Recently Dr Bruce Long elicited feedback on various academic networking platforms (Researchgate, Academia etc.) about his Ethics of AI course. There were several excellent responses and suggestions from academics about how to improve and augment the course.
One computer science and ethics postdoc had most excellent suggestions and probing questions, and the exchange (respondent de-identified) is included here.
A link to the draft is available here
The previous version of the draft had over 1000 views from mid 2019 to now (also still listed in Academia.edu).
Thanks for your response and suggestions. My replies in line with your list:
- Who are the target audience? BSc, MSc, PhD? Computer scientists, historians, mathematicians?
Second year to postgrad normative AI ethics students.
- How long is the course? It is really a lot of ground you want to cover
It's a semester course. It fits nicely into 13 weeks. I do not use all of the listed texts and articles. This is the draft version, so I include a wide range of additional readings for background.
- How do you define AI?
The course covers grades of AI and AGI, and I include machine supervised and unsupervised machine learning and deep learning systems along the usual lines.
We cover machine/deep learning from perceptrons through sigmoid neurons, and up to the current work on AGI and the use of intrinsic plasticity. However, the technical aspects I cover are only pursuant to ensuring that the philosophy and ethics scholars know enough to understand why a certain issue has a normative ethical dimension (there is some metaethics and applied ethics, but the course is mostly about normative ethics.) For computer science scholars a lot of the technical elements could be removed, as they would already have that knowledge.
- Why are you focusing on strong AI? It still does not exists and there are plenty of problems to discuss with current ML-based applications.
I can see how you'd get that impression. The Web part that you ask about and the software qualities link that you refer to are mostly about unsupervised deep learning type systems. The course is a philosophy and ethics course. Many students have a Dennettian position (as do I) which is less speculative and more cautious. However, many students are familiar with the work of Chalmers and Clark [and other scholars] on extended cognition and the singularity. This remains an important topic in cognitive science, normative ethics, and even among computer scientists...
- Also, among others, missing the link between the "ethics theory" and the implications for the system design (the block software qualities does not give a lot away)
You're only looking at a bare course outline. I have a current paper that covers the link between software qualities like interpretability and transparency, and [normative] ethical problems. I introduce the concept of the epistemic opacity dilemma (it's referred to in the draft outline currently as a paradox, but that's a little inaccurate). This is the only course I know of that does this (I was a full stack Web software engineer for many years.)
- How does "ethics of AI" fit to the topic of web based applications? (they seem to appear unexpectedly)
Well, as I am sure that you would be aware, trained deep learning [machine learning/neural net] systems are responsible for many on line Web applications - such as movie and product recommendation systems etc.
If you need to develop a course of your own and so would like some more details, I'd be happy to share a current full version of the outline. If you need someone to teach such a course - I was a normative ethics tutor and instructor at the University of Melbourne in 2017. I can easily conduct a course on line or in person if your institution needs it.
Best Wishes and Thanks Again,