New AI literacy interest group holds teach-in sessions on nuance of large language models

Penn's new literacy interest group on artificial intelligence hosted an inaugural panel and showcase to spread awareness about interdisciplinary topics related to large language models. 

The Penn AI Literacy Interest Group was launched this semester to create a nuanced understanding of recent AI developments. The day-long event took place in the Van Pelt-Dietrich Library on Feb. 13 and included presentations from expert speakers, panel discussions, a special collections showcase, and interactive games.

Applied data science librarian Jajwalya Karajgikar founded the literacy group with the aim of exploring the multidisciplinary impacts of AI in a way accessible to the Penn community. Karajgikar, who is part of the Penn Libraries’ Research Data and Digital Scholarship team, said that she wanted to avoid falling into either overly optimistic or pessimistic views about new technologies.

“The group's mission is to be a critical teach-in space for AI and the societal implications of the ubiquitous usage of AI," she said. "The idea is to not succumb to doomerism." 

Karajgikar assembled an advisory board composed of students across different schools to organize the group's first sessions. The speakers on Tuesday covered a range of topics in AI from technical, historical, and legal approaches.

The day began with an introduction from Karajgikar and College junior Eug Xu, who is the RDDS data science and society research assistant and member of the literacy group.   

Andy Janco, research software engineer and digital scholarship programmer at Penn Libraries, then provided an overview of what large language models are and how they work. He described how models generate approximations of cumulative data online without social context. 

“ChatGPT is a blurry JPEG of the web,” he said.

Penn Libraries copyright advisor Stephen Wolfson explained the copyright implications of generative AI, while a subsequent session led by historian of sciences David Dunning focused on the history of AI in the classroom. 

Karajgikar then hosted a showcase at the Kislak Center for Special Collections, displaying objects such as books on ciphers and science fiction novels from the 1970s. Event attendees ended the day with a social hour, engaging in games and activities related to digital humanities. 

Xu emphasized the importance of having a discussion about AI, given its rising prominence in day-to-day life. 

"It's really important for users of this kind of new tool to understand what that new tool is, first and foremost, and also how to use it in a way that will yield them results that they want," they said.  

Xu added that other crucial topics to discuss are the ethical considerations of AI and the material impact of computing. 

College sophomore and literacy group member Mia Antonacci said that she is interested in looking at AI through a political lens. She stressed the importance of AI ethics across all industries due to the inherent bias existing in machine learning and large language model training data.

"At the pace that we're moving just in the last couple of months with AI, it's very interesting to see the relationship with how much bias we still haven't been able to control for," Antonacci said. "It's crazy that we're advancing so fast, and it's still such an issue."

Karajgikar said that she hoped that attendees would leave the event with a new sense of understanding and curiosity about AI tools. 

“[The literacy group] is a space for … people to consider these systems in place critically and apply that to their daily life,” Karajgikar said. “That's what media literacy is. That's what tech literacy is, and I hope in a library setting, AI literacy can also evolve in that way."

The group is co-sponsored by the Price Lab for Digital Humanities, the Data Driven Discovery Initiative, and the Andrea Mitchell Center for the Study of Democracy. It plans to host a student-led debate discussing the intersectionality of technology, gender, and region in relation to AI on Feb. 27.