FairGPT: owning more than the playlist.

Meliora Managing Partner Clive Dickens presented at Beyond The Music in Manchester on 9 October, as part of the Creative Industries AI Summit convened by Prime Minister Andy Burnham. His keynote, The UK Must Own More Than the Playlist, made the case for one idea: FairGPT.

The UK is one of the world's great exporters of culture. Its music, film, fiction, fashion and design travel further than almost anything else it makes. Yet when AI learns from that culture, the UK owns almost none of the systems doing the learning.

A Commonwealth model

FairGPT is a proposal. It sets out how a sovereign, fair AI model could be governed, powered and trained, and it is built for the Commonwealth rather than for one country.

Outside the United States and China, countries are already building models of their own. France has Mistral, Germany has PhariaAI and Switzerland has Apertus. India has Sarvam, Canada has Cohere and Singapore has SEA-LION. Indonesia has Sahabat-AI and Japan has Fugaku-LLM.

Two countries are missing from that list: the UK and Australia. Both have world-class creative industries. Neither has a model of its own.

FairGPT rests on three pillars: Fair Governed, Fair Powered and Fair Trained.

Fair Governed

Fair Governed is about who sets the rules: rules, regulation and accountability. The keynote borrowed from fields that already manage powerful technology. Breaches are reported, as they are under data protection law. Models are approved before release, as medicines are. And limits are agreed between nations, in the spirit of the nuclear non-proliferation treaty.

Fair Powered

Fair Powered is about capacity: the compute, models and data centres a country needs to run AI on its own terms, and the energy behind them. Australia leads the world in rooftop solar. The UK is a world leader in offshore wind. Switzerland's Apertus shows that a public, open model can be built outside the largest technology companies.

This pillar is still taking shape. These examples are a starting point, not a finished framework.

Fair Trained

Fair Trained is about provenance. Every model learns from something. The question is whether that material was licensed, scraped or stolen.

FairGPT proposes a fair exchange of value. Rights holders license their work, and they are paid when it is used.

That value runs deeper than the music and book catalogues most people think of. The keynote pointed to three archives: the BBC archive, the James Hyman archive of music and pop culture magazines, and the History of Advertising Trust. Each holds decades of creative work. None of it should train a model without permission and payment.

Why Manchester

The setting mattered. Manchester's history runs from the world's first industrial city to the first stored-program computer. Alan Turing worked at the University of Manchester, and in 1948 the Manchester Baby ran the first program stored in a computer's own memory. The city knows what happens when new machines arrive faster than the rules for them.

It will soon host the next chapter. Manchester welcomes the G20 leaders' summit in November 2027, and Prime Minister Andy Burnham has called the UK an "honest broker" on AI. FairGPT is the kind of idea a host country could bring to that table.

What comes next

FairGPT is not adopted policy, and it is not an announced partnership. It is an invitation to build a fair model for licensing, governance and infrastructure, together with creators, rights holders, governments and technology companies across the Commonwealth.

To be part of that conversation, get in touch.

In July we asked the same question from Australia: The Prime Minister, Anthony Albanese, has written the rules for a game Australia doesn't play, yet.