What to do about re-identification? A precautionary approach to big data privacy | Hacker News Reader

What to do about re-identification? A precautionary approach to big data privacy

freedom-to-tinker.com

35 pointsrandomwalker6 comments

A basic problem with de-indentification of data is useful uniqueness. The sensitive data that needs de-identified tends to be the data that provides the uniqueness needed to make the data useful. Sufficiently scrubbed data often doesn't work well for actual testing.

After years in banking and health care, I've seen businesses tilting at that windmill too often. Hire outside consultants to help if you need to do it.

The problem that I've noticed is that we DO hire outside consultants... but too often that provides a security blanket for the decision makers, and they think that just because they outsourced the work, that they have nothing to worry about anymore.

I think the author's approach is pragmatic. I think more decision makers need to realize that data sets can just as easily be a liability as they can be an asset. I work in healthcare and I see slow motion trainwrecks waiting to happen all around me.

Narayanan's article is valuable, but I think he's far out in front of public policy. AFAICT the default in policy is that privacy doesn't matter; that's the issue on the table right now. How to implement it or deal with challenging technical issues seem to be far down the road.

My impression and limited experience is that policy makers don't care about privacy; they don't understand it or its implications, they haven't the slightest understanding of technical issues, and they have no motivation to learn because their consituents don't understand or care.

The way it changes is if people who do understand it, such those reading HN, patiently spread the word to people you know socially, in business and to your elected representatives. They need your help; they lack the resources to study every public policy issue out there, especially those requiring technical expertise.

>AFAICT the default in policy is that privacy doesn't matter

I think that may be true for some, but not for others. It depends a lot on previous experience I think. My uncle was a county clerk and digitized a lot of public records, for which he was rewarded by unhappy citizens filing a FOIA act on his e-mail account (a great reason to never e-mail relatives at their work address if you don't want it all potentially public someday). After that, his office was a lot more cautionary about data releases, even though it was public record.

For some counter-examples on the business side, Netflix got bit by the re-identification bug and now is very careful about data releases. At Kiva, I face this problem all the time of turning down research requests for what would be interesting and useful analysis. We release an anonymized dataset of our loan activity (that's taken a lot of work to properly sanitize). However, we get requests nearly every day to release more information, which we can't do, due to the real possibility of re-identification.

Thanks; that's very informative.

> we get requests nearly every day to release more information, which we can't do, due to the real possibility of re-identification.

Is there a legal liability for re-identification? Either way, it's good to know that you are being careful and thinking through the risks.

Netflix got sued and investigated by the FTC for the data released for the Neflix Prize contest. One the plaintiffs could be linked to their imdb account by similar reviews and there were additional reviews in the Netflix dataset to imply that they were lesbian while still in the closet.

AOL had to pay a settlement after the search data they released identified people.

Reply on news.ycombinator.com