Introduction to Lex: An AI-Powered Writing Assistant

Submitter: Marc Watkins, U of Mississippi

——————————————————

The experiment:

Chatbots make poor interfaces for writers—they lack affordances, responses are isolated, and require users to engage in an iterative exchange with a machine, hoping that it guesses the their intent based on a prompt (Wattenberger 2023). However, ChatGPT is not the only interface a user can engage a large language model. Four months before OpenAI released ChatGPT, I helped start a working group in the Department of Writing and Rhetoric at the University of Mississippi to pilot AI assistance in first-year writing courses. Our group quickly discovered that chatbot interfaces did not offer students abilities they were familiar with in traditional word processor software. To use generative AI effectively, the tool’s interface matters just as much as the underlying LLM. We piloted interfaces by Fermat, and Wordtune, before finding our current tool, Lex. Each tool changed multiple times during our pilot and it quickly became clear that learning to teach students resiliency and adaptability about rapid updates to generative AI was a crucial part of AI literacy.

Lex can use a variety of LLMs, identify what material the author generated, provide in-line edits/ suggestions, and even generate holistic feedback for an entire essay. After introducing Lex, I settled on a gradual release model of instruction to allow students to explore how to use the tool independently.

Results:

Many students’ only previous experience with generative AI was through ChatGPT, and employing the technology outside of a chatbot interface in a more familiar UI gave students more freedom and flexibility about when to use generative features to support their writing. Perhaps one of the best features of Lex is students did not have to use generative AI to produce any content within their essays if they choose not to. Many instead used the underlying features of the tool to brainstorm and get feedback on their overall writing.

Providing students with an ethical framework allowed them to explore how these tools best supported their overall learning by letting them implement the technology alongside their writing and evaluate it through structured reflection. Likewise, many students valued being taught how to use generative AI technology pragmatically and ethically.

Writing is itself a technology, but also a distinct practice. Asking users to abandon the latter forces a change in their behavior. Put simply, I’ve not seen students abandon their practice of writing to embrace text generation via chatbot as envisioned by many. Writing is a habit and our students arrive in higher education with over a decade of consistent writing practice. Allowing them to explore this new technology through more humanistic interfaces offers users a thoughtful bridge between established and emerging writing tools. I hope to see more humanistic interfaces that support writing, not offload it.

Relevant resources: https://docs.google.com/document/d/1ggP_lQ5jeZSWTL2OP1rdgngt5sgbCC81/edit?usp=sharing&ouid=106137383630710974014&rtpof=true&sd=true

Contact: mwatkins[AT]olemiss[DOT]edu