A new initiative, Doing AI Differently, led by The Alan Turing Institute alongside partners from the University of Edinburgh, AHRC-UKRI, and the Lloyd’s Register Foundation, is calling for a shift in how we design AI. Instead of treating AI outputs as cold calculations, the team argues they should be viewed as cultural artifacts—more like novels or paintings than spreadsheets.
The problem, says Professor Drew Hemment, is that current AI lacks “interpretive depth.” It can generate information but struggles with nuance, context, and meaning—especially in areas where human understanding is essential.
Breaking the homogenisation problem
Today’s AI systems are often built on a small set of similar architectures. The report warns this “homogenisation problem” creates shared blind spots and repeated biases across countless tools. The team points to social media as an example of how limited design goals can lead to unintended societal consequences.
Their solution is Interpretive AI—systems designed from the ground up to handle ambiguity, multiple perspectives, and rich context. This involves exploring alternative architectures and fostering human-AI “ensembles,” where technology complements human creativity rather than replacing it.
Real-world impact
Interpretive AI could change sectors like healthcare, where a patient’s story is more than a list of symptoms, or climate action, where local culture and politics shape how solutions are received. By understanding these nuances, AI could produce outcomes that are both effective and trusted.
A global call to action
A new UK-Canada funding program aims to unite researchers to embed interpretive capabilities into future AI. For Lloyd’s Register Foundation, which is supporting the effort, the goal is to ensure safety and reliability in any AI system developed.
“We’re at a pivotal moment for AI,” Hemment warns. The team’s vision is not just to build better algorithms, but to create AI that amplifies human values, addresses our biggest challenges, and strengthens trust between people and machines.
