Successfully reported this slideshow.
Your SlideShare is downloading. ×

Disparate Impact Diminishes Consumer Trust Even for Advantaged Users

Ad

1
WIS
Web
Information
Systems
Disparate Impact Diminishes
Consumer Trust Even for
Advantaged Users
Tim Draws1,2, Zoltán Sz...

Ad

2
WIS
Web
Information
Systems
Trusting Persuasive Technology (PT)
+
+
References: Nickel & Spahn (2012); Purpura, Schwanda...

Ad

3
WIS
Web
Information
Systems
PT’s Disparate Impact?
• Machine learning research documents biases
and unfairness of many d...

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Ad

Check these out next

1 of 11 Ad
1 of 11 Ad
Advertisement

Disparate Impact Diminishes Consumer Trust Even for Advantaged Users

  1. 1. 1 WIS Web Information Systems Disparate Impact Diminishes Consumer Trust Even for Advantaged Users Tim Draws1,2, Zoltán Szlávik1,3, Benjamin Timmermans1,4, Nava Tintarev5, Kush R. Varshney4, Michael Hind4 t.a.draws@tudelft.nl https://timdraws.net 1IBM Center for Advanced Studies Benelux 2Delft University of Technology 3myTomorrows 4IBM Research 5Maastricht University
  2. 2. 2 WIS Web Information Systems Trusting Persuasive Technology (PT) + + References: Nickel & Spahn (2012); Purpura, Schwanda, Williams, Stubler, & Sengers (2011); Sattarov & Nagel (2019)
  3. 3. 3 WIS Web Information Systems PT’s Disparate Impact? • Machine learning research documents biases and unfairness of many different kinds – Usually results from biased data – Disparate impact: disproportionally negative effect on some user groups (e.g., women) • Disparate impact in PT  consumer trust? References: Barocas & Selbst (2016); Baeckström, Silvester, & Pownall (2018); Rossi (2019); Mullainathan, Noeth, & Schoar (2012); Ntoutsi et al. (2020); Toreini et al. (2020)
  4. 4. 4 WIS Web Information Systems Our Study RQ1: Disparate impact  consumer trust? RQ2: Advantaged ≠ disadvantaged users? Method: online between-subjects user study Use-case: personal finance PT
  5. 5. 5 WIS Web Information Systems Method: Procedure Step 1/2 All participants: • “AI Advisor” • General usage statistics • Baseline measurements – Trust, perceived personal benefit, willingness to use
  6. 6. 6 WIS Web Information Systems Method: Procedure Step 2/2 Depending on condition (1 out of 4): • Gender-specific statistics – No bias – Little bias – Strong bias – Extreme bias • Second round of measurements – Trust, perceived personal benefit, willingness to use 20% 20% 10% 10% 10% 20% 25% 15% 30% 10% 35% 5%
  7. 7. 7 WIS Web Information Systems Method • 489 participants – 49% male, 51% female – Randomly distributed over four conditions • Per participant: difference between baseline and second measurement – Change in trust – Change in perceived personal benefit – Change in willingness to use
  8. 8. 8 WIS Web Information Systems Results RQ1: Disparate impact  consumer trust? – H1a: Disparate impact decreases consumer trust. – H1b: Disparate impact decreases willingness to use. −0.50 −0.25 0.00 Control Little Bias Strong Bias Extreme Bias Condition Change in trust Difference between conditions χ2 = 25.06, p < 0.001 Difference between conditions F = 6.906, p < 0.001
  9. 9. 9 WIS Web Information Systems No evidence for interaction between condition and gender F = 2.094, p = 0.096 −0.8 −0.4 0.0 Control Little Bias Strong Bias Extreme Bias Condition Change in trust gender female male Results RQ2: Advantaged ≠ disadvantaged users? – H2a: Gender moderates disparate impact  personal benefit. – H2b: Gender moderates disparate impact  trust (see H1a). Interaction between condition and gender F = 8.525, p < 0.001 −2.0 −1.5 −1.0 −0.5 0.0 Control Little Bias Strong Bias Extreme Bias Condition Change in perceived personal benefit gender female male
  10. 10. 10 WIS Web Information Systems Discussion & Conclusion • Disparate impact in PT can decrease consumer trust and willingness to use • Despite users recognizing their respective (dis-)advantage, all users may lose trust in systems they use due to disparate impact t.a.draws@tudelft.nl https://timdraws.net
  11. 11. 11 WIS Web Information Systems References • Baeckström, Y., Silvester, J., Pownall, R.A.: Millionaire investors: financial advi- sors, attribution theory and gender differences. Eur. J. Financ. 24(15), 1333–1349 (2018). https://doi.org/10.1080/1351847X.2018.1438301 • Barocas, Solon and Selbst, A.D.: Big data’s disparate impact. Calif. Law Rev. 104(671), 671–732 (2016) • Mullainathan, S., Noeth, M., Schoar, A.: The Market for Financial Advice: An Audit Study. SSRN Electron. J. (2012). https://doi.org/10.2139/ssrn.1572334 • Nickel, P., Spahn, A.: Trust, Discourse Ethics, and Persuasive Technology. In: Persuas. Technol. Des. Heal. Safety; 7th Int. Conf. Persuas. Technol. 2012. pp. 37–40. Linköping University Electronic Press (2012) • Ntoutsi, E., Fafalios, P., Gadiraju, U., Iosifidis, V., Nejdl, W., Vidal, M.E., Rug- gieri, S., Turini, F., Papadopoulos, S., Krasanakis, E., Kompatsiaris, I., Kinder- Kurlanda, K., Wagner, C., Karimi, F., Fernandez, M., Alani, H., Berendt, B., Kruegel, T., Heinze, C., Broelemann, K., Kasneci, G., Tiropanis, T., Staab, S.: Bias in data-driven artificial intelligence systems—An introductory sur- vey. Wiley Interdiscip. Rev. Data Min. Knowl. Discov. 10(3), 1–14 (2020). https://doi.org/10.1002/widm.1356 • Purpura, S., Schwanda, V., Williams, K., Stubler, W., Sengers, P.: Fit4Life: The Design of a Persuasive Technology Promoting Healthy Behavior and Ideal Weight. In: Proc. SIGCHI Conf. Hum. factors Comput. Syst. pp. 423–432 (2011) • Rossi, F.: Building trust in artificial intelligence. J. Int. Aff. 72(1), 127–133 (2019) • Sattarov, F., Nagel, S.: Building trust in persuasive gerontechnology: User- centric and institution-centric approaches. Gerontechnology 18(1), 1–14 (2019). https://doi.org/10.4017/gt.2019.18.1.001.00 • Toreini, E., Aitken, M., Coopamootoo, K., Elliott, K., Zelaya, C.G., van Moorsel, A.: The relationship between trust in AI and trustworthy machine learning tech- nologies. FAT* 2020 - Proc. 2020 Conf. Fairness, Accountability, Transpar. pp. 272–283 (2020). https://doi.org/10.1145/3351095.3372834

×