Open-source is the secret sauce that will power the future of AI
Open-source – code made freely available to be redistributed and modified – has arguably been saddled with boring and technical branding, but it is an incredible powerful tool for challengers and essential to unlocking the potential of AI, says Camilla de Coverly Veale
This week the government published their response to AI-superstar Matt Clifford’s AI Opportunities Action Plan. The plan focuses on how the UK can capitalise on the benefits of AI to transform the economy and public services. In his foreword the PM writes: “Shaping a successful AI future for Britain requires us to act and invest now. Failure to do so will condemn us to merely obeying decisions made by others beyond our borders.”
When it comes to nurturing open-source AI in the UK, this warning by the PM could not be more prescient.
Open-source – code made freely available to be redistributed and modified – has arguably been saddled with boring and technical branding. Admit it, your eyes just glazed over a bit. But for those in the know, it is the beating heart of what has made the internet great. And it’s been the internet’s white knight – it was the open-source community in the early 90s that fought off the telcos and then Microsoft when they made their bids for complete control. Now that community is in the next great fight unfolding over who controls AI.
Openness is a powerful tool for a challenger. In 2023 a leaked internal-memo at Google described an AI open-source community daily striking fear into the hearts of the Tech Giants. Open-source innovation is nimbler, more flexible and – because users can fine-tune models to work for them personally – doing things bigger companies would never have thought – or bothered – to do. The components of open sourced AI are also by definition transparent, so the innovation built with it increases access to safer and more trustworthy systems. The leaked memo sums up the threat well: “we have no secret sauce…open source AI will outcompete”.
Making it harder for tech giants to gatekeep
Tech Secretary Peter Kyle should grab this opportunity with both hands and pull it close – particularly in light of his recent comments that the UK “needed to show a ‘sense of humility’ when dealing with companies such as Google, Microsoft and Meta”. Growing the UK’s open-source AI community reduces our dependency, spurs competition and democratises a powerful technology. Nurturing this makes it harder for tech giants to gatekeep and gives British businesses adopting AI more options for less money. And for startups building on top of open-source LLMs – well they don’t have to live every day in fear that their business model could be washed away with the flick of a TechGiant’s wrist.
Without a strong open-source AI community as a counterbalance we will all have to be a lot more humble.
Crowding in support for UK open-source shouldn’t be controversial – the UK Government is itself spearheading a globally renowned open-source AI safety project in Inspect – its AI Safety Institutes’ testing platform to evaluate LLMs. And research by OpenUK shows the UK contains a lot of open-source innovation. But being controversial is not the same as being an afterthought. And for too long open-source has been an afterthought – it’s difficult to see so difficult to help. But our open-source community needs targeted support. Other countries are seizing this opportunity and running with it – we need only to look across the Channel to France. Mistral – perhaps the most famous European LLM and one who has just expanded into the UK – is open-sourced. Recently the FT commented “French AI start-ups favour open-source models…that, they hope, will give them a competitive edge in applying AI to almost every sector of the economy”.
So many of the UK’s big questions over how to capture value in AI – and increase public trust – are partially or wholly answered by boosting open source innovation. And, as the geopolitics around AI tightens, open-source will only work to boost the UK’s international resilience: after all, you cannot exactly stockpile access to AI models.
As the Government digests the Plan and considers how to implement its response, Mozilla stands ready to help. Many of the recommendations look set to support the open-source community by osmosis – but when it comes to implementation we could do more and better targeted.
From the disruption of the industrial revolution to Wilson’s white heat of technology, the Labour movement has always understood the need to harness tech for the many
Crucially, we need to include provision for the open-source community in the upcoming AI Bill. This isn’t about making the Bill cover everything-AI – it absolutely needs to be targeted and nuanced, it’s about ensuring open-source can compete with closed-source AI. If only one approach gets a set of guardrails and the other is left with basic questions such as liability unanswered the first will have a massive competitive advantage.
From the disruption of the industrial revolution to Wilson’s white heat of technology, the Labour movement has always understood the need to harness tech for the many. As a powerful general purpose technology, all of AI is a huge opportunity – but in its response to this Plan the Government will have to pinpoint where it thinks the real value will accrue to the UK: supporting open-source must be at the heart of that.
Camilla de Coverly Veale is head of UK policy at Mozilla Foundation