Artificial Intelligence should not write your research proposal
EFMC - European Fund Management Consulting
Pre-award & post-award services, trainings and impacts of EU funded projects.
These days the temptation to use AI assistant like ChatGPT for proposal writing is increasing as it makes writing?“faster”?and it is a?“free platform”?to write engaging text for readers. The ability to rapidly gather millions of data and generate convincing content, ChatGPT seems like an easy solution to use.
Fascinating as it is, ChatGPT is a great tool that can help to generate ideas and streamline the writing process, but it should not function as a replacement for human writing and editing!
ChatGPT creates an?“Artificial Content” and not an “Authentic Proposal”
Proposals are not supposed to be too theoretical – meaning you should be describing YOUR actual capabilities, approach, methods and resources; and that foundation of data is not something that ChatGPT has currently access to. Even though, ChatGPT may be able to describe how a requirement ought to be addressed, it is not going to describe how you and/or your partners would address it in the real world using the real resources.
领英推荐
AI assistants are not works of logic; they are works of words!
Large language models (LLMs) are good at producing convincing text which looks right and for many applications that is enough (e.g. marketing brochures, translations of web sites, etc.); however, it is not enough for scientific discoveries – as science is not about word sequencing.
Any winning proposal must convince the evaluators -which are still real people- through specific examples, quantifiable metrics, preliminary results and successful past performance that they are the best researchers to perform the work.
Currently, ChatGPT do not have the capacity to invent, to reason, to form bridges between concepts, or more importantly to follow logic to make conclusions..... ChatGPT cannot fact check or experiment to verify if something is correct in the physical reality - we live in. Furthermore, ChatGPT do not form specs, create something and then check whether what they create (software, recipe, etc.) is in accordance with the specs.
In conclusion, for AI to be truly useful as an assistant in scientific research proposal writing, the above features are needed and they are currently not yet met!!!?