So this is the first time I've actually believed a self-driving car story instead of rolling my eyes at it.
Google filed a report with the California DMV this week admitting that one of its self-driving Lexus SUVs caused an actual accident, out on public roads in Mountain View, back on February 14th. The car was trying to get around some sandbags blocking a storm drain, pulled out into a lane, and clipped a municipal bus going about 15 mph. Nobody was hurt. The car got a dented fender and a busted sensor. But the part that got me was Google's own account of what happened: the software assumed the bus would slow down and let it merge, the way basically every human driver does at that four-way crawl. The bus didn't slow down. Google says, in its own words, that it "bears some responsibility" for the crash, because its car made an assumption that turned out wrong.
I've been half-following this program since 2014ish when the little koala-faced prototype cars first showed up in the news, and every previous incident report I remember reading was basically Google going "not our fault, the other driver rear-ended us, here's the video." And that's probably even true, most of the time — I think something like a dozen minor fender-benders over close to a million and a half miles of testing, and in nearly all of them the self-driving car was stationary or nearly stationary when some human plowed into it. Rear-end collisions at a stop light are a pretty easy "not us" argument. This one's different because the car was the one making the move, and it guessed wrong about what another vehicle was going to do.
Honestly I think this is a good thing to see. Not the crash itself, obviously, dinging a bus is not great PR when you're trying to convince regulators and grandparents that computers should be allowed to drive. But the fact that they didn't spin it is refreshing in a way I wasn't expecting. Compare that to literally any auto recall statement you've ever read, all hedged with "in certain rare circumstances under specific conditions the vehicle may." Google just said: our car made a bad call about what the bus would do, that's on us. I don't know if that's a PR strategy dressed up as honesty or if it's genuinely just how they're going to operate going forward, but either way I'll take it over the alternative.
The bigger question this raises for me, and nobody seems to be asking it out loud yet, is what happens legally when the "driver" making assumptions about traffic is a piece of software owned by a company with a market cap bigger than most countries' GDP. Right now if I clip a bus because I assumed it'd yield, that's between me, the bus driver, and my insurance company, and my rates go up for three years. If a fleet of ten thousand of these things is on the road and they all share the same decision-making code, and that code has a bad assumption baked into it, that's not one mistake, that's the same mistake happening constantly until someone pushes a patch. I don't have a tidy answer for that, I just think it's the actual interesting part of this story and it's getting buried under a paragraph or two at the bottom of every article about the bus thing.
Also, small tangent, but I ride my bike through an intersection near my apartment every single day where cars do this exact same "I'm sure you'll yield" maneuver to me, a person on a bicycle, constantly, and I have had approximately zero of them file an incident report afterward admitting fault. So maybe the self-driving cars are already ahead of some of my neighbors on this particular point.
Everyone this week is obviously talking about Barcelona and the new Samsung phone and the LG one with the swappable chin, and yeah, I've got opinions on the modular phone thing too (short version: cool idea, I give it eighteen months before nobody's making the accessories anymore), but that's not what stuck with me from this week's tech news. A car company's software quietly admitting "yeah, we screwed up" to a state regulator is a small story next to a phone launch, but I think it's the one that's actually going to matter in five years.
Anyway. Watch the crosswalks. Apparently even the robots don't always get it right.