VOICES OF THE SEARCHERS
Have You Chatted With Your Documents Lately?
by Mary Ellen Bates
I have been speaking frequently about the impact of AI on information professionals and, until recently, focused primarily on general-purpose generative AI tools such as ChatGPT and Claude.ai and search bots such as Google Bard and Perplexity.ai. Yes, general-purpose chatbots are fine for summarizing text and drafting business correspondence (see, for example, what you get when you ask it to write a letter from a library manager asking for a raise). And search bots do a nice job at aggregating the top search results into a cohesive summary.
However, my perspective shifted when I encountered a new set of AI tools, including PDF.ai and ChatDOC (chatdoc.com), which let you upload a document and use a ChatGPT-like interface to ask questions of the document and get accurate, insightful, context-sensitive answers. For example, you could upload a company’s annual 10-K filing with the U.S. Securities & Exchange Commission, and then ask it questions such as, “What did this company spend the most money on?” or “How much money did this this company make from its best-selling product?” If you have a lengthy product user manual or employee handbook, you can ask it a question about your specific situation or issue and get pointed to the specific place in the document that addresses the problem. You can even take a research paper and ask it for the most noteworthy aspect of the findings.
Most of these tools that enable dynamic interaction with documents are available for a free test period and then cost less than $30 a month. Of course, the first question any info pro will ask is what happens to the files that are uploaded to any of these AI tools: If you upload a confidential or internal document, is it stored on the platform or used to train the chatbot? Many of the services I have investigated claim that they do not retain the document in their cloud storage, although you may want to look at your preferred service’s specific privacy policies.
While being able to have a dynamic interaction with a document could be a real time-saver, a cataloging librarian recently pointed out a library-specific use for these tools. She described a special collection of reports in her library that had not been adequately cataloged and wondered whether she could use a generative AI tool to create cataloging records for the material. As we talked, I realized that these tools to create interactive documents could be used in even a more powerful way by skipping the step of identifying a specific document and taking a user directly into a conversation with the entire special collection at one time. Metadata and cataloging only bring the user so far, and even full-text searching does not always offer meaningful access to the right information that answers the user’s need. Being able to have a conversation with the content, asking it questions that can only be answered by analyzing and synthesizing all of the content being queried, could revolutionize how we provide information access to users.
I was reminded of when I was introduced to HyperCard back in the late 1980s. Even that early, the Mac-based database program enabled hyperlinking within a document as well as with other documents on the same computer. Back then—when the only way to navigate internet content was with simple tools like Gopher, FTP, and Telnet—I had trouble getting my head wrapped around the idea of how these hyperlinks would transform information discovery as the early adopters promised. It feels like we are at a similar inflection point with interactive documents. We have access to new tools that let us discover insights and mine data in innovative and much more efficient ways. We are only just starting to see the use cases for this technology.
One of the best ways to explore the possibilities of interactive documents in your setting may be to establish a generative AI sandbox for your users and see what they come up with (and to get buy-in for funding a robust enterprise tool within your organization). Identify some sample reports or other documents that would be relevant to your users, and subscribe to a couple of the tools for chatting with those documents. Ask your users what their biggest obstacles are in finding and organizing information, and encourage them to think expansively about where these new tools could be used to solve their information challenges. Be patient; just as it took me a long time to grok the value of hyperlinks when I first saw them, it is going to take searchers, researchers, and analysts time to discover how they can best incorporate intelligent documents into their information workflow. As info pros, we can lead this conversation and advocate for tools that can transform the information landscape.