This week in the MIT AI and Ethics group, we discussed the role of ethics in AI and computer science education. In particular, we discussed two papers: an essay on integrating ethics in existing courses from Harvard, and a curriculum and exercises from SCU. Our goal at the end of the session was to provide a set of recommendations to the department, school of computer science, or even the greater institute on how to integrate ethics into education.

Students debate how to incorporate AI ethics education.
Students debate how to incorporate AI ethics education.

In the beginning session, we asked what it means to be responsible in the context of full-system engineering. For example, earlier this year, Uber reported the first pedestrian fatally due to a self-driving car. Despite some faults (no system is truly error proof), some group of people (likely management or manufacturer) decided the car was suitable to put on the road. In this example, who is responsible, and what does it mean to be responsible? During the exercise, some people brought up the analogy between self-driving car accidents and mishaps during surgery – if someone dies after a surgery it doesn’t mean that the inventor of that surgery is at fault. Other people mentioned that it’s difficult to put fault on a developer since the code in question is often the product of a larger system of teams and managers. Similarly, it is hard to pinpoint a single entity who should have foreseen the error. This lead to a discussion of how we can teach computer scientists how to foresee and combat critical edge cases.

We had a diverse set of people and discussions that emerged. In one group, someone mentioned a program at Stanford University where you can pair a CS degree with a minor in a humanities field. They brought up that perhaps MIT should require that students choose their general education courses to be more cohesive and directed (e.g., CS + ethics or CS+ policy).

This led to a debate about whether ethics should be integrated directly into CS courses, or taught separately. Many people disagreed with the integrated approach, noting that it is hard to teach ethics in a CS course. They brought up that ome professors of CS might not want to teach ethics and may brush it off, but having separate classes would allow ethics experts to teach them. However, many stated that having separate ethics courses may not be popular with students or teachers. From prior experience at MIT, one student mentioned that having separate required classes (like communication-intensive requirements at MIT) may cause students to end up being annoyed or not see the value in a course they must take to graduate.

A group of attendees from the Philosophy department brought up a middle-ground solution that they are currently implementing, that involves an ethics curriculum integrated into existing courses but taught by ethics experts. Their focus in on integrating ethics into fundamental design processes, such that it is a part of an engineering mindset from the outset. The curriculum focuses on value-sensitive design, examining stakeholders and affected population to form theories and make ethical design choices. This was contrasted this with a theory-based approach that starts off with a presupposed theory of ethics and then tries to implement it. Some of the participants will be testing out this course over IAP!

While we focused a lot on college-level education, it was also higlightlighted that ethics should be taught early and often in the curriculum. As students are first learning how to code, they should learn a design process (e.g. like writing tests first) that incorporates ethics as part of the workflow.

Another question that was raised was how to make ethics an ongoing educational discipline. It’s important that students realize that ethics is a constant practice and not just a skill that is learned and applied, e.g. “I got an A so I’m good at ethics.” This was further support for an integrated approach where ethics appears consistently throughout many courses, from problem sets to chapters in textbooks or course notes. We briefly discussed the natural extension of how to incorporate ethics education into professional workplaces. If ethics education is purely tackled in the undergraduate curriculum, then we may be neglecting a cohort of people currently in the workforce who could benefit more immediately. We discussed how a pathway for this may be an online course that leads to a professional certification.

Our next session on anonymity and privacy in the context of AI and ethics on Wednesday November 28th.