Forecast Like a Weather Report

The honest, useful way to forecast isn't a confident date — it's a weather report: a few futures, rough odds, and what to watch.

The Future of Technology · Lesson 5 · 7 min read

Two pundits go head to head about a fast-moving technology. One declares, “Self-driving cars will be everywhere within three years — guaranteed.” The other shrugs: “Eh, could happen, could not.” The confident one is precise but probably wrong, with no fallback when reality differs. The vague one can never be wrong — and is therefore useless. Then your boss turns to you and asks what the company should actually do. Neither extreme helps. How do you forecast a genuinely uncertain future in a way that is actually useful? Hold that question — by the end you should have a format that beats both.

A point prediction is almost always wrong — and brittle

The future is noisy (Lesson 2), shaped like an S-curve whose phase you cannot pin down (Lesson 3), and gated by adoption friction you cannot time (Lesson 4). Against all that, a single exact claim — “X by year Y” — is nearly certain to miss.

And missing is not even the worst part. A point prediction is brittle: the moment reality lands nearby-but-different, it gives you nothing — no partial credit, no guidance for what to do instead. Precision is not accuracy. A confident date does not handle uncertainty; it just hides it.

Think in ranges and scenarios

Instead of one future, sketch a few distinct ones — say a fast, a middle, and a slow path. For each, ask two questions: what would have to be true to drive it, and what early signal would tell you you're in it? Then attach rough probabilities — even a gut “60 / 30 / 10” is far better than silence.

This turns a guess into a map. You can prepare for more than one outcome, and you know exactly which tells to watch for as the future picks a lane. Humility stops being a hedge and becomes a tool.

Forecasting self-driving cars over the next several years.
• Useless (precise): “Most cars will be self-driving in three years.” Brittle, probably wrong, no fallback.
• Useless (vague): “It’ll happen eventually, or maybe not.” Unfalsifiable, no help.
• Useful: “~20% fast — robotaxis in many cities (driver: a safety+cost breakthrough; signal: rapid expansion past the pilot cities). ~50% gradual — common in limited zones, rare elsewhere (driver: steady-but-bounded improvement + patchy rules; signal: slow city-by-city rollout). ~30% slow — still mostly pilots (driver: a hard long-tail safety problem; signal: stalled expansion). I’ll recheck in a year and shift probability toward whichever signal shows.” Same future — vastly more useful.

A good forecast is falsifiable — and updated

Two opposite traps. False precision: the brittle confident date. Empty vagueness: “could go either way,” which can never be checked and guides nothing. The cure for both is the same discipline: state probabilities, name what would change your mind, set a checkpoint, and actually update when signals arrive.

The gold standard is calibration: your “70% likely” things should happen about 70% of the time. A calibrated forecaster who keeps score and adjusts beats a confident one who never does. This is the whole ethos in miniature — hopeful, specific, and willing to be wrong out loud.

Your boss asks: “Will this technology take over our market in the next few years?” Two bad answers and one good one:
1. “Definitely, within two years.” — precise, brittle, probably wrong; if it misses, the company has no plan B.
2. “Impossible to say.” — unfalsifiable; the boss learns nothing and can act on nothing.
3. “Three scenarios: takeover (~25%), partial adoption (~50%), stalls out (~25%). Here is the driver and the early signal for each, and I’ll report back in two quarters with updated odds.” — now the company can hedge across all three, watch the right signals, and correct course as evidence lands.
Only the third answer is something you can actually use to make a decision.