Is the integration of large language learning model AI into search engines going to herald as massive a change for knowledge organization and discovery as Mosaic did in 1993? I spent some time with the LLM-powered chat bot ChatGPT to try to get a sense of what we are facing, at the end of which I am not much wiser though I am persuaded it can generate a decent haiku.
I gained a ChatGPT account in return for an email address and phone number. I started asking questions about work-related tasks. It gave me interview questions that were bland arrangements of buzzwords but not that different from what an uninspired search committee might come up with on a dull Monday morning. As I watched the sentences rolling out, I had a strong urge to take off for lunch instead of reflecting on our library’s strengths and needs and candidates’ resumes and how to best elicit a meaningful conversation during an interview. A tempting dystopian next step could be to let ChatGPT take care of the entire recruitment process - compose our advertisement copy, scan resumes, conduct interviews and suggest a lead candidate for us. A cynic might suspect that much of that may already be happening in corporate searches.
Using ChatGPT feels exhilarating, despite the dullness of the text it generated in response to my prompts. I asked about the future of academic libraries, and it dutifully composed entirely predictable bullet points. As for the future of search, it wrote that it would involve personalized results and 24/7 availability; has ChatGPT not yet been told about Google? I got the most joy from demanding particular poetic forms; the sonnets were painful, but the haikus were satisfying
On verifying citations:
Citations verified, Sources checked with utmost care, Accurate research.
On information literacy:
To navigate life, We must learn to seek, find, and Judge information.
On academic libraries:
A haven of books, Silent halls and endless tomes, Knowledge waits within.
It broke down when I asked for a haiku about busy libraries serving Hispanic students.
Search engines are introducing language learning model-powered search – DuckDuckGo has DuckAssist that harvests and delivers content from a relatively small collection of texts, including Wikipedia. Google has released Bard. Bing has, well, new Bing. We have been teaching students to include in their critical assessment of web-sources consideration of elements of the website itself and what it reveals about the author, currency, etc. Other methods are needed to assess information extracted from unknown websites; CRAAP and SIFT evaluation methods are not enough. How can we assess information when the source is hidden? When we cannot know who created the information nor deduce for what purpose?
If a searcher’s satisficing threshold is low, there will be no incentive to go further than the AI-informed search engine result; that’s fine when we want to know how long it takes to soft-boil an egg but poses problems for higher stakes topics.
This is where the traditional gatekeeping function of libraries can help once again. Libraries have tools to aid the discovery of and provide access to high-quality, original, authoritative, and evidence-based content. Libraries will find uses for ChatGPT: its ability to construct coherent texts can help compose rough first drafts of technical reports, executive summaries and abstracts, resumes, and cover letters focusing on specific job descriptions. Its chatbot conversational abilities could be harnessed to, e.g., replace a frequently asked questions page. But literary & scientific works remain human creations, and our libraries will continue to collect and share those works.
NYC Board of Education has banned ChatGPT. Some colleges have incorporated ChatGPT prohibitions into their plagiarism policies. Some instructors are experimenting with using ChatGPT in the classroom to enhance learning.
We need to figure out when composing text using ChatGPT may be helpful and appropriate and guide our students appropriately. Of course, they should know not to cheat themselves out of an education by plagiarizing. We should teach the importance of evidence-based policies & practices and encourage the study of tools of persuasion in contemporary and historical contexts. The ability of ChatGPT to generate search engine optimized (SEO) blocks of text makes it a powerful tool for generating propaganda and marketing content; information literacy skills will be vital. Even commercial search engines will continue to be challenged to prioritize meaningful results as chatbot-generated sites heavy in SEO keywords and light on the content rise to the top of their lists.
I turned to Chat GPT for advice on the future of libraries. It told me, “academic libraries provide a range of resources, services, and support that are not available elsewhere, and are essential to the research and learning process. While the internet has certainly changed the way we access and consume information, it has not made libraries obsolete.” I don’t disagree, but it misses the insights that poets bring.
New eyes each year... by Phillip Larkin. (1979).
New eyes each year
Find old books here,
And new books, too,
Old eyes renew;
So youth and age
Like ink and page
In this house join,
Minting new coin.