The Importance of Animal

There is something an animal can do that the most complete library in the world cannot.

It can encounter something that has never been documented.

That sounds ridiculously simple.

It is.

And it is remarkably easy to forget when the answer arrives before you have even finished thinking about the question.

Knowing Everything We Know Is Not Knowing Everything

Imagine a hypothetical AI with access to every piece of documented human knowledge immediately before commercially practical incandescent lighting.

One constraint:

It has no access to what happens next.

It knows electricity.

It knows chemistry.

It knows combustion, oil and gas lamps, electrical arcs, and the materials people have experimented with.

Every documented experiment. Every success. Every failure. Every theory available at that moment.

Then someone types:

“I want to find a better way to make light.”

The AI can compare existing technologies, identify limitations, suggest materials, and propose experiments.

It may reason toward something genuinely new.

It may even propose the design that eventually works.

But proposing it is not the same as knowing that it works.

It cannot retrieve the results of experiments that have not happened.

Edison makes this example more useful precisely because Edison did not invent the light bulb.

Electric light existed before Edison. So did incandescent lamps.

What Edison and his team helped develop was a commercially practical incandescent lighting system. That meant improving the lamp and developing the generation, distribution, wiring, switching, safety, and economics needed to make the whole thing useful at scale. [1]

Many of the pieces existed.

The working system still had to be developed.

This is a thought experiment about the boundary of knowledge. We are lending the past an AI, not pretending modern computing could have developed independently of that history.

Freeze what is known at that moment.

Now somebody still has to build the proposed solution.

Test it.

Watch what fails.

Change the material.

Change the resistance.

Change the vacuum.

Discover that making the lamp work creates another problem somewhere else in the system.

Solve that problem.

Try again.

A reliable result nobody previously possessed adds something to what we know.

Document it. Make it available to the AI.

Now the AI can use it.

Neither the AI nor the person prompting it automatically possesses evidence that has not been produced.

Sometimes the answer is already in the library.

Sometimes all the pieces are in the library, but nobody has figured out how they fit together.

And sometimes a piece is still sitting somewhere in reality waiting for an animal to fucking notice it.

The Animal Has to Touch the World

This is what I mean by animal.

Not stupidity.

Not anti-intellectualism.

Not rejecting technology, education, research, data, or artificial intelligence.

I mean the living thing that encounters reality directly.

The animal looks.

The animal touches.

The animal tries something.

Something unexpected happens.

The animal becomes confused.

The animal asks why.

Sometimes the observation confirms what we expected.

Sometimes it does not.

That second outcome is valuable because it gives us something to investigate.

A mistake in the test?

A weird circumstance?

Or something our model of the world does not account for?

We do not get to know which just because we like our explanation.

AI Makes This Distinction More Important, Not Less

AI can help generate hypotheses and propose designs. Connected to experimental systems, it can also help test them. That is already happening. [2]

Good. Use that.

The point is to keep observation and testing in the process, whether people, machines, or both do the work.

Here is the workflow that worries me:

Problem → ask AI → receive answer → implement answer.

Then what?

Did it work?

What changed?

What happened that the answer did not account for?

Treat implementation as the final step and we stop looking.

We mistake having an answer for having settled the question.

Sometimes the answer is established, and using it is exactly the right thing to do.

An engineer should use existing research. A business should not have to rediscover accounting. Human progress depends on every generation not having to rediscover fire.

But accumulated knowledge has a boundary.

AI can help us investigate beyond it.

It does not remove the need to find out whether we are right.

A Better Airline Was Not Necessarily the Answer

When Virgin Atlantic entered commercial aviation in the 1980s, it did not discover that people needed transportation across the Atlantic.

Airlines existed.

Aircraft existed.

Routes existed.

So did conventions about what an airline was supposed to look and feel like.

A conventional analysis could recommend better food, better seats, better service, or a different price.

My read of Virgin's approach is that it also questioned the experience itself.

Why does flying have to feel like this?

Virgin drew on its music-business roots and made entertainment, personality, irreverence, glamour, and fun part of its airline identity. Its own history describes using personality, service, and provocative marketing to stand out. [3]

The lesson is not:

Make airlines fun.

That would be copying the answer.

The useful question is:

Which parts of this business are requirements, and which parts are habits?

Safety was necessary.

Professional pilots were necessary.

Aircraft maintenance was necessary.

A boring personality was not.

Recognizing that gives you something to test. It does not guarantee customers will care—or that the economics will work.

You still have to find out.

This Is Where Business Strategy Gets Confused

Benchmark the competitors.

Research best practices.

Analyze historical performance.

Read the case studies.

Apply the framework.

Follow the playbook.

All useful.

They tell you what other people discovered, what appears to work, and what might happen next.

They give you a map.

We confuse having a very detailed map with there being nowhere left to discover.

If ten competitors ask similar systems to analyze the same market using the same documented knowledge, they may get similar recommendations.

Meanwhile, the competitive advantage may be sitting three feet beyond the edge of the map.

In something a customer does that nobody thought to measure.

In a workaround employees stopped mentioning because everybody treats it as normal.

In a problem the industry has become exceptionally good at explaining and strangely comfortable not fixing.

Sometimes “I Do Not Know” Is the Correct Answer

“I do not know” can be an uncomfortable thing to say in a room that paid you for an answer.

Executives want recommendations.

Clients want answers.

Managers want certainty.

So we turn uncertainty into confident language.

We have a framework.

We have benchmarks.

We have data.

We have a 78-slide presentation.

Therefore, we must understand the problem.

No.

Sometimes the most intelligent conclusion available is:

We do not understand this yet.

That is not the end of the work.

It tells you what work needs to happen next.

Form a hypothesis.

Decide what result would show you are wrong.

Run a test that can actually reveal it.

Observe. Document. Adjust.

Figure out whether the result holds up somewhere else, or whether you encountered one weird circumstance.

Being willing to find out includes being willing to lose your favorite explanation.

The Fix Comes From Somewhere

Businesses do not pay for knowledge just because knowledge is beautiful.

Something needs to happen.

Revenue needs to grow.

Costs need to fall.

Customers need to care.

A product needs to work.

A company needs to survive.

That is the difference between knowing a lot about a problem and being able to solve it.

Imagine thirty people can explain why something is broken.

Twenty can benchmark it.

Ten can recommend best practices.

Five can make beautiful presentations about it.

Then somebody tests a different approach.

It works.

Oh shit. You actually have a fix.

They may have used research, AI, and decades of institutional knowledge.

They should.

Sometimes the fix is already in the answer key.

But when it is not, somebody has to find out what the answer key is missing.

And either way, the business needs the fix to work outside the presentation.

Keep the Animal

We should use AI aggressively.

Use the research.

Use the models.

Use the databases.

Use everything humanity has already learned.

There is no virtue in spending three weeks rediscovering something that can be retrieved in thirty seconds.

But after receiving the answer, keep looking.

Touch the thing.

Build the thing.

Run the advertisement.

Talk to the customer.

Launch the product.

Watch what people actually do.

Notice when reality refuses to behave according to the presentation.

Ask why.

For me personally, I know what I know, and I do not know what I do not know.

But “I do not know” and “I cannot figure it out” are not the same fucking statement.

I do not find answers. I figure them out.

Finding out assumes there is something to find. Figuring out begins when retrieval stops being enough. We, as a people, are getting remarkably good at finding things out—and forgetting how to figure things out.

Because no matter how sophisticated our knowledge systems become, there will always be a boundary between:

everything we currently know

and

everything there is to know.

The animal lives at that boundary.

And occasionally it looks beyond the map and says:

What the fuck is that?