if you define driving as a binary “solved” v.s. “unsolved”, then it’s simpler to just say that driving is “unsolvable”. humans have been driving for a century

if you define driving as a binary “solved” v.s. “unsolved”, then it’s simpler to just say that driving is “unsolvable”. humans have been driving for a century and we still have 30,000 traffic fatalities per year in just the U.S.: human-level automation (AGI) doesn’t solve that. sure, you could tune the driving parameters to significantly reduce that count — but not for free and not without creating new (but hopefully smaller) problems. at some point it just mirrors any other unsolved human system, where you’re left debating values and tradeoffs. think the Ford Pinto memo [1] but where the tradeoffs are “do we tune driving parameters to achieve 10 fatalities/year and 10M man-years of travel time per year, or 1 fatality and 11M man-years of travel time?” personally i think the tradeoffs in this article are overblown. 99% of us accept that traffic lights are stupid and will delay every 2am no-other-cars-on-the-road trip by a few minutes. but when it’s a new autonomous system suddenly those delays are to be treated as unsolved problems? don’t get me wrong, i like when people acknowledge that there are things around us that ought to be improved, but the distinction here just seems arbitrary. it’s just some weird status quo bias that seems to carry more weight than i can justify. 1: https://en.wikipedia.org/wiki/Ford_Pinto#Cost–benefit_analysis,_the_Pinto_Memo https://en.wikipedia.org/wiki/Ford_Pinto#Cost–benefit_analys...