Using AI Tools in Your Research

Some AI systems have been developed with the specific purpose of performing tasks or services for users based on prompts or commands given via text or voice. These tools are now standard on many smartphones and other devices, with examples including Apple's Siri, Microsoft's Cortana, Google Assistant, and Amazon's Alexa. 

Recent advances in AI have built upon the concept of virtual assistants, bringing forward a wave of IVA tools meant to assist users with research processes. 

See the examples below for more information on these tools and what they have to offer. 

Elicit

Elicit is an AI research assistant that uses large language models to find research and summarize research based off of text input. 

Uses

• Finding research articles for literature reviews and papers.
• Extracting key information from articles such as measured outcomes, number of participants, or other data points.
• Summarizing abstracts.
• Brainstorming research questions. 

Limitations

• Like Google Scholar, will link to articles that are behind paywalls - always remember to check for these articles in CardinalSearch and request PDFs via Interlibrary Loan as needed.
• Limited to articles in the Semantic Scholar Academic Graph dataset, which does not include all published articles.
• As with all other AI tools built on large language models, Elicit may "hallucinate" information or create inaccurate summaries. 

Consensus

Consensus is an AI research assistant that searches for research articles based off of users' questions and attempts to provide a summarized response based on the most relevant papers. This tool emphasizes searching for yes or no questions. 

Uses

• Finding research articles for literature reviews and papers.
• Getting a quick idea of the current research in a particular area of interest. 
• Visualizing consensus about a topic among a small number of papers. 

Limitations

• Like Google Scholar, will link to articles that are behind paywalls - always remember to check for these articles in CardinalSearch and request PDFs via Interlibrary Loan as needed.
• Consensus is not transparent about what set of articles they are searching and, therefore, what content they might be missing. 
• The Consensus Meter visualizes data from a very limited number of articles and may not be able to correctly classify results, causing incorrect information to be shared with users.