In 2016, an investigation by The Seattle Times revealed that LinkedIn's search algorithm exhibited gender bias by consistently suggesting male names when users searched for common female names, while not doing the reverse for male names. LinkedIn attributed the behavior to an algorithm based on past search frequencies rather than intentional gender bias. Following the report, LinkedIn updated its search algorithm to explicitly recognize names and stop suggesting alternative names for search queries.
An investigation by The Seattle Times in 2016 found a gender bias in LinkedIn's search engine.
Risk classification
- Primary risk domain: 1 Discrimination & Toxicity
- Primary risk subdomain: 1.3 Unequal performance across groups
The search algorithm's effectiveness and accuracy differed by demographic group, performing worse for female names by incorrectly suggesting male alternatives while performing correctly for male names.
Additional risk subdomains
- 1.1 Unfair discrimination and misrepresentation: The search suggestions systematically misrepresented female professionals by prompting users to search for male equivalents instead.
Causal factors
- Entity: AI
- Intent: Unintentional
- Timing: Post-deployment
The gender-biased search suggestions were generated automatically by LinkedIn's deployed search algorithm as an unexpected outcome of analyzing past search frequencies.
EU AI Act risk tier
High Risk: The AI system is used within a professional networking platform that directly impacts recruitment and employment access. Under the EU AI Act, 'Employment and worker management systems, such as AI used in recruitment' are classified as High Risk.
AI system and alleged parties
- AI system: LinkedIn search algorithm
- AI purpose: Content Search; Predictive Typing
- Behaviour type: Assistant
- Alleged developer: LinkedIn
- Alleged deployer: LinkedIn
- Alleged harmed parties: Women
Harm severity
Severity ratings are based on this scale.
Highest direct severity in any category: Minor. Severity is scored from Negligible to Catastrophic in each harm category, for harm the reports describe as caused directly or indirectly by the AI system.
- Physical: direct Negligible, indirect Negligible
- Infrastructure: direct Negligible, indirect Negligible
- Property: direct Negligible, indirect Negligible
- Financial: direct Negligible, indirect Negligible
- Environmental: direct Negligible, indirect Negligible
- Malicious content: direct Negligible, indirect Negligible
- Differential treatment: direct Minor, indirect Minor
- Civil rights: direct Negligible, indirect Negligible
- Democracy: direct Negligible, indirect Negligible
- Privacy: direct Negligible, indirect Negligible
- Psychological: direct Negligible, indirect Negligible
- Epistemic: direct Negligible, indirect Negligible
- Child sexual exploitation and abuse: direct Negligible, indirect Negligible
Differential treatment
Reported: The report explicitly describes differential treatment where searches for common female names prompted suggestions for male names, while searches for male names did not prompt female suggestions.
Directly caused: The search algorithm systematically treated female names differently by suggesting male alternatives (e.g., prompting 'did you mean Stephen Williams' when searching for Stephanie Williams), affecting at least a dozen common female names.
Indirectly caused: This differential treatment potentially reduced the visibility of female professionals to recruiters and other platform users.
Inferred additional harm: It is likely that thousands of female professionals on the platform experienced reduced profile views and fewer professional opportunities due to this systemic bias.
People affected
- Occurrences reported: 1
- People reportedly exposed: 225000000
Potential causes
Management
- Lack of Proactive Risk Assessment: Management did not assess risks of algorithmic bias before launch.
Technology
- Spelling Correction Algorithm: Algorithm corrected valid names based on spelling and frequency.
- Lack of Name Recognition: System did not recognize inputs as names to prevent correction.
Data Inputs
- Biased Search Query History: Algorithm learned from past user queries that favored male names.
- Skewed Profile Frequency Data: Relative frequencies of male names in profiles biased the model.
Human Factors
- Biased User Search Tendencies: Past search behaviors of users skewed toward male professionals.
- Homogeneous AI Workforce: A workforce skewing male may fail to anticipate gender bias issues.
Process and Methods
- Inadequate Bias Testing: Testing processes failed to detect asymmetric gender-biased suggestions.
Information quality
- Classification confidence: High
- Reason for confidence: The reports provide consistent, detailed accounts of the search algorithm's behavior, the specific names tested, LinkedIn's official response, and the subsequent algorithmic fix. The facts are well-documented across multiple reputable news sources.
- Ambiguities identified: There is some ambiguity regarding the exact proportion of male versus female users on LinkedIn at the time, which could influence search frequency data.
- Alternative interpretations: LinkedIn argued the behavior was purely mathematical based on user search frequencies rather than systemic gender bias, though the outcome remains functionally biased.
An investigation revealed LinkedIn's search algorithm exhibited systemic gender bias by suggesting male names for female queries. This commercial algorithmic bias represents a minor societal equity concern with negligible direct implications for national security.
- Overall national security impact: Minor
- Scope: Multiple nations
- Primary target: No clear primary
- Alleged perpetrator: Unknown
Threat characteristics
- Imminence: Long-term. The incident represents a long-term algorithmic bias and design issue rather than an active national security crisis.
- Autonomy: Human-supervised. The algorithm autonomously generated spelling and name suggestions based on historical data, though users made the final decision to click.
- Novelty: Established threat. Algorithmic bias arising from historical training data is a well-documented and established issue in machine learning systems.
Impact by dimension
- Physical security: Negligible. No physical systems, critical infrastructure, or kinetic capabilities were affected by this search algorithm issue.
- Information security: Negligible. No evidence of state-sponsored information operations, intelligence compromise, or malicious disinformation campaigns.
- Sovereignty: Negligible. The incident involved a commercial networking platform and did not impact government operations, territorial sovereignty, or elections.
- Economic security: Negligible. While the bias affected professional networking, it did not pose a threat to critical supply chains, financial systems, or strategic national industries.
- Societal stability: Minor. The algorithmic bias resulted in widespread differential treatment of female professionals on a major platform, representing minor societal equity concerns but no threat to stability.
.avif){kind=small}