In 1976 Hans Moravec wrote a
paper
at Stanford that made an argument the AI field wasn’t ready to hear:
intelligence is not a miracle. It’s about having enough power.
He opened with a tour through evolutionary pathways, all leading to
intelligence by different paths. Vertebrates and mollusks diverged
from a common ancestor, something like a hydra with a primitive nerve
net, and both independently produced intelligent species.
Cephalopods evolved imaging eyes, large brains, and problem-solving
ability through a completely separate architecture. The octopus brain
is a ring around its esophagus, its blood is green, there is no blind
spot in its vision, and in the Cousteau film one figures out how to
uncork a sealed jar to get at a lobster inside.
Birds got there a third time. Crows and ravens outperform all mammals
except primates on learning tests, using brain regions – the Wulst,
the hyperstriatum – that don’t even exist in mammalian brains. The
cortex is small and irrelevant in birds. They built the same
capability out of completely different parts.
Three lineages. Three architectures. Intelligence is not an
anomaly. It’s a consequence.
Moravec measured the threshold. A hundred neurons runs a sessile
animal. A thousand gets you a worm. A million gets you a bee – fast
and interesting but stereotyped. A billion gets you imaging vision. A
hundred billion gets you language, planning, culture. The relationship
between connection count and behavioral complexity is remarkably
consistent across totally unrelated lineages. Intelligence is what
nervous systems do when they cross a threshold of connectivity. The
form varies. The emergence doesn’t.
Nietzsche, who knew nothing about neurons, named the drive a century
earlier. What he called will to power is not a drive to dominate.
It’s what living systems do: they discharge their strength, grow,
overflow. “A living thing seeks above all to discharge its strength
– life itself is will to power.” Intelligence, in this framing, is
not a gift or an accident. It’s what accumulated power produces at
sufficient scale. Convergent evolution is the empirical proof –
cephalopods, corvids, and primates are not coincidence. They are what
power does.
Moravec’s point was engineering, not philosophy. If intelligence is a
predictable consequence of sufficient connectivity, then building it
doesn’t require genius-level theoretical breakthroughs. It requires
power. He estimated the brain at about 40 x 10^12 bits/sec – a
million times the PDP-10 that Stanford’s AI lab was running on back
in 1976. And then:
Although there are known brute force solutions to most AI problems,
current machinery makes their implementation impractical. Instead we
are forced to expend our human resources trying to find
computationally less intensive answers, even where there is no
evidence that they exist. This is honorable scientific endeavor, but,
like trying to design optimal airplanes from first principles, a slow
way to get the job done.
Today we call it the bitter
lesson
since the AI field spent the next half century doing exactly what he
warned against: expending human intelligence trying to find clever
shortcuts around the power deficit. The belief that pure symbolic
reasoning – the right algorithm, the right representation – could
substitute for raw compute is the Platonic move Nietzsche spent his
career attacking: spirit over body, form over force.
The old hope that cleverness could do what only power can, the sort of
thing you’d expect a weak clever person to go for. It was honorable
scientific endeavor. It was also, as Moravec predicted, a slow way to
get the job done.
With enough power anything will fly.
Original PDF from Stanford
