New research published in Royal Society Open Science presents a new framework to help understand the decisions and considerations that go into AI use in practice. We spoke to author Dr Denis Newman-Griffis to find out more about the framework.
Tell us about your paper
AI, or artificial intelligence, has really blown up in the last couple of years since the launch of consumer-facing generative AI platforms like ChatGPT and Gemini. But the way that we think and talk about AI is still based on a model of technical expertise: that only computer scientists and big tech firms have the ability to understand AI and shape how it works, and that using AI is about having the right computing and data infrastructure. This paper shows that there is much more to using AI in practice than just the technology itself, and illustrates some of the key competencies and kinds of decisions that make AI use effective and ethical—or not.
The paper describes a model called AI Thinking, which breaks down the process of using AI into five skills of approaching AI in terms of informing specific processes, how you define what an AI application is meant to accomplish, comparing and contrasting different options for AI technologies, understanding and choosing between data sources for use with AI, and grounding use of AI in the specific contexts where it will be applied.
How did the idea for the paper come about?
My research has always focused on applied AI in one form or another, and I’ve always been struck by the gap I often see between some of the narrower, more technical ways that we talk about AI in research and the huge variety of ways it is used in practice, which raise all kinds of interesting research questions at the interface between people, process and technology.
As part of the GRAIL project I’ve been leading with the Research on Research Institute, where we’re looking at responsible use of AI in research funding and evaluation, I had the chance to speak at the 2023 Workshop on the Economics and Policy of Scientific Research, where I first talked about this idea of the skills involved in AI application as a specific thing to understand and develop. The discussion there made it clear that this idea could help capture something that we were really struggling with in making AI a bit more tangible, especially right after the launch of ChatGPT, when it was clear we were at a moment of transformation but we weren’t sure yet what shape that would take. The subsequent research we’ve conducted in GRAIL and in another project looking at responsible AI across sectors, FRAIM, has really helped to surface a lot of the very diverse aspects that affect use of AI in practice and whether it is effective and ethical or not.
So, I wrote this paper to help bring together what we’ve been learning about AI applications with the ideas of building up better skills around AI, and to make it a bit more tangible what it is we do when we’re using AI.
Who and what is AI Thinking intended for?
AI Thinking is a framework to help structure and understand what goes into using AI in practice. It’s quite a general purpose: it can be useful for AI developers, to help map out how new technologies might be used; for AI users, to walk through the key decisions they need to make to use AI most effectively; for team managers and decision makers in organisations using AI; and even for each of us as people whose data is being analysed with AI. I have already used AI Thinking in my own teaching, and educating AI developers and users is one of the key benefits it can bring. Its biggest value, though, comes from its ability to help bridge the gap between very different ways of thinking about AI and what goes into its use, and to help teams work together to use AI effectively.
The AI Thinking framework helps provide a guide to some of the most common and important processes that go into AI use, and makes that process a bit more understandable. It doesn’t cover everything that goes into each of these cases, of course: instead it provides a strong foundation to build on for the different contexts and situations where AI is used.
The AI Thinking framework (reproduced from Newman-Griffis, 2025)
What are the future directions for your work?
One interesting direction is to build on the competency model that AI Thinking is built around by getting more into the specific skills that AI users, developers, policymakers, and so on need to have in an ‘AI everywhere’ world. Competencies need to be learned, and the best way to support that is by focusing our efforts in training and education on getting people not just the skills to understand AI, but also to put it in context and make practical, everyday use of it.
Another direction this raises is the need to think more deeply about what interdisciplinary practice in AI looks like. AI use is quite interdisciplinary by nature, but we often struggle to work effectively across different ways of thinking and different understandings of knowledge and data. The AI Thinking framework can serve as a good way to ground discussions about AI that cross boundaries between academic disciplines or different sectors in a shared understanding of AI use, and this can help us start to understand how to make interdisciplinarity in AI more effective.
How was your experience publishing with Royal Society Open Science?
Great! It’s not easy to find a place to publish a paper like this that is keen to change the way we think about a particular discipline, especially when it draws on a number of different academic traditions and ways of thinking. It was important to me to find a journal that could cut across disciplinary boundaries and was open to publishing work that speaks to a number of different audiences. I got some great early support from the Royal Society Open Science team in understanding the different sections and article types in the journal and what would be most appropriate for this piece, and I got excellent feedback in the peer review process that really helped strengthen the article. I’m excited to have published this work in a journal with a wide audience and to be working with the Royal Society Publishing team.
Read the full paper, AI Thinking: a framework for rethinking artificial intelligence in practice
Royal Society Open Science is an open access journal that welcomes the submission of all high-quality science. More information about the submission process can be found on our webpage.
Images: Top image from iStock, by monsitj; AI Thinking Framework, Figure 1 from Newman-Griffis, 2025, Royal Society Open Science
