Hacker News Reader

Comment by valuearb | Hacker News Reader

This is not a problem with capitalism.

This is not a problem with regulation.

This is a problem with humans. Some will always try to break the rules, and take risks, no matter what the system. The best systems are the ones that recognize this and have ways of self correcting.

In this case if there was a magical device that measured actual pollutants in every car sold, through out it's life, and could be easily and inexpensively checked annually so that the owner had to pay a tax on excess pollution (or even better, a market price for total pollution), the regulation would be easy.

Automakers could claim whatever they wanted, rig whatever tests they chose, but it would become very clear in real world usage that they had done so. And their brands would be murdered if they got a reputation for lying.

View on HN

Of course it's a problem with capitalism and regulation. The particular form of capitalism we have today provides a strong financial incentive to concentrate on short term returns. And the particular form of regulation we have today created an environment where automakers felt there were few consequences, if any, for gaming the emissions system. These cheating programs didn't spring up de novo all of a sudden, they came as the result of years of incrementally gaming the system more and more with no regulatory consequences. Particularly in Europe automakers have been tweaking the vehicles they send for emissions to improve performance in tests. It's easy to rationalize this sort of thing as just "making sure" that the emissions results are accurate and reflect the best the vehicle can do. But once you get into that mindset it's easy to fall into the trap of pushing the envelope on gaming the system more and more. And once you do something like VW did of building different emissions behavior into the vehicle software you can create an enormous divergence in emissions between the test state and the regular operating state.

Sure, other systems can have similar problems, but that's no excuse.

I would love to see the manufacturer taxed on the difference between advertised and real-world measured pollution (perhaps exponentially, if they emit more pollution than they advertise -- this would greatly incentivize conservative accuracy in fuel economy and emissions specs).

But no matter what scheme we propose, I do worry that this would become an adversarial "gaming" situation a bit like hackers vs DRM -- no matter what the rules/criteria to reduce dangerous emissions, manufacturers will employ people specifically to circumvent them in any and all ways possible. This isn't to say we should stop trying to curb emissions -- just that they must be ongoing and open to change and adaptation to new "attack vectors".

Genuinely interested in what you think the difference between a system that "recognizes this and has ways of self correcting" and "regulation" is...

If you have to pass some standardized tests, the engineers will do what they're supposed to: Meet the goals.

I wouldn't call it 'rigged tests', it's more like they optimize the system under test to meet the spec. It probably happens everywhere, not only in cars. The behavior of your fridge, AC or any other hardware is probably optimized in weird ways just to meet Energy Star etc.

>I wouldn't call it 'rigged tests', it's more like they optimize the system under test to meet the spec.

You would do so wrongly. In these cases, like VW, the car is programmed specifically to detect when it is being tested, and lower emissions. They aren't optimizing to test spec, they are cheating. There's no other word for it.

In the auto industry, it's called a "defeat device", and has a long history of causing woe to car companies who use such devices.

Since when? And which long history? Seriously, I don't know, where can I read up on this?

Wikipedia has a list of a bunch of incidents going back to the 70s, in which the phrase "the largest fine to date" appears often: https://en.wikipedia.org/wiki/Defeat_device

No, he's right. That's "optimize for the test spec". Detect an idiom and optimize it is very common in this situation.

For example, commercial compilers are often written to specifically detect and rewrite idioms/code in oft-used benchmarks (that aren't common to the real world), etc Your browser's javascript engine almost certainly does it.

They aren't cheating, they are just making the test worthless by optimizing it. They still shouldn't do it, but it's super-common.

Exactly -- what we need are better tests that aren't as easy to game. The idea of installing a real-world pollution measurement device to each car is still valid though -- that would just become the new test.

In recent cases, these defeat devices have all been for NOx. NOx is reduced in the catalytic converter, unless there is too much oxygen in the exhaust. The defeat devices turn off behavior that limits exhaust oxygen.

Every car comes with an oxygen sensor, which could be used to infer how much NOx is being emitted. So, the solution may actually just mean a standardized little microcontroller that reports directly through the ODB port on how much oxygen has been in the exhaust. If it's sufficiently simple it shouldn't be beatable.

You don't really want to have to plug something into the ODB port.

I have been a member of the ISO committee that standardized ODB-II and related stuff. I remember the representative of one OEM suggesting that they expected at least California to require emissions testing in real-world conditions at some point. This was about 15 years ago.

There was the expectation that in the future the vehicle bus would have a short-range wireless gateway so that you can link the brakeing systems together of cars in a road train. With this in place a country could set up roadside boxes to read out the current emissions values.

With the change in usage of mobile phones over that time you could pick a different mechanism today.

I guess you still have the issue of potentially needing to prevent cheating, the OBD-II PID values are in a sense "translated" from the OEM specific data on the CAN bus.

This is a problem with humans. Some will always try to break the rules, and take risks, no matter what the system.

A deeper point to ponder: Is that really a problem, or is it just life and the way it should be? Would trying to "fix" this "problem" only result in a more dystopian, authoritarian society with more government control?

This is probably going to sound quite controversial, but I've never been a fan of such emissions regulations. The argument that emissions effects everyone is true, but IMHO besides the point: After all, the best way to not "pollute" the planet is to just not live at all.

I do have an interest in car modding/tuning so I may be slightly biased... but to summarise, I'd much rather die from pollution than live in a society which doesn't have any personal freedom anymore.

/rant

You might be fine dying from pollution, but what about the rest of us who aren't?

If you were only affecting yourself, most people would be happy to let you. But you're not. You're taking down the rest of us with you.

Regulating this seems, well, like a regulation problem.

We already have something similar for odometers and I believe some states assess taxes on miles traveled.

Yea, that's a good comparison.

But odometers aren't magical, they can be turned back. There was a recent scandal/lawsuit where an ex-Ferrari mechanic claimed that Ferrari dealerships were equipped with tools to roll back mileage for their best customers. Example being as the ones who bought Ferrari's elite million dollar cars, so they could drive 10,000 or 20,000 miles and roll them back to 200 miles at sale to sell them "as new".

So regulation still requires a way to detect false measurements at the customer level if it becomes too prevalent.

Unless you're going to trust a TPM style chip to sit the reading continuously and/or broadcast them in real time for external audit, you have that same problem.

Pff no need for that. Worm devices are enough for the task and can be audited openly

> In this case if there was a magical device [..] the regulation would be easy.

You'd also need magical devices that force the people who do the readouts and judge cases based on those readouts act honestly, and so on. Why are we talking about magical devices again?

Cant wait for the realtime GPS tracking reported to a central authority too, to make sure nobody is cheating, of course.

That's silly. You don't need to track every single car, just a reasonable number to get a cross-section. I'm sure that randomly selecting people and offering them money to mount a detector on their tailpipe would be enough. No need for hyperbolic mass tracking.

Reply on news.ycombinator.com