![The Coded Gaze [Joy Buolamwini 2016]
Face detection software: Fails for some darker faces
Bias](https://image.slidesharecdn.com/sofusmacskassyresponsibledatause2019-11forpublic-191130224429/85/Responsible-Data-Use-in-AI-core-tech-pillars-3-320.jpg)
![• Facial analysis software:
Higher accuracy for light
skinned men
• Error rates for dark skinned
women: 20% - 34%
Gender Shades
[Joy Buolamwini &
Timnit Gebru, 2018]
Bias](https://image.slidesharecdn.com/sofusmacskassyresponsibledatause2019-11forpublic-191130224429/85/Responsible-Data-Use-in-AI-core-tech-pillars-4-320.jpg)
![Representative Ranking for Talent Search
S. C. Geyik, S. Ambler,
K. Kenthapadi, Fairness-
Aware Ranking in Search &
Recommendation Systems with
Application to LinkedIn Talent
Search, KDD’19.
[Microsoft’s AI/ML
conference
(MLADS’18). Distinguished
Contribution Award]
Building Representative
Talent Search at LinkedIn
(LinkedIn engineering blog)](https://image.slidesharecdn.com/sofusmacskassyresponsibledatause2019-11forpublic-191130224429/85/Responsible-Data-Use-in-AI-core-tech-pillars-24-320.jpg)
![Intuition for Measuring and Achieving
Representativeness
• Ideal: Top ranked results should follow a desired distribution on
gender/age/…
• E.g., same distribution as the underlying talent pool
• Inspired by “Equal Opportunity” definition [Hardt et al, NIPS’16]
• Defined measures (skew, divergence) based on this intuition](https://image.slidesharecdn.com/sofusmacskassyresponsibledatause2019-11forpublic-191130224429/85/Responsible-Data-Use-in-AI-core-tech-pillars-25-320.jpg)
![Fairness-aware Experimentation
[Saint-Jacques & Sepehri, KDD’19 Social Impact Workshop]
Imagine LinkedIn has 10 members.
Each of them has 1 session a day.
A new product increases sessions by +1 session per member on average.
Both of these are +1 session / member on average!
One is much more unequal than the other. We want to catch that.](https://image.slidesharecdn.com/sofusmacskassyresponsibledatause2019-11forpublic-191130224429/85/Responsible-Data-Use-in-AI-core-tech-pillars-32-320.jpg)
![Differential Privacy
43
Databases D and D′ are neighbors if they differ in one person’s data.
Differential Privacy: The distribution of the curator’s output M(D) on database
D is (nearly) the same as M(D′).
Curator
+ your data
- your data
Dwork, McSherry, Nissim, Smith [TCC 2006]
Curator](https://image.slidesharecdn.com/sofusmacskassyresponsibledatause2019-11forpublic-191130224429/85/Responsible-Data-Use-in-AI-core-tech-pillars-43-320.jpg)
Responsible Data Use in AI - core tech pillars
- 1. Responsible Data Use Sofus A. Macskássy Data Science @ LinkedIn smacskassy@linkedin.com November 2019
- 3. The Coded Gaze [Joy Buolamwini 2016] Face detection software: Fails for some darker faces Bias
- 4. • Facial analysis software: Higher accuracy for light skinned men • Error rates for dark skinned women: 20% - 34% Gender Shades [Joy Buolamwini & Timnit Gebru, 2018] Bias
- 5. Bias
- 6. • Ethical challenges posed by AI systems • Inherent biases present in society • Reflected in training data • AI/ML models prone to amplifying such biases Algorithmic Bias Bias
- 7. Massachusetts Group Insurance Commission (1997): Anonymized medical history of state employees William Weld vs Latanya Sweeney Latanya Sweeney (MIT grad student): $20 – Cambridge voter roll born July 31, 1945 resident of 02138 Privacy
- 8. 64%Uniquely identifiable with ZIP + birth date + gender (in the US population) Golle, “Revisiting the Uniqueness of Simple Demographics in the US Population”, WPES 2006 Privacy
- 9. A self driving car knocked down and killed a pedestrian in Tempe, AZ in 2018. - Who is to blame (accountability) - Who to prevent this (safety) - Should we ban self-driving cars (liability and policy evaluation) The need for XAI Explainability https://www.nytimes.com/2018/03/19/technology/uber-driverless-fatality.html
- 10. A recent research paper shows that a classifier that could recognize wolves from husky dogs was basing its decision solely on the presence of snow in the background. The need for XAI Ribeiro, Singh, and Guestrin. 2016. "Why Should I Trust You?": Explaining the Predictions of Any Classifier. SIGKDD 2016. Explainability
- 11. The need for XAI Explainable AI is good for multiple reasons: - Builds trust (why did you do this) - Can be judged (how much do I believe the prediction) - Can be corrected (new training or tweaks to correct errors) - Is actionable (I know what do to next) - … Explainability
- 12. Data Governance Governance Reflect company policies Ensures compliance Protects data Protects company Involves all orgs in a company https://www.dama.org/sites/default/files/download/DAMA-DMBOK2-Framework-V2-20140317-FINAL.pdf
- 13. Laws against Discrimination Immigration Reform and Control Act Citizenship Rehabilitation Act of 1973; Americans with Disabilities Act of 1990 Disability status Civil Rights Act of 1964 Race Age Discrimination in Employment Act of 1967 Age Equal Pay Act of 1963; Civil Rights Act of 1964 Sex And more...
- 15. Responsible Data Use @ LinkedIn Case studies - Bias - Privacy - Governance
- 16. LinkedIn operates the largest professional network on the Internet Tell your story 645M+ members 30M+ companies are represented on LinkedIn 90K+ schools listed (high school & college) 35K+ skills listed 20M+ open jobs on LinkedIn Jobs 280B Feed updates
- 17. Bias @ LinkedIn Fairness-aware Talent Search Ranking
- 18. Guiding Principle: “Diversity by Design”
- 19. Insights to Identify Diverse Talent Pools Representative Talent Search Results Diversity Learning Curriculum “Diversity by Design” in LinkedIn’s Talent Solutions
- 22. Identify Diverse Talent Pools
- 23. Inclusive Job Descriptions / Recruiter Outreach
- 24. Representative Ranking for Talent Search S. C. Geyik, S. Ambler, K. Kenthapadi, Fairness- Aware Ranking in Search & Recommendation Systems with Application to LinkedIn Talent Search, KDD’19. [Microsoft’s AI/ML conference (MLADS’18). Distinguished Contribution Award] Building Representative Talent Search at LinkedIn (LinkedIn engineering blog)
- 25. Intuition for Measuring and Achieving Representativeness • Ideal: Top ranked results should follow a desired distribution on gender/age/… • E.g., same distribution as the underlying talent pool • Inspired by “Equal Opportunity” definition [Hardt et al, NIPS’16] • Defined measures (skew, divergence) based on this intuition
- 26. Fairness-aware Reranking Algorithm (Simplified) • Partition the set of potential candidates into different buckets for each attribute value • Rank the candidates in each bucket according to the scores assigned by the machine-learned model • Merge the ranked lists, balancing the representation requirements and the selection of highest scored candidates • Algorithmic variants based on how we choose the next attribute
- 27. Architecture
- 28. Validating Our Approach • Gender Representativeness • Over 95% of all searches are representative compared to the qualified population of the search • Business Metrics • A/B test over LinkedIn Recruiter users for two weeks • No significant change in business metrics (e.g., # InMails sent or accepted) • Ramped to 100% of LinkedIn Recruiter users worldwide
- 29. Lessons learned • Post-processing approach desirable • Model agnostic • Scalable across different model choices for our application • Acts as a “fail-safe” • Robust to application-specific business logic • Easier to incorporate as part of existing systems • Build as a stand-alone service or component for post-processing • No significant modifications to the existing components • Complementary to efforts to reduce bias from training data & during model training
- 30. Engineering for Fairness in AI Lifecycle Problem Formation Dataset Construction Algorithm Selection Training Process Testing Process Deployment Feedback Is an algorithm an ethical solution to our problem? Does our data include enough minority samples? Are there missing/biased features? Do we need to apply debiasing algorithms to preprocess our data? Do we need to include fairness constraints in the function? Have we evaluated the model using relevant fairness metrics? Is the model’s effect similar across all users? Are we deploying our model on a population that we did not train/test on? Does the model encourage feedback loops that can produce increasingly unfair outcomes? Credit: K. Browne & J. Draper
- 31. Engineering for Fairness in AI Lifecycle S.Vasudevan, K. Kenthapadi, FairScale: A Scalable Framework for Measuring Fairness in AI Applications, 2019
- 32. Fairness-aware Experimentation [Saint-Jacques & Sepehri, KDD’19 Social Impact Workshop] Imagine LinkedIn has 10 members. Each of them has 1 session a day. A new product increases sessions by +1 session per member on average. Both of these are +1 session / member on average! One is much more unequal than the other. We want to catch that.
- 33. Privacy @ LinkedIn Framework to compute robust, privacy- preserving analytics
- 34. Analytics & Reporting Products at LinkedIn Profile View Analytics 34 Content Analytics Ad Campaign Analytics All showing demographics of members engaging with the product
- 35. • Admit only a small # of predetermined query types • Querying for the number of member actions, for a specified time period, together with the top demographic breakdowns Analytics & Reporting Products at LinkedIn
- 36. • Admit only a small # of predetermined query types • Querying for the number of member actions, for a specified time period, together with the top demographic breakdowns Analytics & Reporting Products at LinkedIn E.g., Title = “Senior Director” E.g., Clicks on a given ad
- 37. Privacy Requirements • Attacker cannot infer whether a member performed an action • E.g., click on an article or an ad • Attacker may use auxiliary knowledge • E.g., knowledge of attributes associated with the target member (say, obtained from this member’s LinkedIn profile) • E.g., knowledge of all other members that performed similar action (say, by creating fake accounts)
- 38. Possible Privacy Attacks 38 Targeting: Senior directors in US, who studied at Cornell Matches ~16k LinkedIn members → over minimum targeting threshold Demographic breakdown: Company = X May match exactly one person → can determine whether the person clicks on the ad or not Require minimum reporting threshold Attacker could create fake profiles! E.g. if threshold is 10, create 9 fake profiles that all click. Rounding mechanism E.g., report incremental of 10 Still amenable to attacks E.g. using incremental counts over time to infer individuals’ actions Need rigorous techniques to preserve member privacy (not reveal exact aggregate counts)
- 39. Problem Statement •Compute robust, reliable analytics in a privacy- preserving manner, while addressing the product needs.
- 42. Defining Privacy 42 CuratorCurator + your data - your data
- 43. Differential Privacy 43 Databases D and D′ are neighbors if they differ in one person’s data. Differential Privacy: The distribution of the curator’s output M(D) on database D is (nearly) the same as M(D′). Curator + your data - your data Dwork, McSherry, Nissim, Smith [TCC 2006] Curator
- 46. Keeping our data safe and secure for members Problem statement • We have a lot of data • Some may have PII data • How do we keep this secure? • Removing PII data • Tracking access
- 47. Policy: Keeping the data safe Solution through Technology • Meta data store • Tag all attributes in all tables • Know which fields are PII • Know which fields need protection • Audit access to data • Obfuscate data wherever possible
- 48. Tracking pedigree of data • Tables can be combined to create new tables • Automatically track pedigree of attributes and their PII value • Assess new attributes for PII as well • Have authors be accountable Name Type PII ? Name string Yes Age String Yes A1 string No A2 url No Name Type PII ? Name string Yes Adult boolean No B2 string No C1 number No Name Type PII ? Name string Yes B1 number No B2 string No B3 string No
- 49. Reflections
- 50. Fairness in ML • Application specific challenges • Conversational AI systems: Unique bias/fairness/ethics considerations • E.g., Hate speech, Complex failure modes • Beyond protected categories, e.g., accent, dialect • Entire ecosystem (e.g., including apps such as Alexa skills) • Two-sided markets: e.g., fairness to buyers and to sellers, or to content consumers and producers • Fairness in advertising (externalities) • Tools for ensuring fairness (measuring & mitigating bias) in AI lifecycle • Pre-processing (representative datasets; modifying features/labels) • ML model training with fairness constraints • Post-processing • Experimentation & Post-deployment
- 51. Explainability in ML • Actionable explanations • Balance between explanations & model secrecy • Robustness of explanations to failure modes (Interaction between ML components) • Application-specific challenges • Conversational AI systems: contextual explanations • Gradation of explanations • Tools for explanations across AI lifecycle • Pre & post-deployment for ML models • Model developer vs. End user focused
- 52. Privacy in ML • Privacy-preserving model training, robust against adversarial membership inference attacks • Privacy for highly sensitive data: model training & analytics using secure enclaves, homomorphic encryption, federated learning / on-device learning, or a hybrid • Privacy-preserving transfer learning (broadly, privacy-preserving mechanisms for data marketplaces)
- 53. Thank you Sofus A. Macskássy Data Science @ LinkedIn smacskassy@linkedin.com