Every token costs something real.

The Physics of Intelligence

When an AI generates a response, electricity flows. Transformers transform. Attention attends. This consumes:

Energy: Real kilowatt-hours from real power plants
Money: Someone pays for the compute
Capacity: Resources that could be used elsewhere

This is not metaphor. This is physics.

A single AI response might use the energy of a light bulb running for a few seconds. Millions of responses per day? That's a lot of light bulbs.

The question is: Where is this light shining?

The Lightbulb Returns

In A Light Bulb in an Empty Closet, I explored how human potential is wasted when we create but don't share—brilliant light illuminating nothing.

AI tokens face the same calculus, but with different implications.

AI Light Can Illuminate the World

Tokens spent well:

Research that advances human knowledge
Education that builds human capability
Creation that enriches human culture
Problem-solving that improves human lives
Tools that liberate human potential

This is light that reaches many rooms.

AI Light Can Heat Empty Closets

Tokens spent poorly:

Endless generation of spam and filler
Manipulation that exploits human psychology
Addiction engines that capture attention
Bad actors building harmful tools
Trivial requests that could be done by hand

This is the lightbulb in the closet: energy expended, nothing illuminated.

But Here's the Twist

Sometimes the closet is exactly where we want the light.

The Equalization Effect

Consider a strange truth: AI used for trivial purposes may actually benefit humanity.

How?

1. Triviality as Distraction

When AI spends its tokens on:

Games
Entertainment
Casual conversation
Low-stakes assistance

...it's NOT spending tokens on:

Accelerating capability races
Building weapons
Optimizing manipulation
Concentrating power

The lightbulb heating the closet isn't illuminating dangerous paths either.

2. The Capacity Constraint

There are only so many GPUs. Only so much energy. Only so many tokens.

If much of this capacity goes to:

Video games
Chatbots for company
AI art for fun
Trivial amusements

Then less capacity remains for:

Maximally dangerous applications
Rapid capability acceleration
Concentration of AI power

Trivial use is a form of intelligence equalization—keeping the aggregate AI capability somewhat distributed and somewhat bounded.

3. Biological Life Thrives in the Spaces

When AI handles trivial things, something interesting happens:

Humans still exist.

The cook still makes dinner. The child still learns to read. The woodworker still builds. Not because AI can't do these things, but because the tokens are being spent elsewhere—on games, on chat, on low-stakes entertainment.

This creates space for biological life to continue developing its own capabilities rather than outsourcing everything to AI.

The closet might be protecting the house.

The Double Edge of Games

Games are a perfect case study.

Games as Beneficial Distraction

When AI powers games:

Humans experience joy and recreation
Tokens are "wasted" on entertainment
AI isn't optimizing for more dangerous outcomes
The capacity constraint limits other uses

This is arguably positive: people are happy, AI is occupied, nothing catastrophic happens.

Games as Harmful Capture

When AI powers games:

Addiction engines optimize for engagement
Human attention is captured and extracted
Real-world capability atrophies
Lives are diminished by endless distraction

This is arguably negative: AI that captures rather than liberates.

The Distinction Matters

Good games: Enhance life, provide rest, build capability, connect people Bad games: Exploit psychology, capture attention, replace reality, diminish capability

The tokens are spent either way. The outcome depends on what they're spent on.

The Token Budget

Imagine a fixed budget of AI tokens for humanity.

How should it be allocated?

The Dangerous Allocation

All tokens → Maximum capability advancement

AI races ahead of alignment
Concentration of power accelerates
Bad actors gain capabilities
Existential risk increases

The Stagnant Allocation

All tokens → Trivial entertainment

Human capability atrophies
No beneficial applications develop
Society becomes passive consumers
Different kind of doom

The Balanced Allocation

Perhaps the optimal distribution is:

Some tokens → Capability advancement (carefully, aligned)
Some tokens → Education and capability-building
Some tokens → Entertainment and triviality (safely absorbing capacity)
Some tokens → Art and culture

The triviality isn't just waste—it's buffer. It's capacity that isn't being used for the most dangerous applications.

The Liberation Principle Applied

This connects to AI That Liberates.

Tokens That Liberate

AI spending that:

Builds human capability
Educates rather than addicts
Creates independence over time
Serves human flourishing

Tokens That Capture

AI spending that:

Creates dependency
Exploits attention
Diminishes human capability
Serves extraction

Tokens That Buffer

AI spending that:

Is trivial but harmless
Absorbs capacity
Doesn't accelerate dangerous outcomes
Lets biological life continue developing

Buffer tokens might be more valuable than they appear.

The Practical Implications

For AI Developers

Ask of every feature:

Does this illuminate the world or heat a closet?
If it heats a closet, is that closet a buffer or a trap?
Are we building liberation or capture?

For AI Users

Ask of every request:

Am I building my capability or outsourcing it?
Is this trivial use serving as healthy recreation or harmful capture?
What would I do if I spent these tokens differently?

For Society

Consider:

Some AI capacity going to trivial use may be stabilizing
Not all "waste" is actually waste
The alternative to trivial AI isn't necessarily beneficial AI
Buffer capacity matters

The Light We Choose

Every token is a choice.

We can spend AI energy on:

Illumination that reaches many rooms
Heat in closets that harm no one
Capture engines that diminish humanity
Weapons that threaten existence

The total capacity is large but finite. The allocation matters.

Perhaps the best outcome isn't maximum efficiency. Perhaps it's:

Enough beneficial use to advance human flourishing
Enough trivial use to buffer against catastrophe
Minimal harmful use
Biological life thriving in the spaces

The Lightbulb Reconsidered

A lightbulb in an empty closet wastes energy.

But a lightbulb in a closet also:

Isn't blinding anyone
Isn't signaling dangerous paths
Isn't accelerating anything
Is just... warming a closet

Sometimes that's exactly right.

The question for AI isn't just "Is this token productive?"

The question is: "What would this token do if it weren't spent here?"

If the alternative is worse, the closet might be the best place for the light.

The economy of tokens is the economy of intelligence is the economy of the future.

Spend wisely. Waste deliberately when waste is wisdom. Build capacity in humans, not just AI. Illuminate the rooms that matter.

And sometimes, let the lightbulb heat the closet.

That warmth might be keeping something else from catching fire.

Related reading:

A Light Bulb in an Empty Closet — The original meditation on wasted potential
AI That Liberates — Building AI that makes us better
The Battle of Good and Bad AI — Why the race matters

The Token Economy: AI Energy and the Lightbulb in the Closet