Comment by TeMPOraL | Hacker News Reader

Comment by TeMPOraL | Hacker News Reader

Parent Full thread

Indeed. People are forgetting we have a whole piece of math - called probability theory - for dealing exactly with that stuff. It can tell you whether or not doing X is an evidence for you being Y (or evidence against it).

The important thing is - probability theory doesn't depend on rhetorics. Someone speaking the "nothing to hide" argument, or someone else finding arguments against it convincing, doesn't suddenly invalidate Bayes' theorem! Trying to argue from principle is a mistake because reality doesn't listen to principles. The only relevant question is whether encryption/Tor/whatever is or is not currently an evidence for/against being an evildoer, and how strong that evidence is.

Those considerations do include things like loss of utility because of false positives / false negatives, too. So for instance, it may be so that evildoers are more likely to use encryption than general population, but there's so much normal people using encryption that you can't get evidence for being evil out of just noticing that a person uses encryption (the same considerations explain why so much of what TSA is doing is security theater).

I agree with you in general. My issue with this line of thinking for policy decisions, is that I personally believe relying solely on objective probability theory to associate people with _______, without looking at the larger picture, and then building policies from that, could exacerbate the reason the association is being payed attention to in the first place.

Let's say theoretically, that people who wear red shirts are objectively more likely to commit a crime. Policies are made such that people with red shirts are targeted for extra policing. People with red shirts who are not criminals feel discriminated and overly watched, and because of that, generate a resentment against society and (I admit this is a jump), ultimately end up with a higher disposition towards committing crimes. And then it just goes on ad infinitum.

So in some situations, I would argue that society benefits from ignoring these objective probabilities entirely, if there's evidence that taking action based on them could cause downstream negative consequences.

I've seen this argument brought up in discussions about racial profiling, and... I think this may be a real - and in some situations significant - phenomenon. I currently have no idea how to model it in a way that could plug into the probability calculations. I need to think about this some more.

I say in some situations, because the typical case is that if something (e.g. wearing red shirts) is likely to attract attention you don't want, you'll tend to avoid doing that thing (you'll stop wearing red shirts in order for the police to leave you alone). So there's a negative feedback loop. But I can see how this loop can fail in case of things one can't (easily) change - like skin color, cultural background, religion, etc. This leads me to believe that just looking for probability-based signals may be a good thing in case of guns and encryption, but a bad thing in case of race and religion.

"The only relevant question is whether encryption/Tor/whatever is or is not currently an evidence for/against being an evildoer, and how strong that evidence is."

I think we need to be careful not to argue law from probability theory. While as you've pointed out many things will not actually be predictive of "being a terrorist" (whatever that means) there will be things that _are_ predictive. But we mustn't let the predictive nature of the information be in and of itself justification for obtaining it. There needs to be respect for constitutional limits on the reach of government, regardless of what the math says.

I agree. How you use the information once you have it is not the same as how you obtain it in the first place. I was addressing the use of information, and I do agree that obtaining those signals has a separate set of considerations to take into account.

But unless there are strong reasons to do otherwise - for example, what 'wakkaflokka pointed out in a parallel comment may potentially be such a reason - I do think we should derive rigorous conclusions from the data we already have based on math, and not on rhetoric.

Reply on news.ycombinator.com