Washington Finds Computer Simulation Unreliable

Toadpipe writes "Washington State Court of Appeals reverses a conviction in which a computer simulation had been the main evidence. Quoting 'At issue was PC-Crash, a computer program distributed by Vancouver, B.C.-based MacInnis Engineering Associates. The program recreates traffic collisions using simulations and reconstructions. "PC-Crash had not been validated for the purpose for which the evidence was offered, simulation and prediction of multiple-occupant movement within a vehicle during a multiple-collision accident," the Court of Appeals said in ordering a new trial. "There is no general acceptance in the relevant scientific community of the use of the PC-Crash program for the purposes to which it was put."' Here is the Court's opinion."

Comments

Digital evidence

[TWX]

I'd like them to stop accepting photo radar data. The cities in my area have switched to digital photography. Currently one's only out is to request the Plaintiff to produce the calibration records for the system for the day of the ticket and hope that they don't have that data.

I'd be okay with photo radar and with red light cameras if they were used to bolster the Prosecution/Plaintiff, like if there were a car accident and the red light camera data were used to show that the cited person (by the officer on the scene) had indeed run the light, and that the officer was correct. The current system of using photography with near-automatic conviction deprives people of privacy. If the police want to cite people for speeding or for any other traffic violation then they need to get out there with people who will be required to testify as to what they saw; people who actively claim the count in the charge, not some computer or desk-jockey who analyses data after the fact.

Of course, I also have the opinion that if there's no victim then there's no crime. Take this as you will.

Re:Darn it

[X0563511]

I think it follows this:

1 hour of MS Flight Simulator (latest) = 1/2 normal flight hour.

You can use it for up to half of the hours needed for a pilot licence.

Im not sure about the numbers, but i do know that a portion of your flighttime under the M$ simulator counts (well, its pretty frikin realistic if you set it to be)

What's so wrong with it?

[Artraze]

The current system basically entails putting an expert on the stand and asking his/her opinion. I fail to see why this same system doesn't apply to the software. Of course it's not going to be perfect, but neither is an expert's testimony. So why can't this be considered something of an "expert" on crash simulation?

Re:Duh

[EvilTwinSkippy]

You don't understand simulations.

When Ford, or Daimler Benz goes out to design a car, they know where every bolt, nut, rivet, weld, and cup holder are. That information is fed into a finite element analysis model that breaks the car down into ever finer blocks of deformable material.

They than take that model and bash it against a series of controlled obstructions.

Even then, those simulations are just used to rule out certain design changes. All final designes are bolted to a hydrolic ram, filled with test dummies, and shot into a wall or another vehicle.

And again.

And again.

Yes, the automaker DOES have a model of the car. Yes, that model could be fed into another FEA. But in order to produce any meaningful result you would have to have equally good data about all the occupants in the car. Where everything on the road was, and at which time in the "event."

And did I mention that the simulation is only as good as the least accurate measurement? At best. And most of the data you would have needed is gone as soon as rescue crews arrive and attempt to move the vehicles out of the way of traffic?

Re:Duh

[EndingPop]

Actually, in today's real world, finite element models are used far more extensively than actual experiments, at least in the auto industry. A prototype costs well over a million dollars and around 6 months to create, while the model is on the order of tens of thousands of dollars and a few weeks to create and run.

Simulations have to be used intelligently, just like lab results. Experiments can be set up wrong, just as variables in a model can be input wrong.

Re:Duh

[foog]

Ever been in a position to compare FEA predictions with experimental data? Analysts rarely get it right the first time, and usually have to severely tweak their boundary conditions, initial values, etc to get in the ballpark of the real world.

You're still talking about the stages where it's more important to be plausible than correct. That's basically where FEA fits in.

I won't argue that it beats guessing, though. And the full-color plots seem to mesmerize people. Note too, though, that "PC-Crash" isn't even an finite element analysis program.

Re:I actually read the article

[amliebsch]

What I found, though, is that a standard trial is a piss-poor way of getting to the truth.

As well it should be! And I say this as a prosecutor. But the reason for this should be obvious: the truth is very often unknowable. Thus, given that it is impossible for the jury to figure out truth, you narrow the scope a little, and only ask them to resolve certain factual disputes, i.e., who is/isn't lying, whether a story is plausible, what a reasonable person would do, etc. The effect of this is that juries are often shielded from The Truth in order to make their decisions on these more specific matters less biased.