Doing AI Differently: Policy Note
Year: 2025
Type: article
Abstract: This policy note summarises key recommendations from the Doing AI Differently white paper – an international research initiative that outlines how interpretive capabilities can be embedded into the foundations of AI development, from frontier models to domain-specific systems already in use.The United Kingdom has declared its ambition to become a global AI superpower. The Modern Industrial Strategy places technology at the heart of the UK’s future prosperity, and the AI Opportunities Action Plan provides a roadmap for this ambition. The success of this ambition hinges on overcoming a critical challenge: the implementation gap. This is the gap between an AI’s performance on technical benchmarks and its effectiveness in real-world settings. As AI undergoes a ‘qualitative turn’ – becoming embedded in the fabric of our culture, language, and society – its success cannot be measured by technical performance alone. Overlooking the complex, interpretive, and deeply human context in which AI operates risks building a brittle and distrusted technological future, repeating the societal missteps of previous technological waves.What is needed is interpretive depth: AI systems that can understand context, cultural nuance, and multiple valid perspectives – rather than producing single “correct” answers. The UK is doing excellent work in driving responsible AI in deployment, but building AI systems capable of interpretive depth requires humanities practitioners as co-designers upstream when these systems are designed and built. This capability is essential for performance and adoption in real-world deployments – from sustainability to healthcare – and for sovereign systems inclusive of and reflective of British culture, diversity and values.
Authors: Hemment, Drew, Nathan, Christopher
Institutions: The Alan Turing Institute, University of Edinburgh
Language: English
Subfield: Safety Research
Field: Social Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W6893394240
FWCI:
Cites:
Cited by:
Related to: