de-biasing
A methodology that corrects systematically optimistic project forecasts by replacing forward projections with the actual average performance of comparable past projects.
As defined in Bent Flyvbjerg on Megaprojects
Where it comes up · 2
- Bent Flyvbjerg on Megaprojects2 claims
- Research Pioneer2 claims
Most important claims · 4 of 4Search all claims for “de-biasing” →
- De-biasing (reference-class forecasting), built on Kahneman's Nobel-winning behavioral economics, ignores planners' optimistic future projections and instead looks at the actual past performance of similar projects, adjusting each variable (passengers, cost, schedule) to match the historical average—e.g., halving a ridership forecast—unless there is a strong, proven reason the new project will outperform, in which case adjusting would 'double-debias.'Bent Flyvbjerg on Megaprojects
- It is impossible to produce an AI system that is unbiased, not for technological reasons but because bias is in the eye of the beholder — different people have different ideas of what constitutes bias, so any system will be perceived as biased by some group no matter how it is tuned.Research Pioneer
- Conventional megaproject forecasting is so complex that anyone who wants to can manipulate the numbers—adjusting a little here and there to produce a desired result—whereas transparent de-biasing techniques are easier for laypeople to understand and make it harder to hide a fishy forecast.Bent Flyvbjerg on Megaprojects
- The ideological lean in systems like Gemini stems not from the political leanings of the designers but from the acceptability constraints of the customer base — a big company cannot afford to offend too many people, so it makes products 'safe', and overdoing de-biasing produces non-factual outputs like Black Nazi soldiers.Research Pioneer