The Right to Obscure: A Mechanism and Initial Evaluation

Eric Hsin-Chun Huang, Jaron Lanier, Yoav ShohamIJCAI 2015

Introduction

The 2014 European Union Court ruling establishing the "Right to be Forgotten" (RTBF) created a significant legal and technical challenge for search engines. The RTBF allows individuals to request the removal of search results deemed "inaccurate, inadequate, irrelevant or excessive." However, current implementation relies on subjective, non-scalable, and opaque manual reviews by search providers. This paper proposes the "Right to Obscure" (RTO) as a transparent, algorithmic alternative that balances individual privacy with the public's right to information without resorting to content deletion.

The Proposed Mechanism: Query-Level Filtering

The RTO mechanism shifts the focus from data deletion to query management. Instead of removing links from a search index, the system modifies how name-based queries are processed.

Protocol and Filtering

A verified user can register "minus terms"—specific keywords they wish to exclude from search results associated with their name. When a search is performed for that name, the engine automatically appends these terms (e.g., [Name] -[minus term]). This effectively hides unwanted results from casual searches while keeping the data accessible for those who specifically seek it.

Conflict Resolution and Abuse Prevention

To prevent "demotion attacks" where individuals with the same name attempt to hide each other's legitimate content, users can register "plus terms." If a term is registered as both a "plus" and a "minus" by different individuals sharing a name, the filter is nullified. Additionally, the system prevents "promotion attacks" by banning high-frequency dictionary words from being used as minus terms.

Experimental Evaluation

The researchers evaluated the RTO mechanism using a custom search engine powered by the Microsoft Bing API. Participants on Amazon Mechanical Turk were tasked with researching four distinct "Data Subjects": Lindsay Lohan (celebrity), Darleen Druyun (public figure), Jane Doe (ordinary citizen), and John Doe (individual with a criminal record).

Results and Impact

The effectiveness of the mechanism varied significantly based on the subject's public profile and the searcher's intent.

Effectiveness for Private Citizens

The mechanism proved most effective for ordinary citizens like "Jane Doe." As shown in Figure 4, filtering significantly shifted searcher impressions from negative to positive by ensuring that damaging content was no longer the primary result encountered.

Impact of filtering on searcher impressions for 'Jane Doe'.

Limited Impact on Celebrities

Conversely, the mechanism had negligible impact on well-known figures. For Lindsay Lohan (Figure 5), the p-value of 0.672 indicates no statistically significant change in searcher perception. This is attributed to the searchers' existing prior knowledge and the prevalence of information across multiple non-search channels.

The Role of Searcher Agenda

The study also found that a searcher's specific "agenda" influences the filter's success. While casual browsing is easily filtered, searchers with a specific goal (e.g., "is this person a criminal?") often bypass the name-only filter by adding their own keywords. This suggests that RTO protects casual reputation while preserving access to vital public safety information for diligent searchers.

Discussion and Limitations

The "Right to Obscure" offers a computationally efficient and transparent alternative to manual RTBF processes. However, several challenges remain. Individuals with common names may face "keyword wars" over shared terms. Furthermore, a database of minus terms creates a "honeypot" of sensitive information. Finally, the authors suggest that identity verification could be cross-referenced with criminal databases to address the ethical complexity of obscuring criminal records.

Future research will explore the application of RTO in domain-specific environments, such as professional networks or social media platforms.

Summary updated

BibTeX
@inproceedings{huang2015right,
  author = {Eric Hsin-Chun Huang and Jaron Lanier and Yoav Shoham},
  title = {{The Right to Obscure: A Mechanism and Initial Evaluation}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2015},
}
Figure