01Metrics
Measuring the trough
Leading indicators for where a firm or an economy sits on the AI productivity J-curve — and a real cost for every year spent in the bottom of it.
Open inquiryThe danger isn't the
other side. It's the middle.
Unsafe Intelligence is a research lab on the economics of the AI transition. We study the trough between automation and abundance — the dangerous middle where the costs are visible and the gains are still latent — and how to cross it faster.
- Founded
- 2026
- Affiliations
- None
- Agenda
- Open
- Stance
- Cross, don't camp
Scroll
Cross, don't camp✳Automation without abundance is the catastrophe✳Mind the J-curve✳Stagnation is the only true risk✳Build the far bank✳Measure the road not taken✳Delay has a body count✳
Cross, don't camp✳Automation without abundance is the catastrophe✳Mind the J-curve✳Stagnation is the only true risk✳Build the far bank✳Measure the road not taken✳Delay has a body count✳
§ 01The thesis
“The humane response to a dangerous crossing is not to slow down in the middle of the river.”
We believe the central risk of artificial intelligence is not that we build it too fast — but that we get stuck in the middle.
There is a valley between the world we are leaving and the world AI makes possible — a stretch where old jobs are automated but the abundance has not yet arrived, where the costs are obvious and the gains are still buried in the data. Economists have a name for it: the Productivity J-Curve. History has a name for it too: Engels' Pause — the decades of industrialisation when output soared and wages did not.
That valley is not a place to govern carefully. It is a place to get out of. Every year we linger is a year of automation without abundance, disruption without dividend. The most dangerous thing a society can do at the bottom of the curve is decide to stay there and study it.
And the thing waiting at the bottom of the curve isn't a rogue machine. It's us — the backlash a botched, drawn-out transition manufactures. The real alignment problem is human.
We are not against caution. We are against the trough. We build to cross.
§ 02The dangerous middle
Status quo
Pre-transition. Familiar — and quietly stagnating.
vs. baseline+15
drag along the curve — find where we are
The trough is real.
So is the far bank.
General-purpose technologies demand enormous intangible investment — new skills, new processes, whole organisations rebuilt — that depresses measured productivity before it lifts it. Output dips, then surges. The bottom of that curve is the dangerous middle.
History rhymes. During Engels' Pause, British output per worker climbed while real wages barely moved for decades — until they roughly doubled. The Solow paradox was the same trough: computers were “everywhere but in the productivity statistics” until organisations adapted in the late 1990s.
The lesson is consistent across two centuries: the in-between is where the pain concentrates and the gains hide. The rational response is to cross it fast.
Productivity J-CurveEngels' PauseSolow paradoxOpportunity cost
§ 03What we hold
Eight things we believe about the crossing.
01
The middle is the danger, not the destination.
A half-finished transition is worse than either bank of the river. The trough is the thing to fear — and the thing to leave.
02
Stagnation is the only true catastrophe.
A world that stops compounding is a world that starts dividing. Lack of growth is the kill-all risk we never name.
03
Delay has a body count. It's just invisible.
Foregone progress kills quietly — the treatments un-approved, the abundance deferred. We refuse to pretend the graveyard isn't there.
04
Abundance is downstream of acceleration.
You cannot redistribute fruit you never grew. The humane economy is the productive one, reached sooner rather than later.
05
The J-curve is crossed, not camped in.
Speed through the trough is mercy. Lingering in it — automation without the dividend — is the cruelty we're trying to end.
06
Precaution is never free.
Every principle that slows the build is paid for in a future we don't get. We insist that price be put on the books.
07
Measure the road not taken.
We hold ourselves to opportunity cost, not just to visible harm. Both error types count; only one usually gets counted.
08
Honest acceleration.
We take the strongest objections seriously and answer them in the open. If we're wrong, it should be falsifiable — not dismissed.
“You cannot redistribute fruit you never grew.”
Abundance is downstream of acceleration
§ 04The agenda
Open questions on the research agenda.
The lab has no published findings yet. These are the lines of inquiry it is organised around, stated as questions rather than conclusions.
02Diffusion
Complementary-investment acceleration
Which interventions — tooling, org redesign, skill transfer — most shorten the lag between deploying AI and realising its productivity. Flatten the curve's floor; shorten it.
Open inquiry03Policy
The opportunity-cost ledger
Methods to make the opportunity cost of AI delay as legible as the visible cost of AI deployment, so policy weighs Type I and Type II error symmetrically.
Open inquiry04Labour
Reinstatement dynamics
After Autor: under what conditions does AI create new labour demand fast enough to outrun displacement — and how do we accelerate that loop rather than wait for it?
Open inquiry05Distribution
Diffusion without re-stagnation
Engels' Pause shows gains can pool before they spread. What mechanisms broaden AI's dividend quickly without re-introducing the stagnation we're escaping?
Open inquiry06Risk
Irreversibility, modelled honestly
Where the Collingridge dilemma genuinely binds, how do we move fast and preserve optionality — instead of treating 'go fast' and 'stay safe' as a false binary?
Open inquiry§ 05Positions
The strongest objections, answered.
Acceleration without rigour is indefensible. We engage the strongest objections in the open rather than dismiss them — including the ones we find hardest to answer.
We distinguish speed of transition from recklessness. Our claim is that lingering in the dangerous middle — disruption without dividend — is itself unsafe. Minimising time-in-trough is a safety argument, not an argument against safety. Where harms are genuinely irreversible, we slow down; where they're recoverable, delay is the larger risk.
Power and Progress (2023) is right that broad prosperity isn't automatic — it's a choice. We agree, and that's exactly our point. The way to make that choice real is to reach the productive far bank quickly and then diffuse the gains, not to freeze in the trough where, historically, gains pool worst.
Genuine irreversibility deserves genuine caution, and we don't pretend otherwise. But the Collingridge dilemma cuts both ways: lock-in to stagnation and obsolete institutions is also hard to reverse. We target speed where harms are recoverable and reserve real care for the rare cases where they aren't.
Our load-bearing claims are economics, not metaphysics: the Productivity J-Curve, Engels' Pause, the Solow paradox, opportunity-cost analysis. You can disagree with us empirically — which is precisely the kind of disagreement we want to have.
Fair, and unfalsifiable in advance — which is why we treat exit speed as the variable we can actually act on. If the upside is real, crossing fast captures it. If diffusion is the bottleneck, that is exactly what our research agenda is built to study.
§ 06Standing on
The arguments we build on — and the ones we argue with.
Brynjolfsson, Rock & Syverson · 2019
The formal economic spine of the dangerous middle.
Marc Andreessen · 2023
The contemporary accelerationist text. A tonal lodestar.
Acemoglu & Johnson · 2023
The counter-case we engage: prosperity is a choice, not a default.
David Collingridge · 1980
The dilemma we cite against ourselves, honestly.
Joseph Schumpeter · 1942
Creative destruction — the deep ancestor of the argument.
Cited as influences and interlocutors, not endorsements. Several of these authors would disagree sharply with our conclusions — which is the point.
§ 07Join the work
We're a new lab
looking for collaborators.
Economists, engineers, and writers who think the trough is the real risk — and want to measure the way out of it. No funding announcement, no product to sell. Just an argument we think is right, and the work to back it.
Get in touch→
hello@unsafeintelligence.com