TECHNOLOGY
In 2016, industry leaders predicted that fully autonomous vehicles would be commercially available by 2021. In 2026, you still can’t buy one.
This isn’t a minor miss. This is one of the most consistent, confident, and expensive mispredictions in the history of technology. And it contains lessons that apply to AI timelines more broadly — lessons the industry is not, in my view, taking seriously enough.
Why It’s Harder Than It Looked
The early benchmarks for autonomous driving were impressive. AI systems outperformed humans on structured tests, handled highway driving smoothly, and navigated predictable urban environments reliably. The extrapolation seemed reasonable: if it’s this good already, full autonomy is just around the corner.
What the benchmarks missed was the long tail of edge cases. A human driver has roughly 16 years of lived experience before they sit in a car. They’ve watched rain, navigated building sites, followed the gestures of traffic police, understood the social dynamics of a four-way stop. Encoding all of that into a system is harder than the early demos suggested.
“The difference between 99% reliable and 99.99% reliable isn’t 0.99%. In a safety-critical system, it’s the difference between acceptable and unacceptable.”
The Lesson for AI Generally
The autonomous vehicle story is a parable about the difference between performing well on a benchmark and performing reliably in the real world. AI systems tend to excel in conditions similar to their training distribution and fail in novel situations. The more safety-critical the application, the more the tail matters — and AI tails are long.
This doesn’t mean AI timelines are always over-optimistic. Sometimes they’re under-optimistic. But it does mean we should weight bold capability claims more sceptically in domains where edge cases carry serious consequences.
Tags: Artificial Intelligence • Opinion • Technology & Society • 192.168.1.22/