AI Technology for Grant Writing: Proposed Rules for Academic Funding

You may, or may not, be surprised to hear that even GPT3 can be used for grant writing to create a plausible sounding grant application for NIH’s R21 program. In the example below, an associate dean at the University of Michigan was trying to evaluate the capabilities of the AI model in formulating a comprehensive grant proposal.

The results were strangely compelling and took about an hour to prepare. Given this example is 3 months old now (an eternity in the fast-moving world of AI and GPT), it is not difficult to imagine engineering a series of prompts to get you 80% of the way toward a first draft for most academic funding calls.

The methodology was as follows:

• ChatGPT was given a background and specific aims to include in the grant proposal.
• The AI model was asked to write the Specific Aim page, Background, Introduction, and sections on safety and the necessity for academic funding.
• After completing these sections, the AI was asked to write an Innovation and Impact section.
• The experimental design was broken down into tasks for each aim, and the AI was asked to provide details for each task.
• Finally, the AI was asked to develop a timeline and a future perspective for the proposal.

You can watch the whole process here and even copy the prompts used if you wish.

However, we should proceed very carefully.

ChatGPT generates text based on patterns and structures it learns from input data. By providing it with information about a specific project or organization, it can help answer questions and generate text for grant applications. While using AI for grant writing can be helpful, it’s essential to be cautious about the output, as the technology can sometimes generate false information or lack proper references.

While GPT may be useful in generating initial drafts, writing cover letters, and creating reports, it’s crucial to understand that AI technology cannot replace human involvement in the grant writing process entirely. You will still need people who can think strategically, analyze complex data, and build relationships with funders (an important and much-overlooked component of successful grant applications).

And then there is a human cost to all of this. If it takes an hour to prepare a reasonable grant application (let’s define that as one that the program manager is happy to send out for peer review), then what’s to stop an individual, lab, or department flooding that funder with – on the surface of it – applications which aren’t junk. Hundreds of extra grant applications will need reviewing, peer review will likely break down, and funders may have to consider other means of allocating funds to applications (like lotteries).

It’s hard to predict where these developments may take us (shorter applications with less narrative text could be one result), but for now we should probably define our own ethical use of AI in grant applications.

1. Use AI to help you prepare an application you were going to make anyway – don’t use it to flood the system with similar-but-different project proposals.
2. Use AI to help you formulate your idea, get creative with solutions, and come up with a fundable, viable research project.
3. Use AI to create a background to your project with certain research papers as basis for that (although this will still need checking).
4. Use AI to – possibly – create a first draft of each section of the grant application.
5. Don’t use AI to create ‘preliminary data’, tables, or figures. And absolutely not to create falsehoods related to your grant application.
6. Don’t rely on AI to be right. Human expert review BEFORE you submit will (for now) always be the gold standard for catching errors, repetition, and omissions.

There are parts of a grant application that AI can’t help with. For instance, AI can’t call the program manager to clarify details, establish a human bond, be memorable to the person who may have to argue for your research project on your behalf. Need to create graphs/images to complement the text? It’s still best to do this manually.

So, while AI can help us get ‘somewhere’ quickly with a first draft, it isn’t the answer to everything just yet. The ‘somewhere’ you’re heading may not even be the right place the funder is interested in. Always count on your colleagues or grant experts to refine your thinking, wording, and overall application. AI can make the process of grant applications quicker, but it takes more than that to make them better all-round.

For expert assistance in preparing your research funding grant, visit GrantDesk.