Those of us who grew up around early personal computers
inherited a fairly clear picture of what computers were and were not.
Machines followed instructions. Humans interpreted them.
Machines executed procedures. Humans navigated ambiguity. The distinction felt
obvious, almost physical, like a wall you could lean against.
LOGO turtles dutifully traced geometric figures across
screens. Around the same time, Julio Cortázar was reminding us, with his usual
mischief, that even human beings occasionally needed instructions for climbing
a staircase. Both images belonged to the same intellectual world: a world in
which following instructions and understanding them appeared to be very
different things.
For decades, that map of intelligence felt stable enough.
Computers were fast, literal, obedient, and occasionally infuriating. Humans
were slow, interpretive, contradictory, and capable of meaning. The division
was not perfect, but it was reassuring.
Then, at some point, I found myself discussing Sartre with
ChatGPT at two in the morning and occasionally encountering observations that
forced me to pause.
The surprise was not that a philosophical mystery had been
solved. It was not that the machine had become human, conscious, or possessed
by the ghost of a Parisian existentialist with good Wi-Fi. The surprise was
simpler and stranger: some responses displayed a degree of subtlety,
sensitivity, and finesse that I had long associated with the better forms of
human thinking.
The old map began to wobble.
Language Was Never Just Decoration
In some respects, I was better prepared than many people for
this development. My background is in linguistics, and I have long belonged to
the camp that sees language not as a superficial layer placed on top of
thought, but as part
of the machinery of thought itself.
Chomsky, Wittgenstein, semantics, pragmatics: the details
matter, but the underlying intuition matters more here. Language is not merely
a vehicle for expressing ideas. It helps organize them. It shapes attention,
abstraction, inference, memory, social meaning, and the categories through
which experience becomes thinkable.
If that intuition was even partly correct, it was reasonable
to expect increasingly capable language systems to display increasingly
sophisticated forms of cognitive behavior. Improvements in language would not
remain neatly confined to language itself. They would spill over into
reasoning, interpretation, synthesis, analogy, and the kinds of judgment that
language helps coordinate.
Even so, I underestimated the magnitude of what followed.
Not because these systems have become human. Not because the
hard questions about consciousness, intentionality, embodiment, or
understanding have suddenly disappeared. They have not. But because the change
does not feel like a simple extension of the LOGO era. It feels like a
departure from it.
The Familiar Surprise
Many people have now had some version of this experience.
The details vary. Sometimes the surprise appears in translation. Sometimes in
writing. Sometimes in coding, teaching, legal analysis, image generation,
research, or creative brainstorming. The specific territory matters less than
the recurring sensation: machines are operating competently in places where
many of us did not expect to meet them.
That does not mean the systems are flawless. They are not.
They hallucinate, flatten, imitate, overconfidently improvise, and sometimes
produce nonsense with the calm authority of a mediocre consultant. But the
failures do not erase the shift. A bad answer from a system capable of
producing many good ones is a different object from a dumb machine doing
exactly what it was told.
This is where the inherited categories begin to strain. We
keep trying to describe a new phenomenon using distinctions built for an older
one.
For a long time, the line seemed easy: computers execute;
humans understand. Then machines became better at tasks that looked
uncomfortably close to interpretation. So the line moved. Fine, perhaps they
can execute and imitate, but humans create. Then machines began producing work
that, whatever its metaphysical status, entered creative workflows. So the line
moved again. Fine, perhaps they can generate, but humans possess meaning,
authenticity, consciousness, emotion, nuance.
The moat remained. Only its location changed.
Human Exceptionalism To Go
At some point, the process began to resemble a customer
reassurance campaign.
Human Exceptionalism® now comes in several premium
formulations: Meaning, Emotion, Consciousness, Authenticity, Nuance. New
Advanced Nuance Complex. Same comforting promise: do not worry, we found
another thing machines do not have.
This is funny because it is not entirely false. Some of
these distinctions may be real. Some may be profound. Consciousness is not a
trivial matter. Meaning is not a marketing garnish. Embodiment, mortality,
desire, social life, history, and lived experience all matter. The point is not
that human beings are “just machines” or that every distinction collapses the
moment a chatbot writes a decent paragraph.
The point is that the frantic search for a final difference
has become a bad habit.
Every time a machine crosses an old boundary, we announce a
new one with fresh confidence. The new boundary may be interesting. It may even
be defensible. But the pattern itself deserves scrutiny.
The Human Exceptionalism Moat is rarely called into question. There is always some new thing that only humans can supposedly do. The only thing that changes is that the moat keeps moving closer to the castle.
The Wrong Comfort
I understand the desire for a stable human essence. Recent
developments are genuinely difficult to assimilate into the map many of us
inherited. Questions about employment, education, economic transition, and
human dignity are real. They deserve serious attention and may ultimately
matter more to everyday life than debates about consciousness.
But those concerns do not require us to invent a new
uniquely human essence every time an old distinction becomes difficult to
defend.
The danger is not that philosophy enters the conversation.
Philosophy should enter the conversation. The danger is using philosophy as
emotional insulation: a way to reassure ourselves that nothing fundamental has
changed because somewhere, behind the latest boundary, there must still be a
protected inner chamber labeled Human.
Maybe there is. Maybe there are many. But if the point of
the argument is simply to preserve the feeling of superiority, it will keep
shrinking as the systems improve. That is not analysis. That is bunker
maintenance.
A better question might be less defensive: not “What can
machines never do?” but “What should human beings choose to become, build,
protect, and value now that some capabilities are no longer organized the way
they used to be?”
On AI, Uncertainty, and the Grace of Not Knowing
The age of LOGO gave us a clean distinction: the machine
follows instructions; the human understands. It was a useful distinction for
its time. It helped generations think about computing, language, education, and
agency. But it no longer explains enough.
We may not yet have the right vocabulary for what is
happening. We may not agree on the right questions. That seems acceptable. New
realities often arrive before the language for them does. The first drafts are
usually awkward. History is not known for sending polished memos.
What seems less useful is the increasingly frantic effort to
redraw the moat every time an old boundary becomes uncomfortable.
Human value may not be best defended by locating one last
exclusive property and guarding it with philosophical sandbags. It may be
better understood as something we build through relationships, institutions,
choices, responsibilities, creativity, memory, care, and forms of collective
life that cannot be reduced to a checklist of capabilities.
The moat keeps moving. Perhaps that is the signal.
Maybe the task is not to find the final line machines can
never cross. Maybe the task is to stop confusing the defense of an old map with
the work of understanding a new territory.
