BETA
This is a BETA experience. You may opt-out by clicking here

More From Forbes

Edit Story

EPA Charges Against Fiat Chrysler Put Software In The Legal Spotlight

Following
This article is more than 7 years old.

The Environmental Protection Agency’s charges against Fiat Chrysler Automobile raise a question: Can code be illegal, or just the actions of the coders?

In accusations that echo the ones that have cost Volkswagen more than $15 billion and led to the arrests of two executives so far, the EPA says Fiat Chrysler buried illegal code in pollution control systems on more than 100,000 Jeep and Dodge Ram diesels to allow excess emissions of nitrogen oxide.

The EPA’s charges against Fiat Chrysler are more muted than the ones against VW. Instead of accusing the company of outright fraud, the EPA says Fiat Chrysler “failed to demonstrate” that it didn’t know, or should have known, that its software violated pollution regulations. Fiat Chrysler denies wrongdoing, saying its emissions systems “meet all applicable regulatory requirements” and it has been working with the government on software fixes.

The difference between the two cases highlights how software fraud is really no different than any other kind of fraud. What got VW into big trouble wasn’t necessarily the code, but how executives hid the code from regulators. That behavior indicates intent, and intent is critical to any criminal prosecution. Where this will get interesting is with artificial intelligence and machine learning, when no one can be accused of writing code since the machine writes the code itself.

With current emission control systems, however, the code is still written by engineers and can be examined by the government. While the EPA’s accusations against Fiat Chrysler are highly detailed, the agency doesn’t come right out and say its code is illegal. That might be a legitimate conclusion in the VW case, since the software allegedly was designed to detect whether a car was on a test stand or being driven on the open road, including code to detect whether the steering wheel was being turned. Fiat Chrysler might well argue its code was designed to adjust engine settings under certain conditions without violating Clean Air Act regulations, such as during engine startup or heavy acceleration.

“Evidence and context comes into play,” said Bryant Walker Smith, a former transportation engineer who teaches technology and mobility law at the University of South Carolina Law School and is an advisor to the Transportation Dept. on automation. Are there going to be memos explaining how engineers designed the software to get around federal testing procedures? As opposed to a design decision that generates inconsistent results?

Simply making a mistake or a design oversight generally doesn’t draw prosecutors’ attention,” said Smith. “Recklessness, lying, covering it up – it’s that evidence of intentional action that’s really going to draw their attention.”

One problem for software engineers is pollution regulations tend to specify the conditions under which emission control systems will be tested, making it easy to write code that fulfills the test requirements but operates differently in the real world. That’s a given with mileage standards, which are rarely achievable in ordinary driving.

Pollution regs include a catchall requirement that the systems must work “under conditions which may reasonably be expected to be encountered in normal vehicle operation,” however. In Fiat Chrysler’s case, the government accuses the company of including Auxiliary Emission Control Devices – mostly software code – that in at least eight cases reduced the effectiveness of the emission control system on the road. The EPA says Chrysler failed to disclose these to the government, a claim Fiat Chrysler denies, and it also says they don’t fit within legal exceptions for code designed to protect the engine or allow it to start up more efficiently.

Proving Fiat Chrysler’s guilt will probably require more than simply looking at the code and how it functions. The government will need to prove that Fiat Chrysler deliberately wrote the code to evade pollution regulations, and didn’t have a valid engineering or safety justification.

What happens when the code simply provides the operating parameters for an artificial intelligence system that produces its own solutions to problems as they come up? Take a self-driving car: What if the engineers program it to deliver its occupants as safely as possible, and the AI system decides that means accelerating past the speed limit to stay away from heavy trucks?

“If the system is trained based on goals, you may not even know the route the system takes to achieve those goals,” said Smith. “You could have systems that are noncompliant with the regulatory regime but there’s no culpability you can attribute to a specific designer or even the company.”

Civil liability will face a similar problem, said Smith, who’s written about the challenges driverless cars will present to the existing civil justice system. Ever since the Pinto exploding-gas tank debacle, carmakers have been careful to avoid putting explicit cost-benefit calculations in documents that can be discovered in litigation, Smith said. It’s too dangerous for a jury to hear that a family of five might have been spared a fiery death if the company had just spent X dollars more on a different fuel tank design.

But software engineers may have to bake those cost-benefit calculations into artificial intelligence programs for autonomous vehicles, Smith said. Because the software is designed to optimize solutions to specific goals, it may need to be told how to weight various probabilities such as the odds of a catastrophic accident versus spending an extra 15 minutes on the road.

“As we move from designing the code that actually operates the system to designing the code that designs the code that operates the system,” Smith said, the very definition of liability will change.

Follow me on TwitterSend me a secure tip