Why “Intelligence Is Getting Free” Misses the Point
From Fei-Fei Li and David Rogier’s Silicon Valley Girl conversation: today’s AI talk mostly means language intelligence—and calling all intelligence “nearly free” is irresponsible.
A popular claim keeps circulating: the cost of intelligence is approaching zero. If that were true, fear would be rational—the industrial revolution automated much physical work; AI seems to be automating mental work. Li’s reply in the interview is blunt: people who say that out loud are being irresponsible.
Most of today’s “AI” is language intelligence
Rogier says one of the most useful things he learned from Li is this: public conversation about AI is mostly about language intelligence. In Li’s words, that understanding is “lossy.” You cannot learn to drive from language alone. You cannot learn to shoot a basketball from language alone.
So we are still closer to AI 1.0 than to a finished picture of intelligence. The field also has real hype. AI does not arrive with its own values; people assign them. That is uncomfortable—and also an opening to design and shape how the tools are used.
The industrial revolution did not erase labor
Li pushes back on a lazy history lesson. The industrial revolution did not automate labor away; it made labor more efficient, scaled it up, and changed labor markets. Assuming that labor somehow contains no intelligence is, in her view, badly wrong.
Physical work, cognitive work, and emotional work are all deeply woven with human intelligence. Nature has not handed us a finished map of how deep that intelligence goes. Anyone who prices the whole thing near zero is skipping the hard part.
Beyond chat: many kinds of intelligence
- Perceptual intelligence — making sense of what you see and hear
- Spatial intelligence — navigating rooms, paths, and 3D scenes
- Embodied intelligence — coordinating a body under physics
- Emotional intelligence — reading people and choosing a tone
- Creativity — a messy blend of brain, life history, and taste
Li readily grants that large language models and their products are powerful—in business analysis, software engineering, and logical reasoning, and in helping people go deeper. The nuance is that much of the value looks like strong collaboration with human intelligence, not a clean replacement.
Why the slogan backfires
She avoids phrases like “automating human intelligence” or “intelligence approaching zero cost” because simplified slogans feed backlash. Headlines about mass layoffs and “we no longer need you” pile up. The answer is still not to hide from the tools—it is to use them carefully, improve them, and keep human agency in the loop.
For readers, the practical takeaway is narrower than the slogan: language AI is getting cheaper and more capable; human intelligence, as a whole, is not a free commodity. Treat today’s models as a powerful slice of the picture—not the whole bill.