Hacker News Reader

Comment by jcromartie | Hacker News Reader

Why shouldn't we use statistical inference to deliver information that enables responders to be better prepared to deal with a situation? I would hate to be a false positive, but I'd rather have the chance of being flagged as "green" when police are headed my way, rather than always being presumed to be a "red" as seems to be the case now. This seems like it would be the opposite of being presumed guilty until proven innocent, for most people.

And this isn't surveillance: they aren't collecting new information, just putting available data sources together. Unless deducing things from existing data is somehow surveillance?

View on HN

We should not use statistical inference because statistics do not always paint an accurate picture.

An extreme example is that many criminals enjoy ice cream; obviously, it's a fallacy to say that all ice cream eaters are therefore criminals. But what about correlations between crime and race, sexual preference, location? If my neighbor is a terrorist, am I his accomplice? Is everyone who lives in a bad neighborhood a crook, or are some of them trying to scrape by legally?

It seems like this argument appears on every single thread about surveillance. Data does not lie, therefore we cannot lose if we use data to solve crimes. The problem, friend, is that data does not lie--humans do. We lie to ourselves all the time, because the patterns fit and we are creatures that survive by pattern recognition. We see patterns even where they aren't patterns.

It's also very much possible to draw the wrong conclusions as you mentioned. So while data may not lie, there certainly are plenty of faulty interpretations.

What scares me is the fact the data aggregation is being done with mere promises of "trust us, we won't look" . Systems built this way are inherently vulnerable to secondary and tertiary usage. That's why I think there is a strong case for end to end encryption so that there is separation but ultimately this is still a problem because there is no way to enforce metadata constraints on data. Someone may design such a system but others will just resort to using something else. Frustrating.

The anecdote from the article strikes me as a really good example of why this sort of statistical profiling is unnecessary. Guy threatens his girlfriend, algorithm flags him as red, so they... call in a negotiator and successfully de-escalate the situation. I'm struggling to see how the system flagging him as red had anything to do with their actions. It was just a case of cool heads doing good police work.

Cops already have a lot of contact with parole inmates who have dramatically reduced rights. So they'll just lump "red" in with the parole inmates. You'll hear cops saying things to each other like "he wouldn't be red unless he deserved it". Generally, having the cops prejudiced against someone before they even meet them is not going to turn out well.

US cops are not trained to prioritize descalatio in general. That's why you have that cop in McKinney doing a diving roll like Farva.

Wouldn't it be better for the police not to presume people to be 'red'?

For example, here in Holland the police are (still) very much public servants alongside 'enforcers of the law'. I actively seek them out if I need assistance of some kind. My experience in Brazil and the USA, on the other hand, was that the police appeared threatening (uniforms, stance, weapons) and the few interactions I had with them were scary and antagonistic. It was my first real experience being scared of the police, and a real eye opener.

The issue is judging everyone before a trial. i.e. the opposite of the presumption of innocence PLUS huge scale. Like mass tracking of people using phones versus being tailed.

Reply on news.ycombinator.com