Implementing Ethics in AI
Download as PPTX, PDF2 likes878 views
The document discusses the implementation of ethics in artificial intelligence, highlighting key ethical risks such as bias, discrimination, and erosion of privacy. It emphasizes the need for AI companies to establish their own ethical frameworks until regulations catch up, and notes that developers lack formal tools to conduct ethical analyses. Recent studies provide insights into current practices and propose new methods for integrating ethics in AI design.
![JYU. Since 1863. 23
Accountability
“It’s just a prototype”
“We have talked about the risks of decision-making support
systems, but it doesn’t really affect what we do”
"It's really important how you handle any kind of data... that you
preserve it correctly, among researchers, and don't hand it out to
any government actors. [...] I personally can't see any way to harm
anyone with the data we have though."
19.11.2019
Source: Vakkuri V., Kemell KK., Abrahamsson P. (2019) Implementing Ethics in AI: Initial Results of an Industrial Multiple Case Study. In: Franch X., Männistö T.,
Martínez-Fernández S. (eds) Product-Focused Software Process Improvement. PROFES 2019. Lecture Notes in Computer Science, vol 11915. Springer, Cham,
Author’s copy available at arxiv.org/abs/1906.12307](https://image.slidesharecdn.com/cewit-2019-abrahamsson-ethics-ai-public-final-191119145259/85/Implementing-Ethics-in-AI-23-320.jpg)
![JYU. Since 1863. 24
Responsibility
"Nobody wants to listen to ethics-related technical stuff. [...]
It’s not relevant to the users”
“What could it affect... the distribution of funds in a region,
or it could result in a school taking useless action... it does
have its own risks, but no one is going to die because of it”
19.11.2019
Source: Vakkuri V., Kemell KK., Abrahamsson P. (2019) Implementing Ethics in AI: Initial Results of an Industrial Multiple Case Study. In: Franch X., Männistö T.,
Martínez-Fernández S. (eds) Product-Focused Software Process Improvement. PROFES 2019. Lecture Notes in Computer Science, vol 11915. Springer, Cham,
Author’s copy available at arxiv.org/abs/1906.12307](https://image.slidesharecdn.com/cewit-2019-abrahamsson-ethics-ai-public-final-191119145259/85/Implementing-Ethics-in-AI-24-320.jpg)
Implementing Ethics in AI
- 1. Implementing Ethics in AI Professor Pekka Abrahamsson 15th International Conference on Emerging Technologies for a Smarter World (CEWIT 2019) Nov-6th, 2019, Stony Brook, NY, USA JYU. Since 1863. 119.11.2019
- 2. JYU. Since 1863. 2 Agenda • Motivation • State-of-the-art tools and methods • Empirical observations • Conclusions • References 19.11.2019
- 3. JYU. Since 1863. 3 Failing AI • “San Francisco Bans Facial Recognition Technology” Source: New York Times, 2019, bit.ly/2Qs9vKi • Amazon scraps secret AI recruiting tool that showed bias against women Source: Reuters, 2018, bit.ly/2Kw77hK • A Popular Algorithm Is No Better at Predicting Crimes Than Random People Source: The Atlantic, 2018, bit.ly/2OrULbA 19.11.2019
- 4. JYU. Since 1863. 4 Key ethical risks from the corporate viewpoint • Bias and discrimination • Erosion of Privacy • Poor accountability • Workforce displacement and transitions 19.11.2019 Source: Schatsky, D., et al 2019. Can AI be ethical? Why enterprises shouldn’t wait for AI regulation. Deloitte report
- 8. Source: https://sloanreview.mit.edu/article/every-leaders-guide-to-the-ethics-of-ai/
- 11. JYU. Since 1863. 11 What should the leaders do? • “Until regulations catch up, AI-oriented companies must establish their own ethical frameworks” Source: Sloan Management Review, 2018, bit.ly/2NXF0Ky 19.11.2019
- 12. Source: Schatsky, D., et al 2019. Can AI be ethical? Why enterprises shouldn’t wait for AI regulation. Deloitte report
- 13. JYU. Since 1863. 1319.11.2019
- 14. 14 Source: Sheard, S.A., 1997, August. The frameworks quagmire, a brief look. In INCOSE International Symposium (Vol. 7, No. 1, pp. 726-733).
- 15. Concerns drawn from Vakkuri, V. and Abrahamsson, P., 2018, June. The Key Concepts of Ethics of Artificial Intelligence. In 2018 IEEE International Conference on Engineering, Technology and Innovation (ICE/ITMC) (pp. 1-6). IEEE. Author’s version available online at https://arxiv.org/abs/1809.07027
- 16. JYU. Since 1863. 1619.11.2019 Conceptualization of the Relations Between Currently Discussed AI Ethics Construct Vakkuri et al. 2020. Ethically Aligned Design of Autonomous Systems: Industry viewpoint and an empirical study. To appear in the proceedings of the Transport Research Arena (TRA 2020) arXiv preprint https://arxiv.org/abs/1906.07946.
- 17. JYU. Since 1863. 17 AI Ethics discussion • Long-standing area of study Old scenarios slowly becoming reality • Recent discourse has centered around a few central constructs Transparency, Accountability Responsibility, Fairness… Trustworthiness • Focus on high-level principles Discussion largely conceptual Little empirical data exists 19.11.2019
- 18. JYU. Since 1863. 1819.11.2019 Tangible Research Framework Vakkuri, V., Kemell, K.K. and Abrahamsson, P., 2019. AI Ethics in Industry: A Research Framework. To appear in the proceedings of the 3rd Seminar on Technology Ethics, Turku, Finland, arXiv preprint arXiv:1910.12695.
- 19. Empirical Study Results JYU. Since 1863. 1919.11.2019
- 20. JYU. Since 1863. 2019.11.2019 Vakkuri et al. 2020. Ethically Aligned Design of Autonomous Systems: Industry viewpoint and an empirical study. To appear in the proceedings of the Transport Research Arena (TRA 2020) arXiv https://arxiv.org/abs/1906.07946.
- 21. JYU. Since 1863. 2119.11.2019 Vakkuri et al. 2020. Ethically Aligned Design of Autonomous Systems: Industry viewpoint and an empirical study. To appear in the proceedings of the Transport Research Arena (TRA 2020) arXiv preprint https://arxiv.org/abs/1906.07946.
- 22. JYU. Since 1863. 2219.11.2019 Vakkuri et al. 2020. Ethically Aligned Design of Autonomous Systems: Industry viewpoint and an empirical study. To appear in the proceedings of the Transport Research Arena (TRA 2020) arXiv preprint https://arxiv.org/abs/1906.07946.
- 23. JYU. Since 1863. 23 Accountability “It’s just a prototype” “We have talked about the risks of decision-making support systems, but it doesn’t really affect what we do” "It's really important how you handle any kind of data... that you preserve it correctly, among researchers, and don't hand it out to any government actors. [...] I personally can't see any way to harm anyone with the data we have though." 19.11.2019 Source: Vakkuri V., Kemell KK., Abrahamsson P. (2019) Implementing Ethics in AI: Initial Results of an Industrial Multiple Case Study. In: Franch X., Männistö T., Martínez-Fernández S. (eds) Product-Focused Software Process Improvement. PROFES 2019. Lecture Notes in Computer Science, vol 11915. Springer, Cham, Author’s copy available at arxiv.org/abs/1906.12307
- 24. JYU. Since 1863. 24 Responsibility "Nobody wants to listen to ethics-related technical stuff. [...] It’s not relevant to the users” “What could it affect... the distribution of funds in a region, or it could result in a school taking useless action... it does have its own risks, but no one is going to die because of it” 19.11.2019 Source: Vakkuri V., Kemell KK., Abrahamsson P. (2019) Implementing Ethics in AI: Initial Results of an Industrial Multiple Case Study. In: Franch X., Männistö T., Martínez-Fernández S. (eds) Product-Focused Software Process Improvement. PROFES 2019. Lecture Notes in Computer Science, vol 11915. Springer, Cham, Author’s copy available at arxiv.org/abs/1906.12307
- 25. JYU. Since 1863. 25 Responsibility • The developers could, when asked, think of ways the system could negatively affect various stakeholders • These were not addressed formally • Developers did not have tools/methods to conduct formal ethical analyses in order to tackle such issues 19.11.2019 Source: Vakkuri V., Kemell KK., Abrahamsson P. (2019) Implementing Ethics in AI: Initial Results of an Industrial Multiple Case Study. In: Franch X., Männistö T., Martínez-Fernández S. (eds) Product-Focused Software Process Improvement. PROFES 2019. Lecture Notes in Computer Science, vol 11915. Springer, Cham, Author’s copy available at arxiv.org/abs/1906.12307
- 26. JYU. Since 1863. 26 Summary • AI is spreading everywhere • People voice out their concerns publicly • AI Ethics is a growing concern for companies • High level guidelines and principles exist but they fail to provide actionable advice to developers • AI is developed by software engineers who need tangible methods to address the concerns • Our studies provide empirical insight about the current state of practice • First AI Ethics in Software Design methods are being published in 2020 19.11.2019
- 27. JYU. Since 1863. 27 Our AI Ethics research results are made public in Arxiv.org 19.11.2019 The Key Concepts of Ethics of Artificial Intelligence arxiv.org/abs/1809.07027 AI Ethics in Industry: A Research Framework arxiv.org/abs/1910.12695 Implementing Ethics in AI: Initial results of an industrial multiple case study arxiv.org/abs/1906.12307 Ethically Aligned Design of Autonomous Systems: Industry viewpoint arxiv.org/abs/1906.07946 Ethically Aligned Design: An empirical evaluation of the RESOLVEDD-strategy arxiv.org/abs/1905.06417
- 28. JYU. Since 1863. 2819.11.2019