Bo Ni, Franck Dernoncourt, Hongjie Chen, Yu Wang, Nesreen K. Ahmed, Zhengzhong Tu, Tyler Derr, Ryan A. Rossi
5 min
Abstract
AI co-scientists that generate hypotheses, retrieve related work, design experiments, execute code, and draft full papers are beginning to change how research is carried out. Despite this rapid progress, state-of-the-art systems remain researcher-agnostic: given a research goal, they optimize novelty, validity, or reviewer score while ignoring the individual scientist who will use the output. This overlooks a fundamental fact about research, namely, that what counts as novel, valuable, or feasible depends on the researcher, including their prior work, methodological repertoire, and the collaborators and communities in which they are embedded. In this work, we introduce the problem of personalized auto-research, which conditions every stage of the research process on a representation of the individual researcher. We argue that personalization is not a convenience layer, but rather the fundamental property that allows an AI system to serve as a genuine co-scientist rather than a generic instrument. To address this problem, we propose a general and flexible framework that threads a graph-grounded researcher context through retrieval, hypothesis search, experimentation, writing, and review. The framework consists of three fundamental components: (i) graph-grounded researcher representations, (ii) personalization across the full research pipeline, and (iii) evaluation grounded in the individual. Notably, we highlight a one-size-fits-all failure mode where distinct researchers issuing the same goal receive essentially the same research, erasing the tacit knowledge through which novel ideas arise. Finally, we discuss fundamental open problems and challenges.
Alex: That sounds broader than the usual assistant tools.
Sam: It is. The authors say existing systems often optimize for the goal alone — they try to make the output look novel or strong in general, but they don't ask whether it's the right kind of output for this particular researcher. Here, the same goal should lead to different research paths for different users.
Alex: But how does the system decide what's actually right for a given person?
Sam: The paper proposes three checks. First, the idea should be achievable with the researcher's real resources and skills. Second, it should fit their scientific identity and community — the kind of work they're known for and the people they work with. Third, it should still bring something new. And the authors are careful to say novelty here doesn't just mean "new to the world." It means new relative to that researcher's own path.
Alex: That makes sense. Otherwise the system could just keep suggesting more of the same.
Sam: Exactly, and the paper flags that as a real failure mode. If two different researchers ask about the same goal, a system that ignores who they are may hand them nearly identical suggestions. That erases the small, personal differences that often lead to original ideas.
Alex: I want to make sure I follow this. If the system already knows the topic, why does the person matter so much?
Sam: Because the topic alone doesn't tell you what can be done next. Two researchers can ask about the same problem, but one might have exactly the right tools, collaborators, and prior work to test a certain approach — while the other simply doesn't. So the same topic can point to very different sensible next steps, depending on who's asking.
Alex: So the paper is really about making the AI act less like a search engine and more like a research partner.
Sam: That's a fair way to put it. And the authors are clear that this is different from the kind of personalization you see in, say, a music recommendation app. There, one answer gets customized. Here, what's being personalized is the entire chain of scientific decisions — the literature search, the hypothesis, the experimental plan, the writing — that eventually produces a research output.
Alex: And how would you even judge whether that works?
Sam: That's something the paper addresses carefully. They argue you shouldn't only ask whether the system can predict the next paper a researcher wrote — because that can reward imitation of history, even when a better idea was available. Instead, they want evaluation that checks whether a suggestion is feasible for that researcher, whether domain experts think it genuinely fits the profile, and whether it leads to better work over time.
Alex: So the big idea isn't just "can AI do research," but "can it do research with the right person in mind."
Sam: That's exactly it. The paper's central point is that personalization isn't a small extra feature you add on top. It's what separates a capable research tool from something that can honestly be called a co-scientist — something that doesn't just know the field, but knows who it's working with.
Alex: Thanks for walking us through it. Thanks for listening to ResearchPod.