Reading Assistants: ExplainPaper and SciSpace
Submitter: Marc Watkins, U of Mississippi
Much of the discourse around generative AI involves text and image generation, with little mention of other applications for the technology. Large language models offer users much more than simple prompt/output interactions. Two apps, ExplainPaper and SciSpace, allow users to upload a PDF and use an LLM to produce concise, targeted summaries of the reading at the user’s reading level, and the case of SciSpace, in the user’s native language. Custom-built interfaces using LLMS to help users navigate large texts present unique opportunities, potentially helping students with reading comprehension, disabilities, and those who are non-native speakers. However, such tools also pose formidable challenges—how can we establish ethical use cases for reading assistants and preserve close reading skills?
OpenAI released API access to the public, allowing developers to explore interfaces that could support augmenting human skills. Both apps allow users to select what words, sentences, or paragraphs they’d like reading assistance through an LLM, offering users the chance to engage the technology based on their own unique needs. When I piloted the reading assistants with my students, I did so through a structured reading assignment, one I was familiar with and knew to be challenging to students from past semesters. I encouraged students to engage the AI only when they reached a pain point within the reading that would normally cause them to pause or disengage from the text.
Students overwhelmingly loved using the tool to augment their reading of the complicated assigned text. Several students who self-identified as having disabilities reflected on how much help the tool gave them. Many other students reported being so intimated by certain readings from the past that they did not even bother trying to engage the text and wished they’d had the tool available to them then.
However, deeper concerns emerged once I spoke to the students in more detail. It quickly became apparent that many students viewed the tool as an opportunity to offload the close reading and analytical reasoning skills I was endeavoring to teach them. From their perspective, they now had a tool that could take any text and produce an immediate summary—personalized Cliff Notes at scale. Part of AI literacy as I teach it is examining the affordances and perils of each tool and I asked my students to reflect more deeply about the implications of using reading assistants and what it might mean for their reading skills. I also asked them to consider if the LLM changed the meaning, nuance, or style of the author’s writing and how that impacts authorial intent. Reading assistants using LLMs provide one of the most compelling use cases for education, but we should establish a new discourse about what guardrails and boundaries need to be established to balance the tool’s helpful nature while preserving reading skills.
Relevant resources: https://docs.google.com/document/d/1njWH04wJMKBEduGRi93Kr_As4UaFWOpJ/edit?usp=sharing&ouid=106137383630710974014&rtpof=true&sd=true