Hacker News Reader

Comment by dendory | Hacker News Reader

"Pre-crime"

You thought it was a movie, now it's reality. Soon, it'll be something you get arrested for. Mark my words.

If you have a record then isn't it understandable that you might attract more attention than someone with a blameless past? It seems logical to me anyway. Obviously not to the extent that you'd be held but keeping an eye on you is fair enough imo.

Do you really think that a previous record and terrorist acts correlate significantly?

Yes. Membership of the domestic terrorist groups in my country correlates very strongly with a history of juvenile anti-social behaviour. These youngsters frequently come into contact with these groups either when subject to vigilante punishment or 'get taken under their wing'. Most terrorists aren't university graduates

> Membership of the domestic terrorist groups in my country correlates very strongly with a history of juvenile anti-social behaviour.

Is the reverse true?

Can you explain what you mean by the reverse?

Domestic terrorists may very likely have exhibited juvenile anti-social behavior. How likely is the reverse? That is, how likely is it that someone who exhibits juvenile anti-social behavior will eventually be a domestic terrorist?

My money is on "it is very unlikely", and that is a problem.

That's analagous to saying it's a problem that Health screenings are often targeted at high-risk groups when the vast majority of those groups won't get the disease in any case.

.....and, where the cost of false positives is sufficiently high and the rate of false positives is sufficiently high, that is a very relevant consideration that is in fact made by the medical community. That is just a basic of modern medicine.

The world is not black and white. These are things that must be weighed and considered rationally with real data, not emotionally with fabricated absolutes. You don't get to discard half of your equation because when you include it the results don't make you feel warm and fuzzy.

P(A|B) = \frac{P(B | A)\, P(A)}{P(B)}

Not a hard concept.

I'm sorry but you have gone off on a tangent there. We weren't arguing about the effectiveness of the policy just the rationale behind it, I believe. I don't have any stats on that, do you?

You were never arguing. You were being educated.

A rationale that includes P(B|A) but not P(A|B) is not a rational one. I have been asking you to evaluate the legitimacy of the rationale by considering P(A|B).

Given that terrorists are very rare then the P(A|B) using your terminology (where A = terrorist and B= bad record) will be very low. But for for a fixed P(A) and P(B) then the higher the value there is for P(B|A) the higher the value of P(A|B) will be. If you have an enornmous haystack to search it's more efficient to look first where you're most likely to find needles.

P(B|A) can be 100%, but if P(A) is astronomically small (and it is) and if P(B) very high (and it is), then P(A|B) will be so low that it becomes clear P(B|A) is absolutely worthless.

As an extreme example demsonstrating the necessity of considering P(A|B), suppose we observe that 100% the 9/11 hijackers had black hair (P(black hair|terrorist) = 100%). That observation is worthless because P(terrorist|black hair) is stupidly low.

Just because we can say something about the traits a terrorist will likely have does not mean that we can rationally use those traits to detect terrorists.

If you are proposing a P(B|A), it is on you to show that the P(A|B) is sensible. Just pointing at a high P(B|A) alone is worthless. That's how you make good "Tonight at 11', is [common household objects] killing you slowly?!?" headlines, but that is about it.

In your extreme example you can see that people in the B' group (those who don't have black hair) can be safely excluded as suspects. So even in this artificial example you can see that knowing P(B|A) is of value.

You seem to have missed the part where the extreme example is not real life. If you exclude people without black hair from suspicion, then the baddies will find someone without black hair. You cannot assume P(B|A) will not be modified by the policies you adopt. ...well, you can if you are the TSA and don't actually have to answer to anybody....

P(B|A) will never be 100% anyway though, so even if the terrorists were unable to adapt it would not matter. The point is that even in the absurd event that you found a P(B|A) that was 100%, it doesn't mean it is useful. Or do you actually think that screening for black hair in the example would be worthwhile?

You seem to have missed the part where the extreme example is not real life Actually, I referred to your example as both 'extreme' and 'artificial' in my previous post. Since the example was constructed to make your point as obvious as possible I was just pointing out that in fact P(B|A) was actually useful even in that example.

Sorry for the delay in responding, I gave up yesterday because there was no option to reply even after a very long time.

Effectiveness is the only rationale that should matter. Also, you might benefit from having a look around http://lesswrong.com/; many of the arguments you will see on HN are grounded in the concepts of Bayesian rationality described there.

Thanks for the link, I'll have a look at it. What I meant by rationale was the reasoning behind the policy, it's actual effectiveness would, I guess, only become apparent in practice.

"If you have a record then isn't it understandable that you might attract more attention than someone with a blameless past?"

Not when we have people who were charged with sex crimes because when they were 17 years old they dated a 15 year old, and who now have to register as sex offenders. The law has become completely out of control -- saying that someone has broken laws in the past is no more of a justification than saying that a person breathes air.

Eventually, having a blameless past will be cause for suspicion, if it's not already. Because at some point, not having a record is is sufficiently different from everyone to not be "normal".

Good point. That reminds me of hearing how traffic cops decide to pull someone over to check for drink driving. One of their criteria was any car that appeared to be sticking too rigidly to the speed limit

Reply on news.ycombinator.com