The popular way to dunk on Ray Kurzweil: he grades himself 86% correct; independent volunteers who blind-scored 172 of his statements landed near 42%. Case closed? No — that scorecard is real and beside the point. It's grading the wrong exam.
Pinning an exact year on a technology 20 or 30 years out is nearly impossible — and it isn't the useful part. The useful part is getting three things right: the direction (where is this going?), the sequence (what comes before what?), and the rough magnitude (how big, how fast?). Nail those and you can position for the future even if you're off on the calendar.
Here's the fair standard, the one we'd apply to any forecaster: if he's within 2–3 years on something decades away, that's a hit. Five years off isn't a disaster either. What matters is whether he saw the order of events coming. Judge the compass, not the calendar. By that standard, Kurzweil hasn't been a coin flip — he's been one of the most accurate big-picture forecasters alive.
Fair challenge, and worth answering head-on — because loosen the criteria enough and any forecaster looks like a genius. So here's the line: "direction and sequence, within a few years" is still a bar you can fail. Kurzweil could have been wrong about the direction — he could have predicted an endless AI winter, or that computing power would plateau, or that machines would never hold a real conversation. Plenty of serious people did exactly that. He didn't. He called the slope, the order, and the rough size — and you can check each against reality. That's not an unfalsifiable dodge; it's grading at the resolution that actually matters. The date is the noise. The trajectory is the signal.
The cost curve. His core claim — that the price-performance of computing falls on a smooth exponential — has held for forty years. Compute per dollar has dropped on the order of ten-trillion-fold since the 1940s. Genome sequencing didn't just follow that curve, it beat it: ~$95 million in 2001 to a few hundred dollars today. When he drew those lines in the 1990s and was laughed at, the lines were right.
The sequence. He called the order: computing gets absurdly cheap → narrow AI beats humans at one thing at a time → the capability becomes broad and conversational → it starts to feel less like a tool and more like a collaborator. That's the last decade, roughly on his arc.
The texture of exponential change. His deepest insight isn't any single prediction; it's that exponential progress feels boring for years and then arrives all at once. Anyone who watched AI look like a toy in 2018 and a coworker in 2025 has lived that. Being right about how a technology arrives — slow, slow, slow, sudden — is worth more than being right about the month.
We're not here to canonize him. The honest misses cluster in two places:
That second one is the whole lesson in miniature: trust his direction, distrust his named technology and his exact date.
A method only matters if you can use it forward, so point the compass at his two loudest current predictions — AGI around 2029 and "longevity escape velocity" in the early 2030s. Don't ask "is the date right?" Ask the compass questions:
So the honest forecast writes itself: lean toward his AI direction, heavily discount his biology timeline, and in both cases watch the bottleneck — compute and energy for the first, the slow grind of clinical trials for the second. You never needed his calendar to position around his compass.
This is where a futurist's batting average stops being trivia and starts being useful:
The rarest thing in forecasting isn't seeing when the future arrives — it's seeing its shape early enough to position for it. Grade Kurzweil on the calendar and he's a coin flip. Grade him on the compass, and he's been pointing the right way for forty years. We'll take the compass.
Educational, not investment advice. Dragonfly Lens reads the curves, not the hype. We are not a registered investment advisor.