You’re sitting down for the big task of pulling together a literature review; looking over the mountains of PDFs you’ve downloaded, you’ve got dozens of open tabs in all the research databases you searched, and you have that sinking feeling in your stomach because you’re pretty sure you lost that one research article that would’ve changed everything. You’re not alone – this is a common experience all researchers share as they’ve come to grips with the realities of researching academically. But, what if you had a research assistant that never got tired, could read thousands of abstracts in seconds and knew exactly what you needed? Welcome to the world of
AI to find research papers with AI assistance. This isn’t a future fantasy; this is the new reality of academic research today. The possibilities are exciting – literature reviews will be much faster, much more accurate, and not nearly as depressing. But are these digital applications going to really provide all those benefits, or just create another layer of confusion in an already difficult process? In the next few paragraphs, I will explain how these tools work, where they’re most effective, and some very real human things to be cautious about when using these types of tools.
The Digital Librarian: How AI Scans the Scholarly Universe
The use of AI in finding papers for research is ultimately a process of finding patterns and understanding what they mean. With traditional methods of searching for information, like the Internet, you rely on searching for specific keywords that are contained within what you have typed in the search box. However, AI search tools attempt to determine the meaning of a user’s question, instead of just matching it with a search result. For example, if you were to ask what effect microplastics have on how seabirds digest food using your own words, like “How do microplastics affect the way seabirds will digest food,” an AI search engine would be able to look at this concept and locate publications regarding polyethylene ingestion, gastric plasticity, and avian toxicology even if none of those words were actually in your original question. As these AI searches are being performed, they are learning from an enormous database of academic articles and publications written in technical language (i.e., using vocabulary that describes science and/or other related subjects) (which ideas are related/connected to each other through authors writing about or referencing other authors), and which of the methodologies used within those fields are employed to write about the same subject matter. They can function as a supercharged contextual filter that searches thousands of sources – I’m talking about millions of papers from repositories like PubMed, arXiv, or Scopus – to find the most relevant example where those two words are used. The focus isn’t on creating the perfect Boolean sentence but rather on speaking with an informed (albeit artificial) colleague.
Through connections is where you will find your true magic! Research paper AI discovery tools do more than just provide you with a list; they also provide you with a map of the landscape. They can group by theme, visually display how trends have changed throughout time, and provide you with a citation backbone of your area of research. By asking AI to “find similar” or “find newer papers that build upon this” from one critical paper, you will be able to follow the evolution of an idea. Where the substantial acceleration happens is in moving from a static to a dynamic and interactive web of knowledge. Thanks to AI automating the exploration process, you can have a curated collection of literature presented in minutes rather than hours manually following the citation trail. It would be like taking a tour through a library with someone who knows every book in that library and has access to all of the hidden threads connecting them all together.
Beyond Keywords: The Superpowers of AI-Assisted Discovery
A traditional review can create frustration when there is a sense of anxiety due to the unknown unknowns; however, utilizing AI to discover research papers creates an unprecedented opportunity for serendipity by design. The ability of these tools to understand context allows them to identify potential review papers that would have never been discovered with a keyword search alone. Even if a paper does not specifically include the language “cognitive load theory” within the abstract, if the results and methodology of that research align with an understanding of the nature of cognitive load theory, then an AI could identify this relationship, thus providing the reviewer a new way to think about the topic and giving rise to possibilities for interdisciplinary approaches that may lead to discovering new innovations or ideas when writing the review.
Personalization and continuous learning are also potent features of the superpowers that AI enables. Unlike many AI-driven research platforms that use thumbs up/down or “highly relevant” tags for user feedback, these types of AI-driven systems continuously refine their ability to provide better and more personalized recommendations for your specific project and your subjective criteria for quality by continuously learning from the user input provided by each user. An even more powerful capability is the fact that many of the best research tools enable their users to seamlessly integrate reading and discovering new information. For instance, after reading a PDF document in an AI-driven environment, the AI will not only analyze the PDF’s text, but will also extract the most important concepts, and then provide the user with a list of related papers (or papers with opposing viewpoints) to explore further—creating a seamless process from finding to reading to synthesizing. Overall, the creation of a closed-loop system (e.g., eliminating the hassle of moving back and forth between a PDF reader, reference manager, and search engine) by grouping all three functions into one single, continuous experience is a major advancement in the research process by making the entire review process significantly easier, faster, and more fluid.
The Human in the Loop: Where AI Stumbles and Your Expertise Shines
Machines that search for scholarly articles are powerful but are not independent thinkers; they are highly advanced pattern-matching systems that depend on the input and training datasets to function. A significant limitation in how an AI works is the bias of the database it uses; an AI will only be able to search within the repositories and publications to which it is able to access. Therefore, if it is not connected to an appropriate niche repository, or does not find pre-prints, it may miss significant works. The age-old advice to researchers to search in multiple sources is still very true! You need to keep in mind the limits of this tool so as not to mistakenly believe that it has provided you with full coverage.
More importantly, AI does not have true critical thinking ability. AI can locate papers that have some semantic relationship to one another, but it cannot evaluate how good the method was used in the study, if the conclusions are valid, or if the study is relevant to ongoing scholarly discussions. AI will display a highly-cited but methodologically flawed study side-by-side with a well-designed, innovative study; neither the AI-created literature review nor the AI-identified papers will be able to supply the user with the data or framework needed to enable the user to evaluate the information contained in the studies. The user must perform the evaluative functions of critical appraisal, synthesis, and narrative construction in order to conduct a literature review. The AI will provide you with valuable information for your “recon” of the academic “battleground,” while you will determine where there are gaps in the information, and what strategy(s) to pursue in order to accomplish your goals. If you depend solely on the suggestions made by AI, you are likely to become trapped in an “algorithmic bubble” where the only studies you observe are the ones similar to the studies to which you’ve previously shown interest, and thus will likely miss both the disruptive studies and the contrary studies.
The Verdict: A Symbiotic Partnership for Speed
Utilizing AI for locating research papers can significantly expedite the overall literature review process — the absolute answer is yes. The asterisk comes from the fact that AI helps to tremendously enhance the discovery and collection phases of literature review. The labor-intensive, time consuming, grunt work of going through the [initially retrieved] literature and working through the citation chains and determining thematic links between articles that formerly took weeks to perform can now be accomplished in days or more typically hours. This ultimately gives the researcher one of the most important assets of all – time. That time can then be put back into the in-depth and thoughtful nature of any human being by the researcher and will be spent on reading, analysing, and writing critically.
A partnership between you and artificial intelligence (AI) will help create the ideal approach. Start your search for research articles by using AI to help guide you through the exploration process. Use it as a tool to explore the terrain and learn about the many connections between various topics that exist in your field, thus creating a rich first complete list of references to use in your review. Once you have established an initial list of references utilizing AI, use your expertise to review and evaluate this list. Remember to look for additional references from trusted sources outside of those suggestions generated by AI. This is where the real work of synthesizing information occurs. In this model, AI provides speed and scale to the process, but you provide direction and depth of inquiry. The result is not only a faster review, but potentially a better review than if you were working alone, as the tedious task of searching will no longer exist and you will have the opportunity to focus on what it means to be an artist in research. The future will see literature reviews being performed by researchers using AI as an enabler, thereby achieving levels of understanding at an accelerated pace.
