AI’s Role in Grant Applications: Is the System in Need of Repair?

The retained GPTOnline cover is an editorial design with a circuit-style face. It accompanies this historical discussion and is not evidence of a grant-writing test or funding outcome.

Historical commentary: This October 2023 discussion draws on robotics researcher Juan Manuel Parrilla’s Nature career column, published October 13, 2023. The panel and grant-writing experiences below are his account, not firsthand testing by GPTOnline.

The ability of artificial intelligence to shoulder a significant portion of the workload in grant applications is indicative of a system that seems increasingly outmoded. It’s high time we simplify the process for scientists seeking research funding.

Let’s face it, grant writing can be an ordeal.

Parrilla described enjoying scientific ideas, diagrams and writing, while finding grant administration burdensome. However, the process of applying for grants requires a mountain of work that extends way beyond merely communicating an idea for a research project. This is a time-consuming and labor-intensive endeavor.

Typically, grant applications require you to present a standard case for support, outlining your proposed research. But, as any seasoned researcher would attest, there are often several other elements involved. These could range from a lay summary, a long abstract, your CV, impact statements, public-engagement plans, to detailed explanations of staff involvement, project management plans, letters of support from colleagues, data handling strategies, and the projected timeline of the project. And let’s not forget the risk analysis! Acceptance rates vary by funding scheme and year; a single rejection percentage should not be treated as universal.

Despite the extensive preparation, the harsh reality is that once the research commences, things might not pan out as anticipated. Milestones might not be met, some projected outcomes may remain unrealized, and if experiments falter, you might not have the bandwidth to execute all the public-engagement activities outlined in the grant application. Yet, at the end of the project, you may still end up contributing significantly to scientific progress, even if the results diverge from your initial proposal. And that should be perfectly acceptable.

From the perspective of panel members tasked with awarding the grants, the process is far from seamless. Parrilla recalled serving on panels where there was not always enough time to read every application in detail. In his account, panel review focused on three main questions: Does the proposal align with the call brief? Is the proposed science sound and innovative? And are the applicants experts in their field? He described using the abstract and proposal to assess the first two points and online information to learn about applicants’ expertise; this is not a substitute for a funder’s formal review criteria.

So, why do applicants need to produce such an extensive array of documents? The process is intended to support rigorous and fair decisions, but paperwork alone cannot guarantee freedom from bias or establish an applicant’s dedication. However, the creation of these lengthy, seemingly redundant documents is consuming an inordinate amount of scientists’ time.

Enter ChatGPT, the artificial intelligence (AI) chatbot that is shedding light on the system’s shortcomings.

Parrilla reported trying ChatGPT after a colleague suggested it for parts of a grant proposal. He supplied his own research ideas and reviewed the generated material, including an explanation of how the proposal fitted the funding call.

He estimated that this reduced that drafting task from three days to three hours. This was one researcher’s account, not a controlled benchmark, a guarantee of comparable savings or evidence that the grant was funded. Editing a draft should address accuracy and compliance, not conceal required disclosure.

The column also cited a September 2023 Nature survey of more than 1,600 researchers, reporting that over 25% used AI for manuscript writing and over 15% for grant proposals. These are historical survey responses, not current adoption figures or a representative estimate for every research community.

While some may view using ChatGPT for grant proposals as a form of cheating, it underlines a larger issue: why are we asking scientists to produce documents that AI can easily generate? What additional value are we bringing to the table? Perhaps it’s time for funding bodies to reassess their application processes.

For an actual application, check the relevant funder’s current AI-use, confidentiality and disclosure rules. Verify factual claims and references, and keep responsibility for the proposal with its human applicants.