Large language models (LLMs) are transforming the ways in which we conduct academic research. These AI models streamline research tasks, from drafting ethics applications and grant proposals to collecting and using large data sets to summarising literature and editing manuscripts. This speeds up the research process, potentially democratising access to complex analytical methods.
The rapid proliferation of both commercial and open-source LLMs has afforded researchers enhanced capabilities for accessing and analysing large qualitative and quantitative data sets, which in turn facilitates interdisciplinary research innovation. We are at an interesting junction in how knowledge is created and shared.
Although mainstream LLMs, such as ChatGPT, Gemini, Claude, Perpexity, Mistral and Copilot, provide general data analysis and synthesis functionalities, a new generation of AI research assistants is emerging. These build on and repurpose the frontier models to deliver task-oriented and domain-specific tools, often featuring advanced search and retrieval functionalities tailored for academic rigour. For instance, Elicit uses different versions of ChatGPT and Claude to train its system to handle different parts of the research process.
- GenAI as a research assistant: transforming start-up due diligence
- Teaching responsible use of GenAI in graduate studies
- Why AI literacy must come before policy
AI research assistant tools and AI-enabled deep search agents provide unprecedented capabilities to access, analyse and synthesise quantitative and qualitative datasets, offering new methods of cross-disciplinary and cross-sectoral research. They can, for example, conduct clustering and thematic analysis of interview and focus group data, analyse survey data and run statistical tests, or visualise textual or numerical data.
AI research agents and what to use them for
Conducting literature searches, reviews, analyses and syntheses are fundamental elements of academic and scholarly work. So, here are the novel and useful capabilities of three AI research agents that I have used.
Synthesise literature and find consensus
The first tool, Consensus, is an AI search engine for research that provides a quick synthesis of scientific literature and identifies the common viewpoint, through its “consensus meter”. It shows how it conducts literature search and analysis and applies inclusion/exclusion criteria. Consensus includes a feature that identifies research gaps, evidence strength and a timeline of key research and development in a particular area. In ranking retrieved studies, it uses research-quality indicators such as citation counts, method used (case report, systematic review, meta-analysis or randomised controlled trials) and journal reputation measures. One of the downsides of this tool is that in some cases in prioritises surface-level claims (the “what”) without explaining the “how”, such as sample sizes, dosage or study limitations.
Extract and synthesise data
If you are looking for a tool to extract and synthesise data, Elicit is a systematic review AI assistant. It allows users to gather, screen, review and automatically extract and synthesise data from up to 500 academic papers. You can create customised reports of study design, data-gathering methods, population or materials/experimental methods across all those studies. Similar to Consensus, Elicit provides a transparent description of search, screening, inclusion/exclusion and extraction. Additionally, you can upload your own papers, which is particularly useful if you’ve already conducted some initial literature review and want to include relevant papers you’ve identified in the systematic review and analysis.
AI-supported literature reviews
Undermind is an AI-assisted deep search agent that mimics human search. It uses iterative, adaptive semantic, keyword and citation searching to support literature search and review. It takes about eight to 10 minutes to conduct a deep search to identify high-quality, relevant papers. Its comprehensive reports function focuses on thoroughness of complex queries and estimates search exhaustiveness. The “discovery curve” identifies relevant papers, even if they are new, interdisciplinary or related to niche areas of research. It also allows you to assess how extensively a research idea has been explored.
My Undermind search for the topic “ethical risks of generative AI in Indigenous data usage” provided a report with eight categories or facets of research in this area that can be explored, ranging from Indigenous data sovereignty to governance frameworks, as well as positive applications of GenAI for Indigenous empowerment.
In addition to these topics, it suggested clusters of research groups and their contributions to this area, a feature that facilitates partnerships and collaboration.
Ensure your AI assistant is credible
A key criterion in determining the quality and credibility of AI research assistants is transparency about the tools’ use of underlying data and the digital libraries and repositories that they use for both training purposes and for supporting the user in searching for and retrieving relevant high-quality research publications.
Elicit, Consensus and Undermind all use Semantic Scholar, a database of more than 200 million scholarly publications, developed by the Allen Institute for Artificial Intelligence (Ai2) in 2015. Both Elicit and Consensus also use PubMed and OpenAlex, two popular open-access digital repositories of scholarly and open-access research publications.
The movement towards incorporating AI agents in academic search engines has prompted the well-established Google Scholar, with 21 years of history, to develop an AI-enhanced version that provides AI-assisted summaries. It is available as Google Scholar Labs to a limited number of users. While it is a useful tool for the quick overview of article summaries, it is an experimental platform that does not have advanced customisation functions and is unclear about data sources and methods.
Researchers need ongoing training in AI literacy
Given the expanding capabilities of AI research assistants, it is more urgent to design AI literacy training opportunities specifically for academic researchers and graduate students. Such training should address the possibilities, constraints and ethical considerations involved in the use of these tools across research and scholarly practice. To use AI research assistants ethically and effectively, researchers need training in information and data literacy as well as digital literacy skills. These skills – such as assessing the credibility and provenance of sources, identifying and evaluating research data, and verifying citations – will become increasingly vital as this technology advances further into research practice.
Ali Shiri is professor of information science in the Faculty of Education and vice-dean of the Faculty of Graduate and Postdoctoral Studies at the University of Alberta, Canada.
If you would like advice and insight from academics and university staff delivered direct to your inbox each week, sign up for the Campus newsletter.
comment