Description
This market addresses the central question in autonomous vehicle (AV) safety. Waymo is accumulating millions of fully autonomous miles, but experts like Professor Philip Koopman of Carnegie Mellon argue that statistical proof of safety requires billions of miles of data.
The baseline for human drivers in the U.S. is approximately 1.35 fatalities per 100 million miles. This means that over a 2 billion-mile span, human drivers would be expected to cause roughly 27 fatal accidents. This market asks whether Waymo can achieve a safety record that is an order of magnitude better than this human baseline according to Koopman's framework for safety. His framework describes "10 critical mishaps" in 2 billion miles, but "critical mishap" is hard to adjudicate, so I'm using fatalities.
Key Definitions
"2 Billion Autonomous Miles": Refers to cumulative miles driven by the Waymo fleet in fully autonomous mode on public roads, as officially announced by Waymo or its parent company, Alphabet.
"Fatality": The death of any person (Waymo occupant, occupant of another vehicle, pedestrian, cyclist, etc.) resulting from a collision involving a Waymo vehicle.
"At-Fault": The collision is determined to be the fault of the Waymo vehicle by the official, final report from a relevant government authority (e.g., NHTSA, NTSB, or the primary law enforcement agency that investigated the incident).
Resolution Criteria
This market will resolve to YES if, upon reaching 2 billion cumulative, fully autonomous miles, the total number of at-fault fatalities involving Waymo vehicles is three (3) or fewer.
This market will resolve to NO if, upon reaching 2 billion cumulative, fully autonomous miles, the total number of at-fault fatalities involving Waymo vehicles is four (4) or more.
Additional Terms
The closing date for this market is January 1, 2040. If Waymo has not officially announced reaching 2 billion miles by this date, the market will resolve to N/A. Resolution will be based on a preponderance of evidence from reputable news sources and official government reports.
1,000According to Gemini (3.1 Pro), as of Feb 2026 Waymo had 200M miles driven and zero at-fault fatalities.
Just as a matter of pure statistics, 4 fatalities in 2B miles is 0.4 fatalities in 200M miles. Since we can only observe integer numbers of fatalities, fatality data alone is not enough to be very confident either way. (Well, we can be fairly confident it's not going to be >40 fatalities in 2B miles, if the distribution of rides doesn't change much.)
One obvious other thing to look at is just accidents (since fatalities require accidents). Here the best I could find was that by May 2025, in the Phoenix area, Waymo had been involved in 71 accidents, and been at fault in 9 of them (13%). This is on ~40M driven in Phoenix up to that point (estimated with various AIs). So extrapolating linearly would give 450 at-fault accidents at 2B miles.
According to various AIs, 0.6% of accidents lead to fatalities. So we expect 2.7 at-fault Waymo fatalities at 2B miles, which makes 4+ appear moderately unlikely. (If modeled with a Poisson distribution, 29% chance of 4+.)
Of course, the distribution of Waymo accidents is likely different (less serious) from the overall distribution, bringing the estimate down. On the other hand, Waymo has so far chosen locations specifically for how "easy" they were to do, and it has not done much freeway driving. Whereas the most likely way for them to reach 2B miles is to expand into less easy areas, and drive on freeways.
Finally, there is the correlation aspect. Waymo cars all share the same software. This kind of thing makes tail events more likely, as now we don't need several independent factors to all go the same way. So I find it hard to justify having this market be >60%.
@DavidFWatson I might be reading something incorrectly but I think the numbers in the title and resolution criteria are different