Generative AI for Research

The use cases for AI in research include:

• Brainstorming ideas
  Chatbots based on LLMs, such as Chat GPT, Gemini, MS Copilot, Perplexity.ai, and others, can help you brainstorm ideas for starting or continuing a project of any scale. Some chatbots, such as Elicit, are specifically trained on academic texts, and therefore may be more likely with ideas of higher scholarly value.
• Building your list of references and getting started on your literature review
  Tools based on neural networks such as Google Scholar, ResearchRabbit, and others, can help you identify relevant papers, articles, and books. Many LLM-based chatbots (see above) can summarize the contents of pdf documents with high accuracy and answer questions about them. Do not upload documents into AI tools unless they are publicly available.
• Academic writing
  While we advise against uploading your entire pieces of work into chatbots due to authorship and privacy concerns, LLMs can help you edit specific spots in your writing that you are unsure about. Specific tasks include rephrasing awkward or ambiguous sentences, checking grammar, and brainstorming culturally sensitive ways to express an idea.
• Showcasing your work
  AI tools can help with titles, summaries, slides, and bullet points

Currently, Montclair State University does not have a contract with any generative AI tools or services other than MS Copilot's Enterprise Data Protection. Therefore, all such tools should be treated as third-party applications, with only Public data allowed to go into them (Data Security Classifications at Montclair)

• Hallucinations. AI generators of text, images, and video content can emulate human language, creations, and behaviors very closely. Yet, the neural networks behind them do not have the same kind of knowledge about the world as humans have. In particular, AI tools of most kinds are prone to hallucinations: they generate content that is plausible, but not actually true or accurate. Ethical use of most AI tools will include human verification and refining.
• Biases. AI tools can inherit and amplify the biases of the real world we live in. Since content generators operate based on probabilities, they are more likely to catch content that is better represented in the public knowledge domain, and less likely to include content that is already underrepresented. When you use AI tools for research, be mindful of the perspectives and sources AI may have left out.
• Authorship concerns. There is no legal guidance as to who owns AI-generated content. To keep your ideas and work safe (or safer, anyway), do not feed your unpublished pieces into any third-party tools. Authorship issues may be particularly worrisome when authors use text generators for editing help. While the input text still belongs to the author, the output you receive could count as open source or the intellectual property of the owner company -- no matter how slight the edits.
• Privacy concerns. No private data, personal or institutional, should go into any third-party tools. See above ("What kind of data can I use with AI tools") for data privacy guidance.
• Variety of perspectives and AI etiquette. Scholars have vastly different perspectives on AI: while some embrace it, others feel uncertain. If you are part of a research team, it is important to obtain everyone's consent before integrating any additional tools into your research.

Multipurpose writing assistants/chatbots

• ChatGPT
• Perplexity
• Gemini (currently not available with montclair.edu accounts)

Literature finders

• Elicit
• Research Rabbit
• Connected Papers

PDF analyzers

• Typeset (called SciSpace since Apr 2024
• many chatbots, such as Perplexity.

*All tools on this list have a free version.