What is AI?

Everyone reaches for an analogy. The car? The printing press? Electricity? The bomb? Which one we pick quietly decides what we worry about — and what we miss.

Explore the four analogies

AI is compressing history

Each bar spans the years from a technology's debut to broad adoption; the dots mark key dated milestones. The printing press took centuries; AI, a sliver. Hover any bar for the figures.

Every earlier revolution gave society decades — even centuries — to adapt. AI gives us years.

And it keeps accelerating

Time to reach ~50% adoption. Each technology rides the last one's infrastructure — so diffusion has collapsed from decades to weeks.

Car
~40 yrs
PC
~20 yrs
Broadband
~13 yrs
Smartphone
~6 yrs
AI (ChatGPT)
~2 months

By the numbers

The same five technologies, scored on dimensions you can actually measure — so the comparison rests on data, not vibes.

Dimension
Printing press
Electricity
The car
Nuclear
AI
Adoption speedtime to broad reach
decades → centuriesestimate
~30 yrs to 10% of homesmeasured
~40+ yrs to 50%measured
decades (regulation-capped)estimate
100M users in ~2 monthsmeasured
Output / cost step-changeproductivity or cost jump
~100–1000× throughputmeasured
economy-wide (diffuse)
discrete sector
early drafting / coding gainsestimate
Infrastructure dependencynew physical buildout needed
low
high
high
high
low
Reversibility / lock-incan society back out?
high
high▼ stalled 1929
high▼ 59%→50% in 1930–33
medium
high
Catastrophic-misuse potentialworst-case severitynegligiblelowmediumexistentialhigh
Economic footprintmeasured vs. projected
book cost ~10×↓estimate
pervasive input (diffuse)
~4.9% of US GDP (peak)measured
~10% of world electricitymeasured
$2.6–7.9T / yr potentialprojected

The units aren't comparable (books vs. households vs. users vs. dollars), so there's no single “impact score.” AI's economic figures are projections, not realized output — and adoption isn't always one-way: car and home-electricity penetration both fell in the early 1930s.

Don't pick one analogy. Map them.

Each invention is strong on one dimension of AI's impact and weak on others. Compare them side by side.

The CarIrreversibility & social reorganization
The Printing PressInformation, knowledge & cognition
ElectricityGeneral-purpose substrate & economic impact
Nuclear TechnologyMoral hazard & governance
Captures wellOnce adopted, it restructured physical and social life — suburbs, commuting, dating, retail — and couldn't be undone.Mechanized the production of knowledge, collapsed costs, flooded society with mixed-quality content, and destabilized the gatekeepers who had controlled it.Not a product but a substrate quietly embedded into everything, with productivity gains showing up only decades later once factories and homes reorganized around it.Catastrophic potential, builders’ regret, and the explicit "shape it from inside vs. refuse to work on it" debate argued by the scientists themselves.
Where it breaksThe car never improved itself. It did one thing — transport — and stayed there.The press didn't act autonomously — it copied what humans set, nothing more.Electricity had no agency or intent — it powered systems but never decided anything.Raw capability alone never determined the outcome — governance and norms were always the real variable.
Lesson for AIThe deepest effects are second-order and arrive slowly: suburbs, oil geopolitics, ~1.3M annual deaths dwarf the obvious first-order use.Disruption lands hardest on incumbent gatekeepers, and adaptation is generational, not immediate. It took centuries for norms and literacy to catch up.Be patient about the productivity payoff. Complementary reorganization — not the raw capability — is what unlocks the value.Who controls it, and under what norms, can outweigh the technology itself. 'Shaping from inside' is a real but contested strategy.

No single analogy is right on every dimension — “where it breaks” is the honest column.

The open future

We can't rewrite history — only shape what comes next.

nowHistory — already writtenThe future — ours to shape

Every analogy here is a technology whose story is already written: the car, the printing press, electricity, the bomb. We know how they turned out. AI's story isn't written yet.

We're still at the very start of its curve — before the norms, the guardrails and the habits have set. That's the part we still get a say in. The question was never really which analogy AI resembles. It's which future we decide to build with it.