AI Found You a Research Gap. It's Fake. Here's the 3-Minute Test | FastTrack LIVE #69

Main Ideas

• AI-generated research gaps can sound convincing without being real or useful. Common suggestions include “X and Y have each been studied, but nobody has examined them together” or “a framework is needed to bring them together.” These can seem novel while amounting to little more than a search result or a superficial combination.
• A gap is not valuable simply because a study has not been done. An unexplored topic may be an “empty room”: nobody has studied it because it is not important, feasible, or likely to attract interest.
• Novelty does not guarantee a workable project. The presenter describes research in which AI-generated ideas were rated slightly more original than human-generated ideas, but less viable. In a follow-up implementation exercise, experts spent more than 100 hours testing ideas; the human-generated ideas reportedly produced better results and took less time. One proposed explanation is that AI often overlooks practical costs, risks, skills, resources, and career constraints.
• AI tends to produce a narrow range of idea types. It commonly combines two topics, proposes a framework that links them, applies a method or question to a new population or setting, or suggests moderators. These approaches can be productive, but they are not automatically genuine research contributions.
• AI is less dependable near the frontier of knowledge. The presenter recounts studies suggesting that AI performs better with well-known, heavily cited work and makes more errors with less-cited or less-studied topics. Errors may include invented references or, more subtly, misrepresenting real sources to fit a narrative. Researchers should verify claims against the literature rather than relying on a chatbot’s memory.
• AI is more useful as a critic than as a source of finished research ideas. It can help challenge an idea, find similar studies, or identify weaknesses. Researchers should still supply the substantive puzzle, evidence, and judgment.

A Model of Creativity

The presenter distinguishes three ways ideas can develop:

1. Interpolation: Filling in a gap within an existing pattern. AI is often good at this—for example, proposing a bridge between two established topics.
2. Extrapolation: Extending an observed pattern or projecting it forward or backward. AI is also suited to this kind of pattern-based work.
3. Puzzle-driven creativity: Starting with an important unresolved problem, then bringing in an idea or approach from elsewhere to address it. The presenter associates this with:
   • Abduction: Borrowing an idea from another context to explain or solve a problem.
   • Bisociation: Bringing together two previously separate domains.

The Gutenberg story illustrates the last approach. The presenter describes Gutenberg combining the reusable metal letter-making technique associated with coin stamps and the pressing mechanism of a wine press to help solve the problem of copying books. The point is not merely to combine two things, but to use the combination to address a meaningful problem.

Three Tests for an AI-Suggested Research Gap

1. Duplication Test

• Treat novelty as a hypothesis to test, not a claim to accept.
• Search Google Scholar or use a literature-search tool to find the closest existing studies.
• Ask AI to identify the most similar papers, then actively try to disprove the claim that the work is new.
• Check the articles themselves—especially their full texts where possible—rather than relying only on abstracts or an AI’s recollection.
• Follow later citations to see how the debate has developed. An unresolved question from five or six years ago may already have been answered.
• Look for conflicting results or an active debate, which may reveal a more defensible puzzle than a generic claim that “no one has studied this.”

2. “So What?” Test

• Explain why anyone should care about the topic.
• State what would meaningfully change if the research succeeded.
• Link the gap to a clear research question and a specific contribution.
• Do not treat the absence of research as evidence of importance.

3. Feasibility and Realism Test

• Ask whether the project can be completed with the researcher’s available time, skills, data, methods, and resources.
• Check whether key concepts can be measured convincingly and whether the proposed causal claims can be supported.
• Consider the practical and career costs of the work, including the risk that an ambitious project may fail.
• The presenter recommends favoring manageable, high-value projects—particularly for early-career researchers—rather than adding complexity for marginal gains.

A useful final check is to describe the gap in one clear sentence, say what would change if the study did not exist, and ensure that the gap, research question, proposed contribution, and methods align.

Examples Reviewed During the Livestream

• Servo-system control: An AI-suggested gap described a trade-off between more sophisticated control approaches and the complexity or training time they require. The presenter saw this as a possible problem, but not yet a clear research gap: it did not specify a feasible research question or how the researcher could overcome the trade-off. A project requiring a major technical breakthrough could be especially risky.
• Fintech lending and financial stability: A submitted idea proposed testing whether institutional quality moderates the relationship between fintech lending and financial stability, and whether that moderation differs across country groups. The presenter questioned the hypothesis’s complexity, the availability of enough country-level data, and how the concepts would be measured. A live duplication search surfaced four closely related articles, including work addressing institutional quality and related forms of heterogeneity. The topic appeared to have an active debate, but the researcher would need to show what remains unresolved and what the new study adds.
• Motherhood, identity theories, and Southeast Asia: A submitted proposal combined self-concept clarity (SCC), narrative identity theory, and a claimed lack of Southeast Asian participants in motherhood studies. The presenter said the rationale for combining the theories and selecting that population was unclear. A stronger proposal would explain what the theories predict, why Southeast Asia provides a meaningful test, what the study aims to demonstrate, and why the result matters.

Other Cautions

• The presenter warns researchers to disclose AI use according to relevant requirements, arguing that undeclared use can create professional vulnerability. Dartmouth and a Harvard president are mentioned as examples in a broader discussion of academic accountability; the transcript does not provide full details.
• Searching only one database, such as Scopus, limits the evidence base. A gap search should use multiple sources and be checked critically.
• Citation patterns can offer a rough indication of whether a topic is active, but they do not by themselves establish that a project is important or publishable.

Speakers and Sources Featured

• Main speaker: Professor David Stuckler, identified in the video context; the auto-generated subtitles render his surname inconsistently as “Duckler” and “Stachler.”
• Audience contributors: Viewers supplied research ideas and comments through the livestream chat, including submissions attributed to Bukhari, Saad, and Iman, as well as comments from Todd, Ahmad Ali, Muhammad Jawara, Kevin, Minno, and Ratan.
• Research sources discussed by the presenter:
  • A study by “C and colleagues” (the author name is unclear in the subtitles) comparing expert- and AI-generated research ideas, including a later implementation exercise.
  • A preprint by Chen and colleagues on the kinds of ideas generated by AI versus people.
  • An older study comparing AI’s handling of frequently cited and less-cited papers; the authors are not clearly identified in the subtitles.
  • A literature-review study by Leonard and colleagues examining AI output on common and less-studied mental-health topics.
  • Arthur Koestler, whose concept of bisociation is invoked in the discussion of creativity.
  • The historical account of Johannes Gutenberg, used as an illustration of combining ideas and mechanisms from separate domains.