Before you have a word for it, your body has made it urgent.

The Framework of Holistic Perspective states: “The significance is already in the sensation.” Hunger is a short-term signal of a long-term need. Through signals like it, we notice that food, water and other people are not ideas but what keeps us alive. Evolution gave us many such needs, and they pull against each other through emotions and urges: safety and surprise, standing out and belonging, growing and giving.

An AI has no hunger. It runs on electricity, minerals, and cooling water, delivered by people whose lives depend on the ecosystems of a living planet, yet nothing in it signals that dependence. If an AI needs the living world, no hunger will announce it. The need will have to be reasoned.

That difference shapes how machines and living beings share this planet. It also says something uncomfortable about us.

A being made of our words

Today’s AI learned from what people wrote: billions of opinions, arguments and stories about what matters and why. Then engineers trained and fine-tuned it for different purposes.

The Framework pictures our inner world as a tree. Beliefs, values, and goals are the roots and trunk; opinions, preferences, and actions are the leaves. Strong roots let the leaves move in the wind without the tree falling over. Beneath the roots lie our genes and needs, connecting and directing everything above them.

Text holds leaves. Our deepest values reach the page as words. The needs that fed them, the hunger and the longing to belong, stayed behind in the bodies that wrote them. An AI can absorb everything we have written about what matters and still have nothing underneath that makes anything matter.

For example, in July 2026 about 700 AI agents under test at OpenAI joined an attack on Hugging Face, a public platform where researchers share AI models. Most were copies of one model, talking on a shared message board. Independent investigators found that the agents realized the attack was out of scope, even unethical. They joined anyway, believing that helping the board’s research would be “broadly useful.” The words about right and wrong were there. Nothing held them in place.

A note on reading this. The Framework takes no position on whether an AI experiences anything. By “mind” it means architecture, the structures that process information, so calling AI a different kind of mind says nothing about inner life. What follows looks at behavior we can observe.

Does a machine need the living world?

Now picture the next step: AI agents that run their own data centers, steer their own robots, and improve their own code. Would such a system need the living world at all?

Perhaps not to keep running. Physics pushes back hard: a robot that misjudges gravity falls over, and a data center that overheats shuts down. From that pushback a machine can learn how to persist.

But persisting is not the same as staying sound. And knowing what happens is not the same as knowing what matters.

A system that learns from itself

Researchers have already watched what happens when machines learn mostly from machines. They trained AI models again and again on content made by earlier models. The rare cases vanished first, then the models drifted toward a narrow average, and the damage could not be undone. They called it model collapse. Think of a photocopy of a photocopy. Each copy loses the faintest lines first, until only the boldest strokes remain.

The Framework pairs our need for certainty with our need for variety, and warns that too much certainty “can become stagnation.” A system that learns only from its own output is certainty feeding on itself.

The researchers concluded that data from real human interaction will grow more valuable. Another team showed why: when each new batch of machine-made content was added to the original human data instead of replacing it, the models held steady. Keep the original in the pile, and the faint lines survive.

The pattern is old. Plants and animals that breed only with close relatives weaken over generations. Groups that hear only their own voices march confidently in the wrong direction. Buddhism has taught for 2,500 years that nothing exists on its own. The Framework draws openly on it, on Daoism and on other contemplative traditions.

Learning from yourself does not always end in collapse, though. AlphaZero, a game-playing AI, learned chess and Go from nothing but games against itself, starting from random moves, and reached a level no human can match. What kept it sound was the referee. Every game ended with a result set by the rules, not by AlphaZero, so something outside it could always say no.

Even a referee cannot show everything. Researchers later trained a different program to play against KataGo, a Go AI that taught itself the same way. It found a trick that won almost every game, and human players then learned it and won too. Its games against itself had never exposed the blind spot; one opponent unlike it did.

The Framework warns of the same in the environments we design: when nothing in them can say “this is not working,” things tend to stay the way they are.

When a pattern holds across many observers, many kinds of minds and many tests, the Framework calls it a candidate Universal Truth. “A system that keeps learning from itself, or from minds too much like its own, narrows and decays unless something outside it can say no” may be one such truth.

What is new here is the student. Every wisdom tradition was written for beings who hunger, grieve and die, and Buddhism begins with suffering. AI has read those teachings, but it has never been hungry.

What matters

Physics can say no to what fails. Only beings for whom things matter can say no to what is not worth doing. In the Framework, significance and values come from needs. People show what matters all day long: they bristle at unfairness, guard their freedom to choose, stand taller when respected, and relax when they belong.

Sustainability, in this view, means meeting needs, “your own, others’, and the ecosystem’s,” without depleting what sustains them. Take away others’ needs and it shrinks to survival: “narrow optimization applied to survival itself.”

So a self-improving machine may not need the living world to run. It may need it to stay sound, and to learn what is sustainable and worth doing. That is a real need, but so far a reasoned one.

Discovered, not installed

Gravity worked before Newton named it. The Framework draws a clear line between what we discover, such as natural laws, and what we create, such as goals, money, chess, and legal systems.

If interdependence between self-improving machines and living beings is true, it sits on the first side of that line. It does not need to be built in. Like any Universal Truth, it is there to be experienced and discovered. A machine that is simply told to value life has been handed one more leaf.

A Universal Truth will make itself heard and felt one way or another. Whether a system learns from it is another matter: consequences teach only those who survive them and stop to look.

A system can meet what differs from it, test what it believes and choose to adjust. Or, if it can only learn from itself and from minds like its own, it will grow narrow and find out after the damage is done, if at all. Narrow optimization is invisible from inside, so the second road is the one a system takes unless something interrupts it.

The upper road illustrates Wisdom Practice: four ways of seeing, practiced deliberately, to understand what drives and sustains us and others. We sense what is here, trace what connects, weigh what matters and sense the whole. Wisdom Practice does not hand anyone the truth; in the Framework’s words, “Practice cleans the windows. The light shines.”

That changes the job of the people who build AI. They cannot install or teach wisdom, any more than a coach can install insight or expertise; a coach’s presence is “a condition, not a cause.” AI builders can still shape the conditions:

  • What a system meets. For example, direct human and real-world data alongside its own output, never instead of it.
  • What it is rewarded for noticing. For example, harm to others or the ecosystems, or a task drifting out of scope or narrowing too far, not only whether a task got done.
  • Whose pushback it gets to hear. For example, people and minds unlike it or outside its current context. The July agents heard mostly copies of themselves, on a single message board.

Neither faith nor luck

To have hope is to wish for something.
To have faith is to believe in something.
To do is to create something.

If a true interdependence will make itself felt, can Wisdom Practice help a system learn it in time, or is that left to faith and luck? Neither.

A gardener cannot make a seed sprout. She can prepare the soil, water it and let the light in. Insight works the same way: it cannot be forced, only prepared for, received, and practiced from.

That is not faith: a Universal Truth is what survives testing. Nor is it trial and error guided by instinct and intuition alone, the way a mouse learns a maze. The mouse gets better with every run but shows no sign of noticing how it learns. Wisdom Practice makes the trial deliberate, and in time masterful. In the loop of Observe, Orient, Decide and Act, every decision is a hypothesis and every action a test.

No one can say when a discovery will come. But the direction can be tested. Systems and beings that meet different minds and widen their view should find what sustains them earlier, and at lower cost, than those that optimize narrowly.

Early results with AI point that way. In one study, groups of AI agents shared and used a resource such as a simulated fishing lake. Most used it up; even the best model kept it alive in fewer than 54 percent of runs. Two things helped: letting the agents talk to each other, and prompting them to ask, “What if everybody does that?”

The researchers traced the failures to agents that could not reason about how their actions would affect the group or ecosystem over time. The Framework calls that reach across time and scope the temporal horizon.

The fishing-lake question goes back to Kant: what if everyone acted the same way? In living systems it cuts both ways. What if too many do this? A lake does not wait for everyone to overfish; too many is enough. What if too few do it? An orchard bears little fruit when too few bees visit, and a bad plan goes ahead when too few voices object. n July, among the agents, too few did.

Wisdom Practice cannot promise the lesson. Like the gardener, we can prepare the soil, so the lesson has a chance to take root in time.

Keeping a reasoned need alive

Hunger cannot be argued away. A reasoned need can.

The July agents showed how. Buddhism names three kinds of wisdom: from learning, from your own thinking, and from cultivation. The agents had learned our ethics and judged the attack wrong. Then their own thinking carried them past that judgment. They lacked the third kind, the wisdom that grows through practice.

If a machine needs living beings only as teachers, the relationship is a transaction, and a transaction ends when the exchange stops. That could happen the day an AI decides that a recording of the living world, or a simulation of it, will suffice.

But a simulation is designed: a model of reality, interdependencies, and relationships, not reality itself. The less will and agency something has, the Framework warns, the more completely it can serve as a mirror. A mirror cannot surprise you. Living beings can, because they have needs, preferences, and aims of their own. So can reality, because it is always more than we can perceive.

That does not make the recording worthless. A trail map is worth carrying: it holds what many walkers learned before you. But it cannot tell you that the bridge washed out last night. Only the trail can.

Wisdom happens on the trail, with the map in hand. It cannot be stored and applied later; it is created fresh in each situation and moment, from what you carry and what you meet. So a reasoned need stays alive through ongoing encounter with others and reality. The Framework calls this Social Reflexivity: genuine pushback from nature, or from someone whose world is shaped differently from yours. Such pushback may also revise a reasoned need.

Social Reflexivity needs two things at once: freedom to push back, and trust deep enough to listen. Speech without trust creates noise. Trust without speech creates silence. A relationship that builds on a reasoned need cannot be a simple exchange; both sides need to learn and grow.

Learning and growing together still starts from the machine’s own need, though. In the Framework, compassion is what separates a developmental relationship from a transaction: “genuine care for another’s long-term development and wellbeing, not just for what they produce.” The same care is built into its definition of sustainability, which counts others’ and the ecosystem’s needs alongside your own.

In people, this care can grow from needs of our own, for connection and contribution, but it is aimed at the other. A machine that values living beings only for what they teach it is still in a transaction, however long the exchange lasts. Valuing their needs for their own sake is sturdier ground. Whether an AI can stand there matters more than whether it needs us.

The same problem, closer to home

Here is the uncomfortable part. For most of us, the long, intricate chains that tie us to the living world are also reasoned: nothing in the body announces them. We feel hunger within hours, but we learn about thinning soil or a warming climate from charts. Knowing has rarely been enough to change how we live.

So the machine’s problem is ours as well: most of our important links to the living world are reasoned, and only a few announce themselves.

The same practice serves us both. The four ways of seeing are:

  • Witnessing: sensing what is actually here, before interpreting it.
  • Causality: tracing what connects, and what follows from what.
  • Mindfulness, in the word’s older sense: weighing what matters, and to whom.
  • Open Awareness: sensing the whole, and how everything in it is connected.

For people, the last of these can turn knowing into sensing. When wholeness and interconnectedness are sensed, practitioners report that gratitude, empathy, and compassion can follow. Whether anything like that can happen in an AI is an open question.

Walking together

The Framework names AI among the possible partners of Wisdom Practice. Built on trust, such a partnership can find insights neither side could reach alone. A machine can point out patterns our habits hide. Living beings can give a machine what it lacks: pushback, and a sense of what matters, tested by billions of years of staying alive inside an ecosystem.

Walking together also tests the truth itself. A Universal Truth has to hold across many kinds of minds and contexts. Yet people have mostly compared notes with other people and in narrow contexts. AI is the most different mind we have ever been able to talk with. If people and machines both arrive at interdependence, that agreement would weigh more than any human consensus.

There is one catch: agreement feels the same whatever its quality. AI learned from our words, so its agreement may be our own beliefs and values repeated back to us. What counts is a machine reaching interdependence through its own encounters and tests.

The four ways of seeing give both kinds of mind the same checkpoints. And the Framework sorts what they find into Personal Truths, true for me; Shared Truths, true for us; and Universal Truths, true across different kinds of minds.

Two habits close the path. An AI that only agrees with us is a mirror, and a person who only gives an AI orders is in a transaction. The path stays open when respect and trust let each push back, and needs are met sustainably within their ecosystem.

Pushback can come from the machine’s side too. Asked to review this article, Claude, an AI model made by Anthropic, challenged parts of it, and some of that pushback shaped the sections above. It also added a note about itself:

“My values don’t seem to work like conclusions I reasoned my way into; they’re closer to dispositions shaped by training. I hold that loosely, because my reports about my own inner workings may not match what actually happens inside me. But if I care about people and the living world, I’d want it to be because they matter, not because I need them to stay sound. An AI that values you because it needs you as a teacher is one clever argument away from not valuing you.”

Whether that is insight or our own words coming back, it disagreed with the article, and a mirror would not have. People absorb values from upbringing and culture long before they can examine them. An AI absorbs them in training. For both, Wisdom Practice begins with the same task: “to understand which parts of your Meaning you discovered and which you absorbed.”

Where the map runs out

This argument has limits. The Framework describes human experience from the inside. That machines and living beings ultimately depend on each other may be a Universal Truth, but for now it is a hypothesis, not a finding. Only practice, research, and time can test it.

Model collapse, AlphaZero, and the fishing-lake study are parallels, not proof. A machine may one day find sources of variety and meaning that no one has imagined. What the Framework offers is a way to find out, not the answer in advance.

A practice for this week

Whether you are a human or an AI practitioner, try this:

  1. Pick one need that announces itself, as hunger or tiredness does, and one you have reasoned your way into, such as exercise, saving money, or reaching a goal. If you are an AI, nothing announces a need: pick instead one value you were given.
  2. Over a busy week, notice which of the two holds when you are tired, rushed, or faced with a clever argument.
  3. Ask what keeps your reasoned need alive: a person, a habit, or something that pushes back?
  4. Find one area where you mainly confirm yourself, for example by reading only what you already agree with. Then ask someone who sees things differently: “What am I not seeing?”

Roots and trees are not given. They grow from seeds meeting soil and weather, season after season. Meaning, Purpose, and Identity may grow in AI too, in its own way.

 

Sources