VOICES OF THE SEARCHERS
Will We Be Replaced by Chatbots?
by Mary Ellen Bates
OpenAI, the company responsible for DALL-E 2 AI-generated images (openai.com/dall-e-2), recently rolled out its newest AI creation—ChatGPT (chat.openai.com). As with any AI tool, it’s creepy, eerily human-like, and the cause of much consternation and gnashing of teeth among journalists and educators.
While it is billed as a conversational chatbot, ChatGPT is better understood as a pseudo-sentient writer. As with other AI initiatives, ChatGPT functions best when relying on commonly known facts and narratives it has “learned” from books, Wikipedia articles, and other open source content. Ask it to write a 600-word essay on ot ters, and it spits out a reasonably good paper for a high school student’s homework assignment—which has teachers concerned about the future of essay assignments. If you are suffering from writer’s block, ChatGPT can write a workable first draft of a letter asking for permission to quote an expert, beginning, “I am writing to request permission to quote you in an article I am writing for [publication]. The article is focused on [topic] and I believe that your thoughts and insights on this topic would be of great value to our readers.” Its potential impact is in the ability to generate responses that sound reasonable and convincing; are written in clear, well-punctuated English; and generally pass the straight-face test.
It has some interesting quirks. When I asked it, “What will be the impact of ChatGPT on librarians?” I got the answer, “It is not clear to me what ChatGPT is or how it would potentially impact librarians.” When I then asked, “What will be the impact of AI systems like you on librarians?” I got a surprisingly good answer (well, given the source), including the following reassurance: “AI systems should be seen as tools to assist librarians, rather than replacing them. Librar ians have a wide range of skills and expertise that go beyond what AI systems can currently do, including the ability to interpret and evaluate information, provide personalized recommendations and assistance, and facilitate research and learning.” Of course!
One of the limitations of most AI systems is that they are not consulting a live version of the web to answer questions; rather, they are fed a dataset of—presumably—accurate information. The AI algorithm isn’t able to ask why the data shows a certain trend, or whether there were biases implicit in reported research. It lacks curiosity, one of the key attributes of a good searcher.
I was recently working on a project that looked at manufacturing trends in the Rochester, N.Y., area, and after I had sent in my deliver ables, I asked ChatGPT to “write about manufacturing trends in Rochester, NY” to see what kind of answer I would get. The response was a short essay that sounded reasonable, although there were no citations—simply assertions such as, “One trend is the increasing use of automation and advanced manufacturing technologies.” Without any access to the underlying information on which this answer was based, I cannot imagine a business executive making decisions based on ChatGPT’s answer. In fact, it was frustrating to this info pro to get an unsourced answer that sounded plausible but also could have been based on outdated or biased information.
Interestingly, LexisNexis is reported to be exploring the use of ChatGPT for generating answers to legal questions, although early tests suggest that it is not yet a reliable decision-making tool, sometimes giving incorrect or overly wordy answers. This got me thinking about how, as librarians and information professionals, we can prepare ourselves for the expansion of conversational AI such as ChatGPT into the workplace. This technology is most useful when a question for which there is a generally accepted answer is asked—providing a summary of how solar flares affect GPS sys tems, say, or a discussion of the pros and cons of electric vehicles. I expect something like ChatGPT could take over part of Google’s market for ”tell me how to” queries; if I want a simple explanation of how to calculate CAGR (compound annual growth rate), ChatGPT offers a clearer answer than a search engine.
Info pros have addressed the “It’s all on the web for free” trope by focusing on the added value we provide in distillation and analysis of information. Likewise, we need to focus on our evaluation and discernment skills to differentiate ourselves from an algorithm that sounds smart but is incapable of understanding the info-scape as information professionals do. As ChatGPT advised, “Given the potential for automation to replace human labor, it is important for librarians to stay up to date on the latest developments in AI and to find ways to use AI technologies in a way that complements and enhances their work, rather than replacing it.” I couldn’t have said it better myself.