|
MF: What motivated you and Jamie to develop this course?
JS: Jamie and I are working at the intersection of biodiversity conservation, climate change, infectious diseases, and sustainable development. We’ve often worked with fairly advanced statistical analyses, geospatial modeling techniques etc. and wanted to look at the potential of AI to contribute to helping address global good or the sustainable development goals.
We were cognizant of the fact that AI also comes with a lot of trade-offs. There might be some potential for it to help address global good—to help alleviate poverty, alleviate hunger, address climate change, secure clean drinking water, and access to clean air. But, at the same time it can contribute to issues around climate change, create more demand for electricity, more demand for water, more demand for data centers, it can propagate biases that might reinforce issues of inequity across racial, gender, and socioeconomic lines. That tension is interesting, particularly because so much of the public discourse has been so polarized: either AI is portrayed inherently as the best thing that’s ever happened for humanity, or as something that’s going to doom us all.
When we were developing the course, we both thought that there are lots of cases where it is quite useful, and also probably lots of cases where it isn’t. So we wanted to ask: How can we think critically about where and how to apply AI in a way that’s advancing global good rather than hindering us from meeting that abstract goal?
MF: Would you say that the course was an effort to teach AI literacy?
JS: It definitely was. The students had a combination of technical skills development and information about the present day landscape of sustainable development. We split the lectures into those that were focused on how to build a chatbot, how to use image recognition software and audio recognition software, and how to use AI to make an automation type of workflow. The other lectures asked how we can apply that knowledge to specific sustainable development goals.
MF: What were the main goals of the class? What did you want your students to walk away with?
JS: The first learning goal was to give students a familiarity with currently available AI tools and how we can apply them. So rather than just using a chatbot, how do we build a chatbot? Or how do we build an automation workflow? Or how do we build something that does image recognition?
A second learning goal was to give them background information about the Sustainable Development Goals, how to think about them in a critical sense, the history of these goals, how they are structured, and how they have guided international action and international development.
And then the third goal was to think critically about the intersection of AI and the Sustainable Development Goals, particularly with a focus on areas where it could provide benefits or help advance those specific goals or areas. We also asked where it might not enable progress towards meeting those goals, or actually hinder progress towards those goals.
The last goal was how to assess the accuracy and sustainability of these tools, as well as monitoring any biases that might be coming through. So it was really about both developing skills and then thinking critically about how to use them in the context of sustainable development.
MF: For the final group project, students worked with an external NGO. How did you establish these collaborations and what did the projects look like?
JS: Early in the semester, we paired students with organizations that are working to implement the Sustainable Development Goals. It was a combination of organizations Jamie and I had worked with, like the World Bank and the Natural Capital Project, and local partners that we found through the Program in Community-Engaged Scholarship (ProCES), like New Jersey Future, the Camden Dream Center, Food Bank of South Jersey, and Sustainable Princeton.
The way we pitched it to the partners was to recognize that they’ve probably had some conversations in their organization about using AI to do a particular thing, but at the same time limited resources to explore and implement their idea. So we asked them, “What’s that project that you’re curious about doing with AI?”
For example, the Food Bank of South Jersey said that they’d often get calls after hours from partners like Costco or Wegmans, trying to deliver food or donate food. They wanted a tool to manage off-hours requests. The students worked to develop a prototype that would take in information from whoever it was, log it in a spreadsheet, and send out confirmation emails. It was an automation and chatbot integrated together.
MF: So, obviously, in this course, AI tools were an integral part of class work and you encouraged students to use it. But you still wrote an AI policy. How did you draw that line between desirable and undesirable uses of AI?
JS: One of the easiest ways to build an AI tool is by asking a different AI tool, “how do I build this tool?” We didn’t care as much if students were learning the code necessary to create their prototype, we just wanted them to make something that works. But we also wanted to design assignments that would be more difficult to have AI write.
In the final project there was a group component, i.e. the AI prototype, a report, and the in-class presentation. And in addition, there was an individual paper component where we asked students to describe what had happened over the course of the semester: what was the initial ask from the partner, what were the obstacles you encountered, how did you address them as a team, how did your team decided to divide up the work amongst different people, and what skills you developed over the course of the semester?
We also asked them to reflect on how the class caused them to rethink their different perceptions on AI. That information would be difficult to generated by AI. We hoped it would cause students to think critically about what they were engaging in, and that it would let us see whether or not they were really engaging with the project. We thought that that structure of the assignment would be less conducive to them just using generative AI.
MF: What did you learn about Princeton students’ ideas about AI by teaching this class?
JS: I think the overwhelming sentiment that the students came into the course with was a skepticism of the utility of AI for global good. At the same time, there was a sense that it’s inevitably going to permeate many dimensions of society. So the initial mindset was, “this is something I want to learn more about because I think it’s inevitable, even if I don’t think it’s a good idea.“
Over the course of the semester the one thing that I think most groups came to realize is that AI as an abstract concept can be quite problematic for global good, or for reaching the sustainable development goals. But AI narrowly defined can be quite good. A lot of the students who worked in these group projects had, like, really clear examples at the end. They often came to the conclusion that they had built a tool that made a positive impact, at least in a local context, or at least in a specific context.
I think there’s still a lot of larger societal concerns for students, for instance about who has control over AI technologies, about persistent biases in AI technologies, about potential risks, and about lacking the proper governance or regulatory structures. I think a lot of students are thinking really critically about who holds the levers of power when it comes to how these technologies are being deployed, in addition to just, like, whether or not the technology is good or bad.
|