Everyone in the field of AI is aware that some of the objectives of AI could pose risks. A few of the risks exceed just car accidents as automated vehicles are beta tested or the replacement of cubicle jobs by processes running in data centers. Various stories, such as Frankenstein, 2001: A Space Odyssey, Terminator, and Transcendence have made readers and movie watchers aware of some of the sequences of technological events that could lead to risks of significant magnitude and permanence.

From a risk management point of view, should ethics classes be taught universally? Should legislative steps be taken to require ethical information to accompany all technical presentations of AI approaches, designs, and implementations? Should universities require ethics classes for all high powered technologies such as AI?

Are there classes in AI ethics taught in high schools and universities yet?

If so, where? If not, why not?

One Possible Syllabus

In response to the query in the comment, here is one possible AI Ethics course syllabus. University deans could decide whether it should be an academic requirement for 2nd year students enrolled in their school.

  • Brief history of ethics in economics, law, science, and geopolitics
  • Triumphs in ethics applied to technology in the past
  • Ways current society is benefiting from past ethical integrity
  • Past negative effects resulting from ethical negligence
  • Team project: A corporate AI policy per board of directors request
  • Distinguishing plausible futures from artifacts of sci fi creativity
  • Ways of evaluating outcomes
  • Relationship between creators of a system and the system created
  • Developing assessments of cost, loss, value, and benefit functions
  • Essay assignment: Does humanity have a manifest destiny?
  • Dealing with predictive uncertainty in technological ethics
  • Legislative and judicial considerations
  • Considering career options in an ethical context
  • Final exam and team project due

Such courses and relevant studies do exist, but they are much more common at departments of Philosophy or similar places that don't really deal with the technical part of AI development. In an way, it's their job to discuss whatever hypothetical philosophical problems we can think of, so it's relevant for them.

Computer science departments typically do not offer anything similar (except when you take some side courses like philosophy and they suggest some "closer to your experience" topics for a a term paper, I had something similar in my student years). So in general, the closer you get to the real AI development, the less likely for you is to face this discussion.

Why so and what I think about it. Honestly speaking, I think the current situation is natural (i.e., unsurprising) in general, and I don't expect it to change (unless we have some power political-driven force from above). I don't think there is much incentive do discuss ethical issues of AI within core computer science/software engineering community.

First, practical AI development is currently focused on "weak AI", the systems that are percieved by the specialists more like regular machines -- think bridges or buildings. Yes, they can do harm if they collapse, but the solution (within this logic) is to make them robust and reliable, so there are no real ethical issues here apart from possible poor quality. Designing a sort of "mastermind AI" is not a goal of most AI researchers, so the problems discussed in fiction you mentioned are not directly relevant.

Second, I must say that great writers write great stories. We shouldn't consider their worlds as direct predictions of our future. Yes, some scientists are very concerned about AI going out of control, too (like late Stephen Hawking), but as technical people we aren't supposed to form our syllabi on the basis of vivid pictures painted by the artists.

Even your current examples like self-driving cars in my opinion don't introduce many ethical issues from the point of view of a developer. Obviously, a developer's job is to design AI as secure as possible, like a bridge designer's job is to design a robust bridge. Real ethical issues occur on higher levels, like policy making. If we have a hypothetical situation when an AI-driven car has to kill someone (think trolley problem), I think the job of an AI designer is to follow whatever regulation is enforced, and perhaps the AI system will have to pass certification like brakes or airbags.

(Added) Consider this. Suppose I need to design an AI-controlled car. Roughly speaking, the behavior of this AI will be based on training data (of "right" and "wrong" maneuvers) or on explicit rules. In the first case, my tech job is to make sure the input data is correct, and the resulting AI follows it closely. In the second case, I must comply with the rules derived by the specialists (since I am no expert in safe driving, I cannot make these rules). In both cases, there is very little room for my personal ethical choice.

I am not trying to take ethical responsibility off the shoulders of AI researchers. However, I sincerely believe that our AI systems are designed in such a way and exist in such an environment, where software engineers don't make "ethical" or "unethical" decisions. Though I would agree that such a course would be truly beneficial for the people engaged in the political side of this issue.

|improve this answer|||||

All the evidence that I have seen suggests that training in ethics has no detectable effect on the behaviour of the trained (!), and so although I sympathise with your goals, I believe the time would be better spent on technical training, the better to advance AI and, hopefully, downstream benefits like more tax revenue for ethics departments.

|improve this answer|||||

@Douglas great question, one so pressing I fear this answer will not be totally sufficient, but I hope to research more and write up a full report on my website.

Ethics classes are becoming more integrated into CS majors and I have anecdotally heard from both Bath, UK, and Stanford, US, Universities that these classes are being included.

Your proposed syllabus, for me, may be too specific. One needs to identify why certain philosophy classes maybe useful to these students. E.g. if we prescribe to the inherent dangers of AI: what kinds of ways of working can mitigate some of these risks?

Given this question I would advise identifying useful processes within design, business and technology methodologies that could help.

For example along with your CS classes you could also run debating classes to help students communicate their ideas, reflect, but crucially, to take the opposite stance and debate from there. This would demonstrate seeing from other peoples point of view and how one deconstructs an argument.

For those students more design lead, then design and UX critiques could be a useful way for a group to poke holes in an idea for it to be iterated on and made (hopefully) better, or more adequate at least. (There is also a lot to say about human centred design here which I won’t have time to go into.)

As a lot of my work comes back to ethics, I would also advise philosophy classes to reflect on concepts like: theory of mind, the greater good, virtue etc. Concepts that would imbue the student with a reflective questioning on their place in the world and therefore what they should bring into it.

Hope this is a useful start. Again this is all back of the napkin stuff which I'll write in full soon. To reiterate, I am not advocating all of the above necessarily but eluding to the question we should be asking, what is the world we are trying to create? And therefore how do we get there?

|improve this answer|||||

Not the answer you're looking for? Browse other questions tagged or ask your own question.