Complete AI Guide for Researchers | How to use AI ethically and responsibly | Dr. Mushtaq Bilal

Researchers used to find literature by walking through physical library catalogs—index cards in drawers, then shelves, then journals and books—an approach that demanded time and specialized training. Over roughly the last two decades, that workflow has shifted toward digital search tools like Google Scholar and PubMed. The key distinction is that these tools outsource “labor” (finding and organizing information) while leaving “thinking” with the researcher: the goal is to generate a research idea first, then use digital databases to locate supporting evidence.

That labor-versus-thinking split becomes the ethical framework for using generative AI. The first rule is to outsource only labor to AI, not cognition. AI systems can assist reasoning, but they don’t “think” independently; they can’t replace the researcher’s responsibility for the direction, judgment, and validity of the work. The second rule is to use AI for structure rather than content. Generative models produce text by prediction, and predicted text tends to be familiar rather than novel. More importantly, AI-generated content can introduce hallucinations—confidently fabricated claims or citations—and the burden of correctness still lands on the person submitting the work.

The third rule is to treat AI as a research assistant, not a research supervisor. A useful mental model is to imagine each AI app as a capable undergraduate helper: it can summarize, draft, and organize tasks, but it still requires guidance and verification. That means not copying and pasting AI output into a manuscript without reading and checking sources, because the author’s name is on the final product.

The fourth, and most urgent, rule is not to over-rely on AI and to keep common sense front and center. Several real-world examples illustrate what happens when people stop verifying. Steven Schwarz used ChatGPT to generate fake legal citations and then submitted the fabricated reports in court, leading to sanctions and a $5,000 fine. In academia, Professor Jared Mum reportedly treated ChatGPT as if it were a plagiarism detector, running student work through it and failing an entire class—despite the fact that AI detectors cannot reliably confirm whether text was AI-generated. Another scandal involved researchers publishing an AI-generated image with explicit anatomy, triggering widespread backlash and media coverage.

After laying out those guardrails, the session shifts to practical use of SciSpace (SciSpace agent) for researchers. The standout capability is “semantic search”: unlike Google Scholar or PubMed keyword matching, SciSpace can interpret the meaning of a research question and then search across academic sources. In demonstrations, it builds a research plan, retrieves and ranks relevant papers (including from Google Scholar, PubMed, and preprint archives), extracts insights from abstracts, and provides narrative answers with real citations that can be checked. Features like “deep search” add clarifying questions to tailor the literature review, while “high quality review” synthesizes top papers into structured summaries and tables.

SciSpace also supports workflows that typically consume hours: adding columns such as methods or limitations to a paper table, saving notes to an AI notebook, chatting with uploaded PDFs, explaining tables and math, and even generating podcasts for commuting. The overall message is clear: AI can accelerate literature discovery and organization, but ethical research still depends on human judgment, source verification, and original thinking.