The Volkswagen emissions scandal remains in the news. I think the reality of this situation has been misinterpreted, and much of the commentary about it is plain wrong. I am not defending what VW did but I think there are much wider implications, and they need discussion, a point which most of the media neither understand nor are currently exploring.
Firstly, what I think we know. VW systematically have chosen to not meet emission standards in normal usage in a range of their diesel engine cars in multiple geographies. VW’s Chief Executive and his predecessor who resigned over the matter have admitted this. The main problem is with Nitrous Oxide emissions where there are stringent tests in the US, but not in Europe yet. But other emissions are also well out of line with the standards set for normal driving conditions.
I am now choosing my words with care – unlike much of the media who say that VW somehow cheated the test. That is not true. Their software detected the test conditions were in place – essentially a rolling road which is unnatural compared to normal driving characteristics – and set the engine management systems to pass the emission control standards.
All tests generate unnatural behaviour. Think of exams and cramming, revision that sticks in your short term memory for the 24 hours necessary to pass the test, and then gets forgotten. Is that cheating? No, it’s passing the test. If you want a test which shows how much knowledge a student has retained then there should be an unannounced test. Or what about your driving test? Don’t tell me you drive the same every day as in that 45-minute period that got you your licence. These examples are somewhat more benign than the VW example, where it is estimated tons of poisonous emissions have entered the atmosphere. This is cynical cheating on VW’s behalf, and dangerous to health. But they passed the test.
Emission tests are flawed, especially in today’s world. If the testers need a car to pass emission tests under normal driving conditions, then the tests need to be under normal driving conditions. VW’s current software wouldn’t pass that test. The software engineers would need to identify something different, perhaps the presences of an emissions collection device on the exhaust pipe to determine that they were under test – which could be done because the back pressure would increase slightly. This could lead us to a somewhat worrying “arms race” between the testers and the tested – but so far at least, the cynical cheating is confined to VW. A great deal of trust is needed to stop the result of this becoming disproportionate.
The point is that we want software to behave differently under different conditions. That what it’s for. In a legitimate way my car changes its behaviour according to how I’m driving. If I’m slow and relaxed it changes gears early, doesn’t overreact to me pressing the accelerator harder and “learns” how I’m driving today. This can be magnified by a switch which changes the gearbox characteristics between Economy, Sport and Manual. The fuel consumption in different modes can change dramatically – I get over 50 miles to the gallon on motorways in economy; in Manual mode on suitable roads I may get less than 30 miles per gallon (and a lot more fun!).
Fuel economy in cars is well known to be indicative rather than real. As the “simulated urban driving” fuel consumption test is the same for each car tested it helps with comparative figures, but is far from an absolute indicator of consumption. Most drivers get nothing like the quoted mileage from real world driving. Is that cheating?
As software is installed in ever more places the problem exemplified by VW’s approach is going to get worse. Few people can understand the code that is embedded in a car, even with democratisation of IT knowledge, and it will remain a specialism. The software is commercially confidential, and the source of competitive advantage. But this is not just about cars. With the Internet of Things emerging, the opportunity for software to manipulate things where trading standards are involved increases dramatically. Alerts to the manufacturer about failing parts that may still have considerable life left in them comes to mind – the potential scam being on unnecessary maintenance bills. I’m sure there are many more.
We will have to rethink so much in a software defined world. If we don’t address this logically and rationally, the risk is that the “luddites” will pass legislation that restricts so much of the potential benefit in their attempts to control what I hope are a few unscrupulous manufacturers who try to take advantage.