Dr. Piper Vasicek writes about how AI could be used to generate the footnotes that readers actually want.
We live in a world full of language-producing technology that would have been a great premise for science fiction ten years ago. This technology has the power to enable learning to a degree never seen before with a personal tutor essentially available to anyone with access to the internet, but with the increasing obvious potential to do our thinking for us. So how is it actually useful and desirable to employ this technology in a way that can enhance understanding, learning, and the joy of experiencing a good story and avoid the possible negative side-effects that employing it badly might cause? In their paper “Stepping into the Margins: How Readers Want AI to Generate Footnotes,” Piper Vasicek, Courtni Byun, and Kevin Seppi discuss which footnotes readers want and how AI could be utilized to create them.
For the purposes of the paper it was written that “we do not distinguish between footnotes, endnotes, and intra-textual commentary. Our goal is to understand how to best support reader understanding. Therefore, we use the term footnotes to include any notes, commentary, citations, etc. that are not part of the main body of the text, but that are meant to elucidate the main text in some way.” This category therefore encompasses everything from a link to related scriptures, a humorous aside at the bottom of a memoir’s page, a rambling three-page footnote about nuances not discussed in the main text, and much more.
The researchers conducted interviews with thirteen readers in order to ask questions about what kind of footnotes they currently use and what they would like to see from an AI footnote. They wrote that “our participants tended to be highly educated. All of them have completed at least some college, and 4 of them have doctoral degrees. The specific population we interviewed also tended to be more religious–most of our participants use footnotes for personal religious study in addition to using footnotes in their career/main field of study, and many are also hobby readers.”
Though responses from the sample varied, the researchers put each response into four categories of footnote in order of “how much external data would be required to produce them.” The categories are as follows:
- Requires only the text being read
- Requires one or more outside source
- Requires a database of related texts
- Requires access to social media, the web, and other “real world” sources
For each category, there were different suggestions for what footnotes could do. For the first category, AI footnotes could produce a summary of what has just been read, produce a character biography, or rephrase a difficult sentence. For the second category, AI could connect to dictionaries or foreign language books to give a better context for a word or let one understand how central a concept is to field. For the third category, AI could include both a citation as well as relevant text from the source so it can be easily evaluated or give recommendations for similar books based on writing style or certain scenes. For the fourth category, AI could be utilized to inform when a piece of writing was connected to a real world event or share the current discussion surrounding a piece of writing. Another category was multimedia footnotes, where participants discussed footnotes linking to data visualization, videos, images, or fan art.
Furthermore, some participants discussed how they would want to implement AI footnotes into their scripture study. Ideas ranged from having the etymology and history of religious words as footnotes to summaries of relevant information of historical figures in the Doctrine and Covenants.
Additionally, the participants were asked to discuss “what ideal AI footnotes would look like.” Participants wanted credible sources, good filtration, appropriate length of footnotes, and a lack of intrusion into one’s personal study.
The authors then went on to discuss which types of AI systems would work best for each category of footnote and which databases would be required. Ultimately, they laid an incredible foundation for those interested in creating AI driven footnotes.
In “Faith, Moral Compass, and the Gift of Possibility in an Age of Artificial Intelligence” by Elder Gerrit W. Gong of the Quorum of the Twelve Apostles, he says “This gift of possibility is realized as we ground AI personas and models with integrity of moral purpose and practice.” Dr. Vasicek through her research with Large Language Models has exemplified this principle as she works to discover exactly what people need from AI and how those solutions could be realized.
Read the full article here.