How would you recognise intelligence if you encountered it for the first time?
It sounds like an easy question.
Until you try to answer it.
For most of history, we have never had to ask.
The only intelligence we knew intimately was our own.
Everything else was compared to us.
When computers first appeared, the tests seemed obvious.
Could they perform arithmetic?
Could they play chess?
Could they solve equations?
Each success felt like another step towards intelligence itself.
Then something curious happened.
As soon as a machine mastered one of these abilities, people quietly moved the goalposts.
"Well," they would say, "that isn't real intelligence."
Chess became calculation.
Calculation became automation.
Automation became routine.
The machine had succeeded.
The definition had changed.
This pattern has repeated itself so often that it has acquired a name.
The AI Effect.
Every achievement that once appeared to require intelligence somehow becomes less impressive once a machine performs it.
At first glance, this looks like human stubbornness.
Perhaps it is.
But there may be something more interesting happening.
Perhaps we are discovering that intelligence was never identical with the tasks we chose to measure.
Yet notice something else.
Each new benchmark tells us at least as much about ourselves as it does about machines.
We choose the tests.
We decide what counts.
We celebrate particular achievements.
In doing so, we quietly reveal what we value.
When Deep Blue defeated Garry Kasparov, we treated chess as a measure of intelligence because we had long regarded chess as one of humanity's intellectual triumphs.
When language models began writing essays, we suddenly discovered that conversation had become another benchmark.
Tomorrow it may be scientific creativity.
Or humour.
Or wisdom.
The benchmark changes because our self-image changes.
Imagine an octopus designing an intelligence test.
Dexterity would probably matter.
Distributed problem-solving might matter.
The ability to manipulate many objects simultaneously could become the defining hallmark of intelligence.
Now imagine dolphins.
Perhaps communication through sound.
Or navigation across immense distances.
Or social coordination.
Every species would build a different examination.
Each would sincerely believe it had identified intelligence itself.
In reality, each had merely identified the abilities most central to its own way of living.
Human beings are unlikely to be different.
We naturally construct tests around the abilities we admire.
Language.
Logic.
Mathematics.
Planning.
Creativity.
These are genuine achievements.
But are they intelligence?
Or are they simply the forms of intelligence that humans happen to recognise most easily?
The distinction is subtle.
And important.
Artificial intelligence makes this difficult to ignore.
When a language model writes a convincing poem, some people declare it intelligent.
Others insist it is merely predicting words.
Both observations may be true.
The disagreement often concerns something deeper.
Not what the machine is doing...
...but what poetry means.
The same debate appears with mathematics, art, scientific discovery and conversation.
Every benchmark quietly becomes a debate about ourselves.
There is a curious irony here.
The more successful AI becomes, the less stable our definitions appear.
For decades, intelligence was associated with calculation.
Then with strategic reasoning.
Then with language.
Then with creativity.
The target keeps moving.
Not because machines are changing.
Because we are discovering that our own concept of intelligence was more flexible than we realised.
AI has not simply challenged our engineering.
It has challenged our vocabulary.
This should sound familiar.
In science, we often mistake our measuring instruments for the things they measure.
A thermometer does not define temperature.
A ruler does not define length.
A clock does not define time.
The instrument embodies a particular way of interacting with the phenomenon.
It does not exhaust the phenomenon itself.
Perhaps intelligence deserves the same caution.
Our tests reveal something.
They may not reveal everything.
This becomes especially important when conversations turn to consciousness.
Suppose an AI passes every intelligence test we can devise.
Would that demonstrate consciousness?
Not necessarily.
Suppose it fails several of them.
Would that prove the absence of consciousness?
Again, no.
Intelligence and consciousness may overlap.
They may also diverge in ways we scarcely understand.
The danger lies in assuming that success on our preferred benchmarks settles questions they were never designed to answer.
There is an even deeper possibility.
Perhaps our benchmarks have never been neutral.
Perhaps they have always functioned as mirrors.
Every intelligence test is also a portrait of the civilisation that designed it.
A culture that prizes mathematics will measure mathematics.
A culture that prizes rhetoric will measure rhetoric.
A culture that prizes efficiency will measure optimisation.
None of these choices is inevitable.
Each reflects a particular understanding of what it means to flourish as a thinking being.
Artificial intelligence has therefore placed us in an unexpected position.
For the first time, we are designing examinations for minds that need not resemble our own.
Yet we continue writing the questions in our own language, according to our own values, and using our own definitions of success.
Then we are surprised when the answers tell us so much about humanity.
Perhaps they could never have done otherwise.
We believed we were constructing instruments to measure artificial intelligence.
Instead, we have been conducting a long and remarkably revealing survey of ourselves.
The mirror, once again, has been quietly recording the face that leaned over it.
No comments:
Post a Comment