[splat]

> That means that 96% are guilty. That's beyond a reasonable doubt and probably not far from the best accuracy you can expect from an inherently fuzzy system.

I disagree. A 1 in 25 chance that someone is innocent certainly seems like a reasonable doubt to me. Even 1 in 100 seems like a reasonable doubt. 1 in 1000 is maybe unreasonable, but 1 in 25?

At any rate, the formulation that the legal system prefers to use is the other way round. Rather than how many innocent people we should put in jail, the question is how many guilty people we should allow to go free to prevent a single innocent person from going to jail. The idea that an innocent person could possibly be convicted is totally at odds with the way the legal community sees itself. (Which is why, incidentally, it is so difficult for the law to accept that innocent people have been wrongly convicted.) So I disagree that the system is not at fault. The legal system must recognize that false positives exist and it needs to handle them intelligently.

By the way, there's a wonderful article by Eugene Volokh speculating on what an acceptable number of guilty men to let go free is [1].

[1] http://www2.law.ucla.edu/volokh/guilty.htm