Alternatives to Generative AI for Library Research
Introduction
The presentation explores alternatives to generative AI tools for library research, examining various AI-based research assistant tools that help locate, analyze, and synthesize academic articles. Unlike traditional generative AI like ChatGPT, these tools emphasize assisting researchers in discovering and synthesizing scholarly literature rather than creating textual outputs. The speakers aim to demonstrate three primary research assistant tools—Consensus, Research Rabbit, and Scite—highlighting their practical applications in library research and how they can empower researchers in their academic pursuits.
Definitions and Key Concepts
The presentation opens by distinguishing between two types of AI in the academic context:
- Research Assistant AI: Tools designed to assist researchers by locating, analyzing, and visualizing academic articles. They are more focused on synthesis than generation.
- Generative AI: Tools focused on creating textual content or answers, which is increasingly becoming a part of research assistant AI systems.
Rationale for AI in Libraries
Duncan and Fay explain their motivation for exploring these tools. With the rise of ChatGPT and its widespread use, particularly for generating fake or unreliable sources, librarians received many inquiries from faculty concerned about the reliability of AI-generated content. Thus, The presenters explored how AI can help researchers (faculty and students) conduct efficient library research without falling into common pitfalls.
Part 1: Defining Research Assistant AI vs. Generative AI
The presentation begins by distinguishing research assistant AI from generative AI:
- Research Assistant AI refers to AI tools designed to assist researchers in locating, synthesizing, and analyzing scholarly materials. These tools typically aim to streamline the research process, focusing on information retrieval, pattern recognition, and generating summaries of existing literature.
- Generative AI, like ChatGPT, focuses on producing new content, including written text, based on prompts provided by the user. While this technology evolves, concerns still need to be addressed over its propensity to fabricate information, such as generating nonexistent sources, misleading citations, or even biased content.
Duncan and Fay highlight the blurring of these two categories as more research assistant tools begin incorporating generative capabilities. This hybridization is becoming a key trend as developers integrate text-generation features into AI systems initially designed solely for data retrieval or analysis. This trend could significantly change how researchers conduct their work, offering new data synthesis and interpretation possibilities.
Part 2: Motivation for Presenting AI Tools in Research
The speakers explain their motivation for the presentation, stemming from the explosion of interest in ChatGPT and other generative AI platforms in 2023. This surge in interest was fueled by the increasing capabilities of these AI systems, particularly in generating human-like text. However, this also led to concerns about the reliability of AI-generated content, especially in academic research. Faculty and students were curious about AI but often needed clarification on misinformation or overly ambitious claims regarding AI's capabilities.
A significant issue was fabricated sources generated by AI models, which raised alarms among librarians striving to maintain rigorous research standards. To address these concerns, Duncan and Fay explored how AI can legitimately support researchers without sacrificing scholarly integrity. They aim to highlight tools that can assist in research workflows without generating misleading information.
Part 3: The Rapid Evolution of Research Assistant AIs
One key observation the presenters make is how rapidly the field of research assistant AI is evolving. A year prior, tools like Elicit were free and widely accessible; however, they have since become commercial products, limiting access for students and researchers.
Additionally, they noted a need for AI tools for humanities research in early 2023, a significant limitation for researchers in these fields. However, this has since changed with the emergence of platforms like JSTOR AI, explicitly catering to humanities research, demonstrating the rapid evolution and diversification of AI tools in the academic research space.
This fast-paced evolution has made it difficult for libraries to keep pace with developments. However, Duncan and Fay emphasize the importance of being adaptable and continually reassessing AI tools for their usefulness in an academic context.
Part 4: Key AI Tools for Library Research
The speakers then introduce three specific AI tools designed to support research:
The Two Research Assistant AI Tools
- Consensus AI
- Focus: Designed to answer research questions quickly by identifying scholarly claims.
- Corpus: Primarily pulls data from Semantic Scholar, an extensive academic database focused on sciences and social sciences.
- Key Feature: It uses a "consensus meter" to show whether articles collectively support or refute a given research question. This helps visualize the degree of consensus among scholarly articles.
- Strength: Helps locate articles, tags high-quality papers, and provides simple visual summaries of the research.
- Limitations: Currently focused on sciences and social sciences, needing more coverage of humanities or arts.
- Research Rabbit
- Focus: A literature-mapping tool that visualizes connections between academic papers and their networks.
- Key Features:
- It offers interactive visualizations and networked nodes that show relationships between papers.
- Builds knowledge maps by referencing papers from Semantic Scholar and PubMed.
- Users can explore an author's research output and examine connections between similar papers.
- Enables collaboration by allowing researchers to share literature maps with peers.
- Strength: Visual, interactive format allows users to see relationships between articles, making it easier to understand the broader research landscape.
- Limitations: Its learning curve can make it challenging to use and may not be intuitive for all users.
Part 5: Demonstration of Tools
Duncan and Fay provide a live demonstration of the tools, showcasing their practical applications:
- Consensus: Demonstrating how it returns a limited selection of papers, providing a consensus summary on whether "learning styles affect academic success in college." They caution that while the visual consensus meter is helpful, users should always read the actual papers for context.
- Research Rabbit: They show how to create a collection of research papers on "learning styles and college students," using the tool's mapping features to find similar work and identify research networks across time. The visualization feature highlights relationships between papers, and the interactive timeline shows how the research has evolved.
Part 6: Pros and Cons of Research Assistant AI Tools
Pros and Cons of Research Assistant AI Tools
Pros:
- Interactive Visualizations: Tools like Research Rabbit offer a highly interactive interface that helps researchers visualize connections between papers.
- Custom Summaries: AI tools can generate summaries, helping researchers quickly grasp the content and relevance of articles.
- Identification of High-Quality Sources: Features like consensus meters or innovative citations allow researchers to quickly identify highly-cited or reputable papers.
Cons:
- Learning Curve: The tools work differently than traditional library databases, confusing users who are more familiar with established methods.
- Account Requirement: Most of the AI tools
Part 7: Future of AI in Libraries: Scopus AI and JSTOR AI
The presenters speculate on the future of AI in libraries, pointing to the release of Scopus AI and JSTOR AI as indicators that large academic publishers are integrating AI into their databases. While initially skeptical, Duncan and Fay now recognize that significant players in scholarly publishing are likely to continue developing and acquiring AI tools.
They note that JSTOR's AI platform, currently in beta, is stimulating because it focuses on the humanities, a traditionally underserved area in AI-driven research tools.