4 ms·
Yes, but that won't solve the issue. The issue is regulators write regulations for one thing, but want a subtly different thing. If you and I start a game of
by londons_explore 7y ago
Yes, but that won't solve the issue.
The issue is regulators write regulations for one thing, but want a subtly different thing.
If you and I start a game of poker, but rather than winning through skill, I call the FBI and get them to search all your cards, have I broken the rules of poker? It wasn't a written rule not to call the FBI...
Regulators across the world have similar gaps in the rules, and both humans and automated design tools will start to exploit them, possibly unknowingly. Instead they need to make far more watertight rules. They should simply require all cars to measure their own emissions, and bill the companies for any cars which either mis-measure emissions, or measure more emissions in their lifetime than an allowed limit. Then manufacturers will have an incentive to upgrade cars in the field.
- adrianN 7y agoThat's a good idea, but as far as I know many (most?) emissions are a bit tricky to measure. You need fairly bulky equipment.
- thefounder 7y agoWell we know how self regulation worked for Boeing. Now I could see software that registers less emissions than they actually produce and we are back to square one.
- legulere 7y agoThe problem simply is that most things are very hard to measure directly if not impossible. "When a measure becomes a target, it ceases to be a good measure." https://en.wikipedia.org/wiki/Goodhart%27s_law https://en.wikipedia.org/wiki/Goodhart%27s_law
- mikeash 7y ago“It wasn't a written rule not to call the FBI...” But it was a written rule not to use a defeat device. This isn’t a case where manufacturers found a loophole and regulators got upset because they wanted manufacturers to obey the spirit of the rules rather than just the letter. This is a case of manufacturers blatantly breaking the explicit written rules.
- londons_explore 7y agoSame thing. If I am designing a printer and you say "it must be able to output 10 pages per minute printing black and white text, and make it as light as possible", then I, the printer designer go to work, tweaking the design of the printer to meet that goal, while keeping the weight down. Seeing nothing about the color printing speed, I don't care about that. It's just black and white speed that matters! I make the black cartridge super fast and super wide. Whoa - gotta keep the weight down! Oh well - I'll make this colour cartridge smaller/cheaper/slower/lighter. Suddenly, we have a printer that does 10ppm of black and white, and only 0.1ppm of colour, because it has to crawl along. We now have a 'defeat device'. It performs great in test conditions (black and white text), and badly in general use (colour pages are common). Did the designer explicitly make a defeat device? Not really... They just only made it work well in the test conditions, and performance in all other conditions suffered.
- pretendgeneer 7y agoThat is nothing like what these devices in the car's are doing. The defeat devices are if you can only design a printer that can print 1ppm of arbitrary pages, but you manage to tweak the printer so you can print 10ppm of the test pages. Then claiming you have the best 10ppm printer.
- mikeash 7y agoThe law says you can’t make a device that causes the vehicle to perform well on the test but perform poorly in other routine scenarios. With your example, you’d be stuck arguing to a judge that color printing is not a normal part of your device’s operation and therefore your loophole exploitation is legal. Which isn’t going to get anywhere at all. The law is not stupid nor is it evaluated by machines.
- the8472 7y agoEither the spirit of the rules already does matter or the regulations were written in a resilient way. Otherwise the companies would have gotten away without any fines or recalls for their defeat devices. Even game-theoretically perfect regulation does not work if they're violating the rules.