”Trust in Numbers” Ethical (and practical) Approach to Identity – Driven AI/Machine Learning
“Trust in Numbers” Ethical (and Practical) Approach to Identity-Driven AI/Machine Learning
Convener: Mike Kiser
Notes-taker(s): Matt Domsch
Tags for the session - technology discussed/ideas considered:
Discussion notes, key understandings, outstanding questions, observations, and, if appropriate to this discussion: action items, next steps:
Matt Domsch (notes)
**** Slides and academic paper available late November after conference publication
Conference link: https://digitaleweltmagazin.de/digicon/program/
- Well-being - defining your ethical stance. Online Ethics Canvas. Who might you harm and how?
- Accountability - to whom? We often cede accountability and control to technology. "Because the algorithm said so" isn't enough. Users must also hold you accountable.
- Transparency - Users must be able to see how you got to your answer. Explain the "why".
- Fairness - Bias and things you assume.
- User Data Rights - best current practice
Wendell: Similar to Ethical OS? and many others? John Rolls.
Adrian: like privacy policies? There are so many to choose from.
Wendell: If you'd go before an Institutional Review Board(IRB), this qualifies.
Determine the relative strength or weaknesses of an approach, compared to other algorithms, and compared to your own position over time.
Wendell: fairness can't be expressed as a fixed point.
Adrian: will run a session on ML in medicine later today. None of this applies, because Medicine is science, nothing in secret.
The problem is we're turning it into a trade secret and business model, with ethics whitewashing afterwards.
Wendell: this is a way to pierce the trade secret system.
Adrian: but that's not the way open science should work. You've already ceded the ground. We're pre-judging the outcome of how money is made vs how science is done.
Wendell: how is this related to other ethical sourcing: blood diamonds, oil money, coffee, ...
Mike: is there another measurement method?
Adrian: Efficiency? Are we reducing friction only to have to control for bias?
Wendell: IRBs serve as a "confessional effect".
Wendell: is this like Mayak's trancendental effect, beyond human comprehension? Formal systems theory from MIT in the 1970s.
This is pretty subjective. Can we build a more objective measuring stick for each of these dimensions?
Similar to the Me2B Harms discussion yesterday.
Adrian: similar to introducing apps into an industry. 2 models:
1. mininmum bar, do no evil, gives maximum flexibilty to any business.
2. Consumer Reports style of strictly objective criteria
Ethics is neither a race the bottom, nor is it objective through standards and independent entities.
Institute for the Future Ethical OS.
Kantara is one of 3 FISMA certifiers.
Wendell: there are other maturity frameworks that this is similar to this work.
Design thinking 5 levels.
Adrian: Twitter #darkpatterns