Published on 09 March 2023, by Hannah Wood
Service delivery, Use of technology, Resource Discovery
About the use of ChatGPT in searching
The evolving world of artificially intelligent tools
The emergence of ChatGPT poses new challenges and opportunities for the Library and Information Science (LIS) sphere. A not-so-new technology that’s now freely available for anyone to use and try out for themselves. It’s an indication of changing times.
There has been a lot of media hype in the past couple of months, paired with competing companies launching their own chatbots, with mixed results. Half-baked products which neither performed well nor were particularly accurate have been released and hastily retracted. Microsoft’s Bing was pulled for providing inappropriate responses to users, and Google’s Bard has been announced, with a mixed reception. Whether Bard will join Google’s list of failed products is yet to be determined.
It’s been a bit of a wild west in the world of artificially intelligent products!
Journal publishers and researchers have taken note of the usefulness of ChatGPT. Some authors have, perhaps unwisely, credited ChatGPT with authorship.
Questions around accountability have rightfully arisen, and some publishers have now banned ChatGPT as being listed as an author. Some have also taken steps to manage the use of ChatGPT and similar softwares in research.
As more of these innovative products hit the market, we may see further questions arise concerning authorship, copyright, accountability, and research transparency when it comes to AI tools.
As more people use them, demand will grow. This means that more tools will be marketed, and more people may be asking us about what these tools are, and how they can be used.
Much like the advanced search databases we’re accustomed to, the tools are only as good as the knowledge used to utilise them, as well as the quality of information available.
Without getting too philosophical, I see these developing technologies as a reflection of human ingenuity and curiosity. ChatGPT has processed an untold amount of human-generated information, and it is merely reflecting that information back to us in a more user-friendly format.
What is ChatGPT?
ChatGPT is a narrow artifical intelligence (AI), a generative pre-trained transformer technology, and a sort of natural language processing software. A lot of technojargon! Here are a few definitions:
Narrow AI tools which complete singular tasks without human assistance, and which doesn’t necessarily impact other tasks. In the case of ChatGPT, it means that while it can recall previous information in a chat series, it can’t recall or use information found within other chats to influence its responses.
For example, if I opened a new chat sequence about spaceships, ChatGPT cannot recall or use that information in future new interactions. It is purely language and text-based too, which also makes it a narrow AI.
Generative Pre-trained Transformer (GPT)
GPT is a fancy way of saying that ChatGPT has been given a vast wealth of different sources of information to process and draw from, but it does not have a live connection to the internet.
This means that it ‘learns’ in a supervised way, somewhat limiting the risk of inappropriate responses. It also means that it must be regularly updated with new information, and its reliability to answer questions concerning topics post-2021 is a little questionable.
Natural language processing (NLP)
NLP tools which can read and ‘understand’ human context in information.
As mentioned before, because ChatGPT and other tools like it use vast quantities of human information (which may be prone to error), it can inevitably make mistakes. Sometimes it can randomly make things up and present them as facts.
This is known as hallucination. Being able to spot this can be tricky, which is why it’s always going to be important to double-check sources and information! The need for critical thinking and searching skills will always be required.
Statistics can be another challenging area for ChatGPT. For example, asking ChatGPT how many chickens there are in the world poses unusual problems despite it being a seemingly straight-forward question. How are chickens counted? What types of chickens are included in the data? How do different farms and chicken organisations collect, present and store data about their chickens, and has it been made available to the chatbot? There are many variables, and with variables, comes risk of error.
Finally, we don’t really know much about what sources of information ChatGPT has access to.
How it can be used in searching
Asking it straightforward questions can be a useful starting point for searching. It is also very useful for listing notable journals, which cuts a little bit of time. Creating lists of synonyms for words also helps me broaden searches when required.
Perhaps most usefully, it can organise information with headings and subheadings. I’ve used it to help me organise information on search topics into easy chunks. This has also been helpful in providing areas for searching that I might not have originally considered. While this adds a little extra time to my searches, it does mean that the quality of my searches has generally improved.
It can also be great for listing resources to search. Databases, websites and official organisations which I might not have necessarily known about before are now more accessible to me.
While it doesn’t have access to medical subject headings, it is fairly useful at building very basic search strategies. I don’t really use ChatGPT for this however, and advise using with caution!
Other ways to use ChatGPT
Being relatively new to the world of Knowledge Management, I’ve found ChatGPT useful in providing definitions and advice about using certain KM tools. I have used it to draw up a plan for a hypothetical peer-assist session; this could be used for other KM tools too, which helps me find new opportunities to highlight the tools to my users.
It’s also handy for generating fancy acronyms; when asked to come up with a list of potential acronyms for a librarian and KM group, it came up with some funny and creative names!
I’ve also used it to set up an imaginary library service. It came up with a targeted library strategy, a business case for extra funding, and highlighted desirable characteristics for a good performing public library service. It also allowed me to practice interview questions. This has implications for quick interview question generation, as well as potential for being used to assist with job descriptions.
Assistance with coding is available on ChatGPT (handy for basic web design and troubleshooting), but I offer words of caution since it’s not always certain where it draws its information from.
ChatGPT has been interesting to use, and it has some helpful functions. Like every tool and technology however, it’s not perfect. I’d recommend trying it out and having a play, its versatility means that it can be used in many different ways, and it’s certainly worth exploring.