Generative AI: Research application and assessment

This briefing note outlines the opportunities and considerations associated with the use of generative AI tools in research funding application and assessment. It also gives examples of potential future uses. Please note, that these need discussion and ideally agreement across the research system to ensure implementation is valuable. AMRC members who are considering the use of generative AI in their research funding processes are encouraged to join our Research Management Working Group or Research Directors’ Forum to facilitate such discussions between members and receive updates on this topic. Visit our Networks and groups page to find out more.

Applications and applicants

While generative AI is still rapidly advancing, some researchers are already using these tools to assist them in developing funding applications. This could include:

Drafting outlines for proposals

Generating standard outlines of sections such as background, objectives, and methodologies
Creating budget templates or providing ballpark estimates of project costs based on input parameters

Editing and proofreading

Refining spelling, grammar and punctuation to ensure accuracy
Improving text clarity and readability
Assisting with translation for applicants whose first language is not English

Data visualisation

Creating visuals or summaries from raw data

Administrative tasks

Automatically filling in the applicant's details from across several funding forms

Generation of potential research hypotheses or identification of research gaps

Literature review synthesis
Trend analysis
Idea storming

Used responsibly, generative AI could provide a range of benefits for researchers and the wider research community. For example, it could save research teams time in writing the initial draft of an application and improving equity in research funding by supporting individuals for whom English is not their first language.

When using generative AI, applicants must do so in accordance with relevant legal and ethical standards, and it is still their responsibility to ensure their application is original, accurate and abides by funder guidelines or policies.
In addition to this, funders, reviewers, and applicants should consider the technology’s limitations:
- the confidentiality, processing and retention of information shared with public generative AI tools, and the potential repurposing of scientific ideas with other users of AI
- the risk of bias and potential need for mitigation in the tools’ outputs.

When discussing the use of generative AI with applicants, funders should state:

How generative AI will and/or won’t be used in the assessment process.
Whether there are any specific restrictions on the use of generative AI in developing applications or other relevant policies. Our current guidance is to allow the use of AI as it may have positive outcomes, but to ask for its use to be declared given that it is not currently possible to identify and police its use.

The following are suggested questions to include in a grant application form, based on those used by Cancer Research UK:

Have any generative AI tools been used when completing this application? (Yes/No)
If yes, please clearly attribute outputs from generative AI tools used in this application. List the generative AI source and where possible name the specific model/s and software used and specify how the content was generated (such as listing the prompt used).

Over time, answers to these questions could help you better understand the influence and impact of AI on application quality and your research portfolio. It is important to explain how this information will be used and who will have access to it, to encourage applicants to answer as honestly as possible.

Research assessment

Some funders are using generative AI at various stages of the research funding process. It is already being used by a number of organisations, within or beyond AMRC members to:

Research strategy and call design

Design grant and application form structure
Search for criteria that deliver 'success'

Expert review

Select expert reviewers

Research reporting

Create lay summaries
Translate lay research abstracts or articles into other languages

When choosing a tool, you must ensure IP and personal and/or scientific data are adequately protected. This may require you to use a confidential in-house tool instead of a public one. It is also important to remember that the technology lacks the ability to think critically to understand context, and that any bias in its training data could be reflected in its outputs.

In 2023, AMRC came together with other UK research funders to release a statement on the use of AI in research application and assessment. This outlines the opportunities, considerations, and risks of using the technology and sets out that expert reviewers will not input content from funding applications or reviews into generative AI tools, nor will such tools be used to develop expert reviews.

Building on this, some funders have established more detailed policies for generative AI use in funding application preparation and assessment:

Developments and future discussions

As the technology progresses, other uses may arise and become more common. Some currently being piloted or considered include the following:

Proposal evaluation

Automated screening
Quality assessment

Project evaluation

Automated data collection and reporting

Optimising resource allocation

Monitoring resource allocation against project needs
Suggesting reallocations or adjustment

Performance analysis and trend detection

Analysing project outcomes against predefined metrics and goals

Reviewer identification

Database mining
Using natural language processing to match keywords and topics in the research proposal with the expertise of potential reviewers

In the future, AI algorithms could be trained to streamline application sifting by assessing application contents against historic data and filtering out ones that are ineligible or unlikely to succeed prior to more in-depth review, reducing expert reviewer workloads. In these instances, a human would still need to review all the rejected bids to ensure consistency and fairness, and funders would need to be transparent about AI use and able to explain the rationale behind the decisions taken by the algorithm.

It’s clear that generative AI has the potential to support researchers, funders and expert reviewers throughout the funding process, but new uses of this developing technology will need to be carefully considered and discussed across the research system before implementation. Through discussion, and ideally alignment, we will be able to identify where implementation will be most valuable, potential risks of implementation, and help to avoid unintended consequences such as people losing valuable skills.

As a funder, you will need to work with stakeholders (including researchers and donor communities) to discuss and clearly communicate why changes are being suggested, the benefits, and the ways this may affect them.