Thread by @WellsLucasSanto

Let's goooo!!! The second of two papers on AI education is coming up in a bit. As an AI educator focused on inclusion and co-generative pedagogy, I'm *really* excited for this talk on exclusionary pedagogy. Will tweet some take-aways in this thread:

https://twitter.com/FAccTConference/status/1369649782759768064

First, a mention for those who don't know, I've been a CS educator since 2013, and in 2017 I moved into specifically being an AI educator, focusing on inclusive, accessible, and culturally responsive high school curriculum, pedagogy, and classroom experiences. Informs my POV

.@rajiinio starts the talk off by mentioning that there's an AI ethics crisis happening & we're seeing more coverage of the harms of AI deployments in the news. This paper asks the question, "Is CS education the answer to the AI ethics crisis, or actually part of the problem?" 🤔

They pointed to 3 areas in existing CS pedagogy that led to today's ethics crisis. Emphasizes that a lot of CS pedagogy today is incredibly individualized, aimed with equipping the sole computer scientist merely with tools to "fix" a technical problem as an "individual savior".

They mention a need for more epistemological pluralism, which is lacking in existing education, which values programmatic intelligence. This leads to the trap of "techno-solutionism" (that fixes only need to be technical in nature), which is currently heightened.

For their research, they looked at 254 ethics courses from 132 universities, mostly from CS, but some from HSS (humanities and social sciences). Found "mechanisms of exclusion," that there was a tendency for CS education to push aside HSS to solve the problem on its own.

Through their survey of courses, they found that participants of the CS discipline tried to isolate themselves other disciplines who could actually help, because they didn't value other schools of thought as valid ways of knowing or lack of interest from learning from each other.

@rajiinio, on "mechanisms of exclusion": "Disciplines don't *value* each other's way of knowing ... disciplines don't talk to each other, and there's a lack of translation to understand and collaborate with the other group."

On methodological dogmatism: Only one course even looked at how different disciplines had different strategies or methodologies for arriving at a conclusion or "truth". CS courses might only focus on HSS approaches' weaknesses (not strengths), and not their own weaknesses.

On lack of joint outputs: only 5 courses even allowed for cross-disciplinary teaching. Point to how many of these courses were ABET required or required prohibitive prerequisites which led to them being closed courses available for students of those disciplines only.

On siloed citations: CS teachers would often only assign CS authors, HSS teachers would often only assign HSS authors. "The same case study might be used in two courses, with completely different conclusions." (!!!)

@rajiinio: "AI ethics is inherently interdisciplinary" ... "Contrary to popular belief, CS can't solve the ethics crisis on its own. Exclusionary behavior results in a loss of values, assumptions, and methods that make the field unable to address its problems."

@rajiinio: "The expansion of the AI field is required to solve the AI ethics problem, but this exclusionary behavior narrows the lens of the field in a way that actually prohibits it from being able to think critically about how to address its problems."

They look to a shift in the climate change literature as inspiration, about how it originally looked only at greenhouse gases, but expanded to look at the social impacts and challenges, driven by students, to tackle climate change.

One possible approach to do this shift: demonstrate the positive outcomes of collaborative pedagogy. Allow students to interact with those from other departments, work on projects together, encourage cross-disciplinary collaborations.

Another approach: "educate students on frameworks of interventions based around existing problems, not anchored on the skills of those assumed to be in the position to address the problems" (aka present different skillsets)

Third approach: Discuss the various stakeholders when exploring the curriculum for AI ethics--make it clear that different stakeholders exist, beyond technologists. (This is reminiscent of the use of Persona Cards from an earlier talk, to get students to think about those POVs.)

Fourth approach: Work directly with the populations affected.

Fifth approach: Assess their own disciplinary limitations & where their disciplines can't solve the problems at hand, to encourage bringing in methods and approaches from other disciplines.

Q for Deb on who's teaching this work (CS ethics). Deb mentions a paper from @cfiesler specifically on looking at that. @rajiinio mentions self-critical approaches that should be used by CS professors to critique their own discipline, even if the course is taught from 1 POV.

@rajiinio mentions again about the climate science field, where much of the work early on was looking at the "technical" side of greenhouse gases, but soon realized that the economics, social aspects were needed. We can take inspiration from that with teaching CS ethics.

A through-line in this work is that we have to emphasize the shortcomings of our discipline, whether it's in being self-critical when teaching, or acknowledging that our sole discipline's methods may not be able to solve the problem at hand. That's how we motivate collaboration.

Related, I wanted to plug in a thread here that Deb & many others have been commenting on, about how even within our community there's a need for a FAccT dictionary and cross-discipline translation guide. Thinking about that re: language for collaboration

https://twitter.com/rajiinio/status/1369450790398607361?s=20

Also, re: incorporating the perspectives of different stakeholders into the classroom, this is something the first AI education paper did with "Persona Cards". See this thread for a lil more on that approach (and check out their paper!):

https://twitter.com/WellsLucasSanto/status/1369052511278002176?s=20

Q for @rajiinio on advice to educators on teaching AI case studies that have harms for underrepresented students that could be traumatic. She points to the work of @DrDesmondPatton and the SAFE Lab at Columbia as an example of folks who have done this teaching well.

I've got *lots* of thoughts on this, especially with creating classrooms environments that foster safety for the students & acknowledge the students as full people & creates space to have these sorts of dialogues inclusively. Rooted in teaching w/ justice & empathy in mind.