OER and Accessibility AI Hackathon

The edtech landscape, particularly with generative AI, is evolving so quickly that developers, let alone educators, are struggling to keep up. Despite this, a group of innovative build4good college interns and mentors/experts took up the challenge last month.

For two days in our Washington DC offices, diverse teams from the US and the UK—comprising student interns, educators, developers, designers, and data scientists—gathered to create edtech tools to use open educational resources (OER) more efficiently and better serve students with disabilities.

This inaugural OER, accessibility, and inclusion AI hackathon employed design-thinking processes and teamwork to inspire innovation and build open-source edtech prototypes. Participants tapped into OER Commons, a public library of instructional materials, for various subjects from math, geography, biology and history across grade levels. The event also aimed to spark interest in public interest technology careers among college students in the build4good internship program, a New America initiative for tech-focused students who want to apply technology to solve social problems.

The core challenge: Could these teams make digital educational materials more accessible and adaptable, especially for learners with disabilities, using AI tools? These questions arose from our existing work at New America on accessibility and edtech, and open educational resources,

The hackathon was designed to explore if AI could solve long-standing challenges with OER. Among them: could AI help curate and personalize OER content efficiently so that resources became easier to find? Could it help educators to evaluate OER content? Could it support educators in adapting it to diverse learner profiles, particularly those with disabilities?

Anyone who is working in education today knows that simply saying the word “AI'' can lead to both anxiety and excitement—and both emotions were present in the room on these hot June days when the hackathon was underway. Participants from AI-for-Education.org, CAST, Fab, Inc, ISKME, Oak National Academy, OpenSciEd, Pennsylvania Association of Intermediate Units OER, Team4Tech, Understood.org were at the hackathon, while Washington State OER Hub contributed materials. Student interns received a crash course in everything from the importance of open content to universal design for learning principles.

Five teams were formed, each tackling a specific challenge. Using big flipbooks of paper, they first developed personas of various individuals to design for, such as an older teacher with multilingual students or a visually impaired student. They then moved to more intense work, still leaning into the discussions but now with laptops open, tapping away, and big screens looming above, showing lines of source code.

One team grappled with how to use voice-assisted technology paired with a tool like ChatGPT to help sort and search for educational materials that visually impaired teachers and students might use. They were able to figure out how to use chatbots to respond in languages other than English.

Another team tackled a common website problem: What can be done about all the images and photographs that accompany text but cannot be read by a screen reader for the visually impaired? When testing the current built-in readers, one of the build4good interns became alarmed. She tested out a commonly used screen reader and the robot was having to sound out every line and every letter in a long line of a URL embedded within a webpage. “This is awful,” she said. Her teammates agreed. They explored using AI to automatically create alt-text descriptions that teachers could be embedded into curricular materials. They soon realized that images can be complicated, especially if they are images that help to explain scientific concepts with detailed diagrams.

As the hackathon progressed, teams gained a greater appreciation for making instructional materials more accessible and how AI could help. In the final hours, all feverishly worked on their prototypes, with multiple side conversations about everything from the pros and cons of college courses, the shortages of computer science teachers in high schools, the biases but also the advantages within AI, and what it looked like to organize one’s thoughts into something someone else could understand.

Before long 3:30 pm arrived, and it was showtime. Teams presented their prototypes with new questions now buzzing in their heads about how to cope with even more specific challenges. One team, for example, demonstrated an alt-text generator prototype for PDF educational resources that were not screen-reader friendly (or, in other words, educational resources that a text-to-speech generator could not easily read out loud in ways that would make sense to an average human). Despite challenges yet to be solved, such as ensuring image readers don’t reveal answers to students when they are taking exams, the teams made progress.

Participants praised the experience. One build4good intern said, "It was great to actually build something." Another intern noted, “... before this I wasn't aware of the extent of the possibilities and ways we can leverage AI; conversely, I didn't fully consider the possible consequences that AI can further increase disparities.”

One educator participant was not sure how she would make a difference. “However, I quickly learned the importance of understanding what is needed for the project's success. With the help and expertise of the interns and programmers, my voice contributed to creating a useful and necessary tool for visually impaired students and teachers.” Another mentor/expert commented, “The project share-outs at the end were amazing -- getting to see what the interns had whipped up (and that so many of them were actually functional) was great.”

Next steps involve continued collaboration with mentors and experts to further test and refine the prototypes. More to come!