Reasoning About Futures

Scenario thinking does not tell you which future will happen. It does something more useful: it prepares you for several, makes your assumptions inspectable, and tells you what to watch for.

Base rates and reference classes

The strongest correction to a confident forecast is a base rate, how often things like this have happened before. Before asking "will this specific AI capability arrive in two years?", ask "of comparable capabilities predicted to arrive in two years, how many did?" That outside view, anchoring on a reference class of similar cases, reliably beats reasoning purely from the specifics of the case at hand, which tends to be swayed by the vividness of the current example.

The biases that distort technology forecasting

Even with a good method, predictable cognitive biases pull forecasts off course. Knowing their names makes them easier to catch:

• Availability bias. Recent, vivid examples (a striking demo, a viral result) feel more probable than they are.
• Recency and straight-line extrapolation. Assuming the current rate of change continues unchanged, through plateaus, bottlenecks, and S-curves that flatten.
• Hype cycles. New technologies tend to run through inflated expectations, then disillusionment, before productive maturity; mistaking the peak for the plateau produces wild forecasts.
• Motivated reasoning. People and organizations with a stake in an outcome forecast that outcome, a reason to weight incentives, not just arguments.
• Overconfidence. Stated confidence intervals for technology timelines are almost always too narrow; the true range of outcomes is wider than it feels.

Further reading

• Superforecasting: The Art and Science of Prediction, Tetlock and Gardner on what measurably distinguishes accurate forecasters: base rates, updating, and calibrated confidence.
• Gartner hype cycle, a model of how expectations for new technology inflate and then correct.
• Scenarios: Uncharted Waters Ahead, the classic case for mapping multiple futures instead of forecasting one.
• Judgment under Uncertainty: Heuristics and Biases, Tversky and Kahneman on availability, anchoring, and the biases that distort prediction.