What if you're accused by a machine?

You stop at a bar after work have a drink. You drive home and get pulled over. The officer says you were swerving and administers a breathalyzer. He claims you're over the limit but you're convinced you're not. What do you do? You have the right to face you're accuser but that's not really the officer. He just read numbers off a machine.

This has bothered me for a long time because I'm a software engineer. Virtually all software has bugs. The more complicated the software, the more bugs it has. Hardware's the same way. How do you know the breathalyzer worked properly? (Or the radar camera, for that matter?)

Well, one gentleman has fought this in Florida. As it turns out, though the machines are certified by the state, the judge was disturbed to find out that the state's breathalyzers electronics appeared different from machine to machine. The state argued that they can't turn over the software for the EPROMs (Electronic Programmable Read-Only Memory) because it's a trade secret. Well, that sucks. Because some company doesn't want their trade secrets turned over, you're not allowed to face your accuser. That struck the judge as unfair.

The judge has ordered that the source code and schematics of the breathalyzer be turned over to an expert witness (PDF). Only the expert witness can receive this information and he can't let anyone else see it. This is both good and bad. It's a small step in the right direction, but it does mean that every time someone is pulled over, if he or she can afford an expert witness, this process has to be repeated. Unless you're poor and can't obtain such a witness, of course.

It's going to be fascinating to see what comes of this. The state has fifteen days to provide the software. The state claims they don't have it. If the manufacturer won't turn it over, charges will be dropped. If the manufacturer does turn it over and the witness finds a defect, there's going to be hell to pay. Personally, I'm hoping for a defect. It's not that I want drunk drivers to get off. It's that I want society to start thinking about what it means for individuals to be accused by machines we're not allowed to interrogate. It's like being allowed to face our accuser but not being allowed to cross examine him.

Did the state certify all machines? What was their certification process? How often are the machines tested? How are the officers trained to use them? Will mouthwash set them off?

If machines are going to accuse you, their hardware and software should be open-source and widely available for anyone to look at and ensure that it's bug-free. Anything else is an offense to the Constitution. Claiming that a company's trade secrets trumps our right to face our accuser is disgusting. (Hear that, Diebold?)
Friend of mine in grad school was called in for jury duty, and as part of the jury selection process asked if he'd convict someone for drunk driving based on a breathalyzer. He told them it would depend upon the reading versus the standard deviation associated with measurements.

He was quickly tossed off of jury duty.

About this time in the mid-90s, the county that Dallas resides in decided to mostly stop doing breathalyzers - too expensive, and instead focus on giving out public intoxication tickets that only required reasonable suspicious on the part of the ticketing officer.
"He was quickly tossed off of jury duty."

Are you sure it was because of what he said? Because that sounds like exactly the right answer to me, and probably is exactly what I would say too.
I've read a number of times that lawyer's do not want jurors who can discern the truth. They want jurors who can be pursuaded. Our adversarial system is not predicated on finding out what really happened. It's predicated on winning.
Additionally, the attorneys may be allowed a certain number of free rides, where they're allowed to toss a prospective juror without stating cause.
That's the only question he was asked and answered, and lawyers had a very limited number of people they could excuse from jury selection.

While it is a great answer, if you can have someone who is going to convict based on 0.081, why would you go with the guy who wants to know just how close to 0.08 you can be given device error?
I had the same gripe with photo radar. You can't face your real accuser. I wanted to fight this when I got my ticket via photo radar because I thought I wasn't speeding and that it triggered incorrectly or maybe off another car and took a picture of mine. However, upon lots of investigation of others who tried to defend themselves in Oregon courts, it always ended badly for the defendant and I just didn't have the money if I wanted to really fight such an issue. It was easier to pay the ticket and the increase in my insurance over 3 years.

I really hope they find a defect too.
You can't really tell much from looking at schematics and code, unless there is an obvious flaw. What matters is reliability and variability, not design. The most meaningful way to evaluate whether these devices are doing their job is to test over a large sample of machines and test subjects, and see the error between machine reading and blood tests. So I don't think this is an open source issue at all.
For any sufficiently complicated program, it's impossible to follow all paths through the program and demonstrate that they do exactly what you want them to do.

This, of course, isn't what the halting problem is, but it's a symptom of the same thing: we can't prove what's going to happen. We can only be certain of a bit of reliability in a constrained number of cases.
Isn't that the point of stateful simulation? To ensure that you have all states in all modules documented, and that the means of state transitions are known. Traveling off the known states, or illegally transitioning states should trigger an exception.

I understand that this is difficult, but health & safety software is written in this manner. Software in regulated industries like medicine, avionics, classified systems, are allegedly written in this manner. I'd expect that software written for systems that will represent a legal expert, would follow the same thing.

Whenever I've done forensics, I design my examination and my reports to be bulletproof. I may end up being sworn in as a court-recognized expert witness. I honestly thing that any system of software that will act as an automated expert witness should follow the same standard.

Of course, I also expect that I'm wrong... I was wrong about the software in those fucking voting machines... MS Access? MS ACCESS? Still bitter about that.
Under UK Law. if the result of the breath test is under 50µg/100ml (the legal limit is 35µg) then you can demand a blood or urine sample be taken and used instead.

This nicely sidesteps the problem. If the case is borderline then you can ask for an alternative test to be made. This means you don't have to totally trust a 'black box' at all, so it doesn't mean that in order to get convictions that you have to make a closed source machine judge and jury as it were. It's demoted to just another data point, along with the other tests, that can be used to disprove reasonable doubt.
Here in Norway they never convict on just a breathalyzer test.
If there's any suspicion of intoxication above legal limits they ALWAYS take a blood sample.

Incidentally, a few cases of speeding have been tossed out because of how the radar or laser equipment was used.

Automatic traffic cameras?
Here there's a legal requirement that the driver of the vehicle can be identified.
In theory that means you could put on a mask of Bush and test how fast your brand new Coënigsegg goes without any other side effects than a large fuel bill... (In practice you may need to have reported it stolen first... )
I like it. Oh it makes my legalese senses tingle. Hmm....

You would have been really fun to talk with when I was studying the DMCA in copyright law.
I agree with this principle -- I think software running on machines used for public purposes (voting, breathalyzers) should be open source for the same reason the text of laws and constitutions citizens have to follow must be available to them for inspection.

Though, it doesn't seem to me that the software could be open source in the more inclusive sense of accepting change submissions from the public. Though wouldn't it be neat if we could submit changes to our laws?

I found your blog very interesting. It may be of the interest to you to check out other great site that talks about bad health care
Post your alternative medicine complaints and help others like yourself.Come share your bad experiences with clinics offering
alternative medicine and services at