Rethinking Privacy in an Era of Big Data
Some years ago an engineer at Google told me why Google wasn't collecting information linked to people's names. “We don't want the name. The name is noise.” There was enough information in Google's large database of search queries, location …
Context & Ripple Effects
Google's stated preference not to attach search data to names sits uneasily beside earlier coverage showing how little input may be needed to identify people from online information: Google users could be identified with two keystrokes. The privacy question is therefore not limited to whether a database holds a name, but whether its other fields can make a person legible.
The story also extends a longer debate over data retention. A 2007 account argued that the absence of governing law shaped Google's retention practices, while coverage of Google+ in 2011 kept privacy an active product question.
First-order effects
- Google's privacy posture is judged against the identifiability of its search and location records, not simply its decision to omit users' names.
- Users face a weaker distinction between named and unnamed data when combinations of behavioral records can identify them.
Second-order effects
- Privacy advocates and policymakers gain a clearer basis to press Google on retention and data-linking practices, because anonymization claims turn on what the full dataset can reveal.
- Services built around social identity, including Google+, must contend with a higher privacy bar when account data can be connected to broader behavioral records.
Third-order effects
- If identification increasingly arises from combinations of non-name data, privacy governance will shift from regulating explicit identifiers toward limits on collection, retention, and linkage.
- Big-data firms' competitive advantage in aggregated behavioral information will be paired with greater scrutiny of whether aggregation itself creates personally revealing records.
The trend: Privacy is moving from a name-and-account model toward an inference model, in which datasets are assessed by what they can reveal when combined.