3 ms·

I think the degree to which driving is a social behavior is underappreciated. By social behavior, I mean we learn to read the "body language" of the drivers (c

I think the degree to which driving is a social behavior is underappreciated. By social behavior, I mean we learn to read the "body language" of the drivers (cars) around us, pedestrians, bikers, etc. Have you ever thought to yourself, "that dumbass is just about to pull into my lane" or something like that? That's you being a social animal, making a social judgment about a fellow animal. This is ancient stuff. Nature is full of animals who have no spoken language, and even no verbalization at all, who nonetheless operate together physically to great effect. Think of a herd of buffalo, or a school of fish, or a murmuration of swallows. Now, you want to put a robot in there, and have it keep up with the animals? That's a hard problem. I would argue that we work the same way when driving. So a robot trying to operate in human traffic is going to face the same hard problem. Instead of tackling that problem, self-driving cars fall back on the written rules of the road. But it's crucial to understand that the rules were written by humans for interpretation by humans. They are not a fool-proof algorithm for safe robot driving, and they were never intended to be that. Sure, you can argue that conservative driving is safer. I'd argue that that is a cop-out that simply delegates the argument to how we define "conservative." It's conservative to drive 5mph in a 25mph zone; it's also totally legal. But is it good driving? We can design a conservative robot to put into a school of fish. The real fish will avoid it and leave it behind. The robot fish is safe, in that there are no collisions. But is it fulfilling the mission of being a good fish?

dsfyu404ed 8y ago
>I think the degree to which driving is a social behavior is underappreciated. I regret that I have but one upvote to give. >We can design a conservative robot to put into a school of fish. The real fish will avoid it and leave it behind. The robot fish is safe, in that there are no collisions. But is it fulfilling the mission of being a good fish? Has anyone even built a robot fish that can operate fairly seamlessly in a school of fish? Obviously there's less resources being devoted to this because there's less incentive to do so because there isn't the same demonetization potential as self driving cars but it seems like technology capable of acting like a fish in a school is a prerequisite of a properly self driving car.
leesec 8y ago
Luckily not all self driving approaches are trying to hard-code rules into their algorithms. See comma.ai , they use ML to have the driving agent learn from millions of miles of human driving data.

dsfyu404ed 8y ago

>I think the degree to which driving is a social behavior is underappreciated. I regret that I have but one upvote to give. >We can design a conservative robot to put into a school of fish. The real fish will avoid it and leave it behind. The robot fish is safe, in that there are no collisions. But is it fulfilling the mission of being a good fish? Has anyone even built a robot fish that can operate fairly seamlessly in a school of fish? Obviously there's less resources being devoted to this because there's less incentive to do so because there isn't the same demonetization potential as self driving cars but it seems like technology capable of acting like a fish in a school is a prerequisite of a properly self driving car.

leesec 8y ago

Luckily not all self driving approaches are trying to hard-code rules into their algorithms. See comma.ai , they use ML to have the driving agent learn from millions of miles of human driving data.