3 ms·
I see this rather differently. In fact AI is already as good as human drivers, statistically. Only a few years back, we could point at people getting killed b
by ajross 5y ago
I see this rather differently. In fact AI is already as good as human drivers, statistically. Only a few years back, we could point at people getting killed by software bugs like the Tesla path-under-semi-trailer issue (not a lot, but it did happen and they were real bugs).
That's not happening any more. All we have left is laughing at stuff like this, where the visualization (not even the autopilot!) gets confused by seeing real (!) traffic lights on a truck, so it paints them in space, but then has to re-recognize them because they are moving.
At some point, the luddites will just run out of ammunition. It's sort of happening already.
- lstamour 5y ago> AI is already as good as human drivers But the AI drives slowly and gets confused easily. Regular drivers routinely have to go around self-driving cars. Not to say they won't improve, but it seems like current AI is assistive to the point where it might be harmful when drivers rely on it in speeds and situations where they shouldn't. I'm sure it will keep improving, but I feel like this is one of those situations where the amount of data and training required, and the amount of iteration on the software required to handle edge cases is not impossible but is exceptionally difficult.
- ajross 5y ago> But the AI drives slowly and gets confused easily. Probably better to say that the AI drives slower than you would and gets confused differently. Real drivers die every day to terrible mistakes that look obvious in hindsight. Most of them probably thought AI was terrible too. The bar here is much lower than you think it is, and frankly existing automation has already crossed it.
- visarga 5y ago> the amount of iteration on the software required to handle edge cases is not impossible but is exceptionally difficult. If you want to fix this one exception (false positive) you will introduce unwanted false negatives. That's how precision and recall work, there's a trade-off. So I am not sure it's useful to fix it, how many times does this happen? Will fixing it introduce more frequent bugs?
- cvak 5y agoDo you have source for the statistics? I remember reading somewhere that it was just a number one of the AI projects said, with no verification, also with no knowledge of the setup.
- akersten 5y ago> where the visualization (not even the autopilot!) gets confused by seeing real (!) traffic lights on a truck, so it paints them in space, but then has to re-recognize them because they are moving. Just my hypothesis: but I think the autopilot really did see them as traffic lights, and just got lucky that they weren't powered and ignored them as out of order. Were there a cross street, I suspect the car would have stopped and treated it as an uncontrolled intersection...
- ajross 5y agoJust as with real brains, "the autopilot" isn't a unitary thing. Different parts do different things. The ML that recognizes objects is different from the software that makes decisions as to which recognitions to honor. In this case, all we have is evidence for the first half: the vision system detected (correctly!) images of traffic lights, which then got presented by the UI visualization (correctly!) in the right place. But the software model is that traffic signs are static, and these were moving, so the visualization had no way to present that. It just left them there until the ML told them they weren't there anymore. What would the AP have done? We don't know. But I don't know why we should simply assume it would have done the wrong thing. Driving is filled with false-positive indicators that it already knows to ignore successfully.
- nuker 5y ago> Only a few years back, we could point at people getting killed by software bugs like the Tesla path-under-semi-trailer issue (not a lot, but it did happen and they were real bugs). That's not happening any more. Will you bet with your life on it?
- XargonEnder 5y agoThe only way to know if I will for sure is to give me a Tesla and see what I do...