FACT-IR. Fairness, Accountability, Confidentiality and Transparency in Information Retrieval

Date: December 4, 2019
Venue: UiS, Stavanger, Norway

Please cite, link to or credit this presentation when using it or part of it in your work.

#InformationRetrieval #IR #Accountability #Confidentiality #Fairness #Transparency

Transcript

1. FACT-IR Fairness, Accountability, Confidentiality and Transparency in Information Retrieval Darío Garigliotti University of Stavanger, Norway December 4th, 2019
2. Outline • An introduction to responsible Information Retrieval • Fairness in IR • Accountability in IR • Confidentiality in IR • Transparency in IR • Responsible IR and the IR community 2
3. An introduction to responsible Information Retrieval 3
4. Information Retrieval systems • Information Retrieval (IR) systems deal with the organization, curation, and promotion of most of information being consumed today 4
5. Information, information, information everywhere! • IR systems connect people with information • IR systems reflect the content and interaction data, and the impact where they are used • IR systems are shaping not only the information consumption patterns, but also the social interactions 5
6. The need for responsible IR • Due to these social and political implications, it is becoming evident that many issues are concerning all aspects of IR system development and deployment - Gaps in information access across communities - Misinformation, polarization - Collection of personal data - Opaque methods for decision making • Since nowadays a variety of IR systems are used everywhere, those issues have potentially a wide ranging impact 6
7. FACT-IR • In this presentation, we describe four focus areas or main concepts on responsible IR: - Fair IR - Accountable IR - Confidential IR - Transparent IR 7
8. Some examples 8
9. Some examples 9
10. FACT-IR • These examples illustrate concerns about bias, unfair representation, harm, misinformation, privacy • F,A,C,T concepts are multi-dimensional • In particular, they depend on multiple stakeholders • The concepts are often related in a given scenario 10
11. Fairness in Information Retrieval 11
12. Fairness in IR • IR systems should avoid discrimination across people and communities 12
13. Fairness in IR 13
14. Fairness in IR 14
15. Fairness in IR • IR systems should avoid discrimination across people and communities - Avoid unfair conclusions even if they appear true - Avoid discrimination even when sensitive attributes are removed - Avoid bias even under vox populi, ensure diversity 15
16. Fairness in IR • Defining fairness is a complex, multi-dimensional problem - Domain dependent - Time dependent - Stakeholder dependent 16
17. Fairness in IR • Research directions and challenges: - Definition of fairness - Metrics and evaluation - Intervention criteria - Data availability 17
18. Accountability in Information Retrieval 18
19. Accountability in IR • IR systems should be able to justify their actions, and to be reliable at all times 19
20. Accountability in IR 20
21. Accountability in IR 21
22. Accountability in IR • Research directions and challenges: - Definition and criteria - Intervention criteria - Metrics and evaluation - Agency, freedom 22
23. Accountability in IR • Impact: - Short- and long-term influence of information in users - Representation of people in the information - Emotional impact in developers, content moderators, data annotators 23
24. Accountability in IR • Interplay with other concepts: - Fair and harmless - Accountable and transparent - Tension with confidentiality 24
25. Confidentiality in Information Retrieval 25
26. Confidentiality in IR • The output or actions of IR systems should not reveal secrets 26
27. Confidentiality in IR 27
28. Confidentiality in IR 28
29. Confidentiality in IR • Research directions and challenges: - Protection scope - Confidentiality criteria beyond topical - Architecture and methods - Metrics and evaluation 29
30. Confidentiality in IR • Relation with other concepts: - Biases and sensitive data in different demographics - Confidentiality criteria in different stakeholders • Tension with transparency • Awareness 30
31. Confidentiality in IR 31
32. Confidentiality in IR 32
33. Confidentiality in IR • Impact: - By better protecting sensitive content, more information can be made available - Increase trust from individuals and organizations 33
34. Transparency in Information Retrieval 34
35. Transparency in IR • IR systems should be able to explain why and how the results are obtained 35
36. Transparency in IR • Explaining decisions made by the system, like retrieved search results, recommendations, answers - In particular for sensitive decision making • Tracing origin of results • Clarifying results such that they are trustworthy 36
37. Transparency in IR • Research directions and challenges: - Definition, criteria, purpose - Intervention criteria - Metrics and evaluation - Tension with confidentiality 37
38. Transparency in IR • Impact: - Increase trust from individuals and organizations - Contribute to accountability - Contribute to reproducibility 38
39. Responsible IR and the IR community 39
40. Impact in IR community • Definitions, criteria, and methodologies for the F,A,C,T concepts should consider the multiple stakeholders • IR research community should aim for diversity • Transparency could improve teaching and reproducibility in IR • There may be resistance from some stakeholders 40
41. Broadening the IR community • IR community needs to promote the collaboration with other disciplines, like social sciences, psychology, economics, and law • Also, there should be collaborations with governmental institutions and organizations, regarding ethical frameworks and regulations 41
42. General Data Protection Regulation (EU) 2016/679 (GDPR) • A regulation in EU law on data protection and privacy for all individual citizens of the European Union (EU) and the European Economic Area (EEA) • Some key parts: - Clear, easy conditions to give and to freely withdraw consent - Right to data access - Right to erasure of personal information - Timely notification in case of security breach 42
43. Events in FACT-IR research • Some workshops related with FACT-IR (recently co- located with the SIGIR 2019 Conference): - FACTS-IR 2019 Workshop on fairness, accountability, confidentiality, transparency, and safety in IR - EARS 2019 - The 2nd International Workshop on ExplainAble Recommendation and Search - NewsIR'19 - 3rd International Workshop on Recent Trends in News Information Retrieval 43
44. Some references • James Allan et al. Report from the third strategic workshop on information retrieval in Lorne (SWIRL 2018). SIGIR Forum, 52:34–90, 2018. • Ricardo Baeza-Yates. Bias on the web. Communications of the ACM, 61(6), 2018. • EU. Regulation (EU) 2016/679 of the European Parliament and of the Council of 27 april 2016 on the protection of natural persons with regard to the processing of personal data and on the free movement of such data, and repealing directive 95/46/ec (General Data Protection Regulation). Official Journal of the European Union, L119:1–88, 2016. • Rishabh Mehrotra et al. Auditing search engines for differential satisfaction across demographics. In Proceedings of the The Web Conference, pp. 626–633, 2017. • Alexandra Olteanu et al. (Eds.) FACTS-IR: Fairness, Accountability, Confidentiality, Transparency, and Safety in Information Retrieval. To appear in SIGIR Forum, 2019. • Terrance DeVries et al. Does Object Recognition Work for Everyone? In Proceedings of CVPR Workshops, pp. 52-59, 2019. 44