Google Explains How It Forgets

Google can forget, but unlike the rest of us, the process is not automatic.

Yesterday Google told a European government data protection working party how it handles requests for search result link removals. The removals began in June after a May European court ruling (see our coverage) upholding a Spanish man’s right to be forgotten.

The working group had earlier sent Google a questionnaire on the practicalities of the removals and met with Google and two other unnamed U.S. search engines. Google’s reply revealed that it is handling the requests on a case-by-case basis, with decisions resting on recently-hired staff.  Companies that help individuals request link removals have begun receiving rejections, The New York Times reported.

The cover letter said that the company’s “approach will not be static” and that it expects to be in dialogue with data protection authorities. It spelled out the criteria by which staff decide whether to honor an individual’s link removal request. These hew close to those set out in the May court ruling.

The company noted some complications it has encountered, such as the fact that different EU countries have different policies on publishing full names in court documents. As anyone who has used Google News has discovered, the company also finds it difficult to establish what sort of online media count as “reputable” news organizations.

Googled also laid out its policy of alerting users that name-containing searches may have had results modified by legal action:

“The notification is intended to alert users to the possibility that their results for this kind of query may have been affected by a removal, but not to publicly reveal which queries were actually affected.”

Perhaps the most illuminating passage  was the admission that there is not yet a good way to convert the court’s order into a computer algorithm for filtering the public interest from the private:

“We are not automating decisions about these removals. We have to weigh each request individually on its merits, and that is done by people. We have many people working full time on the process, and ensuring enough resources are available for the processing of requests required a significant hiring effort.”

Some numbers help put that effort in perspective: the company received about 91,000 requests in the first 7 weeks the relevant form was available and is still working on the backlog. So far, it has approved 53 percent of requests, rejected 32 percent with an explanation of the reason why, and requested further information in 15 percent of the cases. It has also reversed some of its decisions already, in some high-profile cases involving the newspaper The Guardian. A UK House of Lords subcommittee recently called the court’s criteria “vague, ambiguous and unhelpful,” the BBC reported.

First published on IEEE Spectrum’s Tech Talk blog: [html]