Qurated: The AI Superforecasters Are Here

The Machines Learned to See the Future — And We Should Have Predicted This

The most important shift isn't that AI can now forecast the future. It's that AI can now forecast it as well as the humans we once considered irreplaceable — the elite superforecasters who beat CIA analysts with classified data. That moat has evaporated. What you do next depends entirely on whether you understand why.

The Core Insight

Superforecasting was never magic. It was a process: break vague questions into tractable sub-questions, gather diverse evidence, weight it probabilistically, update ruthlessly, and average across perspectives. Humans found this exhausting. Machines find it trivial.

The lesson generalizes far beyond forecasting: any cognitive skill that decomposes into "search broadly, reason carefully, update honestly" is now automatable. The rare human edge lives in the parts that don't decompose — taste, question-selection, and knowing which future is worth predicting at all.

A Mental Model: The Forecasting Stack

Think of prediction as four layers. Ask which layer you actually add value to:

Question framing — What exactly are we predicting, and does it matter? (Still human.)
Evidence gathering — Scraping, reading, aggregating base rates. (Now machine.)
Probabilistic reasoning — Combining evidence into calibrated numbers. (Now machine, often better.)
Decision — Acting under the forecast, owning the consequences. (Still human.)

If your job lives in layers 2 and 3, you are competing with something that never sleeps and never flinches. Move up or move out.

Why the AIs Win

Three unglamorous advantages, each replicable in your own thinking:

No ego. The AI updates the instant evidence shifts. It has no reputation to defend, no prior public prediction to rationalize.
Wide aperture. It considers hundreds of reference classes a human would never surface.
Relentless base rates. It anchors on how often things like this happen, not on the vivid story in front of it.

You can steal all three. The difference between a good forecaster and a bad one was never intelligence — it was discipline against your own biases.

The Actionable Playbook

Adopt the machine's habits manually:

Before any prediction, ask: "What's the base rate for events like this?" Start there, adjust second.
Write your forecast as a number, not a vibe. "Probably" hides your errors from you.
Keep a prediction log with dates and confidence. Review quarterly. Calibration is a trainable muscle.

Redeploy your human edge:

Spend your energy on which questions to ask, not on grinding out answers a model can produce.
Cultivate judgment about what's worth knowing — the one layer AI cannot yet own.
Treat AI forecasts as a committee member, not an oracle. Ask it to argue both sides, then decide.

The meta-move: Run any important decision through an AI forecaster and your own reasoning. Where they diverge, you've found either your blind spot or its blind spot. That gap is the most valuable data you'll get all week.

The Uncomfortable Corollary

If prediction is commoditized, then advantage migrates to action. Everyone will soon have access to roughly the same high-quality forecasts. The winners won't be those who see the future most clearly — they'll be those who act on shared foresight with the most speed, courage, and skin in the game.

Knowing the odds was always the easy part. Betting well was always the hard one. That hasn't changed — it's just been thrown into sharp relief.

One line to keep: When a skill can be described as a checklist, expect a machine to run the checklist better than you. Build your life around the things that resist the checklist.

Sources & Further Reading

The AI Superforecasters Are Here — Astral Codex Ten: https://www.astralcodexten.com/p/the-ai-superforecasters-are-here