Mark Surman is President of Mozilla, a global nonprofit-backed technology company that does everything from Firefox, a free open source Web browser, to advocating for a more open, equitable Internet. His current focus is ensuring the various Mozilla organizations work in concert to make trustworthy AI a reality. Surman led the creation of Mozilla.ai (a commercial AI R&D lab) and Mozilla Ventures (an impact venture fund with a strong focus on AI). Before joining Mozilla, he spent 15 years leading organizations and projects that promoted the use of the Internet and open source as tools for social and economic development. Surman was a speaker at the Tony Blair Institute for Global Change’s “Governing in the Age of AI” event in Feb. 9 in Paris and an attendee at the AI Action Summit February 10 and 11. He spoke to The Innovator about how countries and companies can use open source AI to obtain economic positioning and growth.
Q: Talk about open source AI was an important part of the conversation at the AI Action Summit, with everyone from European Commission President Ursula Van der Leyen to former Google CEO and Chairman Eric Schmidt promoting it. Does this signal a turning point?
MS: AI is in such a different spot than during the last meeting. At the UK Bletchley Park AI Safety meeting [in November 2023] open source was wrongly and cynically painted as the bad guy. Just 15 months later, the tide has shifted. With the world gathering at the AI Action Summit in France this week countries are embracing openness as a key component of making AI safe in practical development and deployment contexts. This is an important turning point.
Q: The ROOST announcement during the AI Action Summit seems to reinforce that notion. Can you talk a little about that?
MS: ROOST (Robust Open Online Safety Tools) was announced during the summit as a nonprofit dedicated to making AI safety tools open, accessible, and widely available. Anyone—startups, nonprofits, researchers—can use ROOS. It is launching at exactly the right time and in the right place, using the global AI Action Summit to gather a community that will create the practical building blocks we need to enable a safer AI ecosystem. Mozilla helped seed this work, supporting early research at Columbia University that laid the groundwork for ROOST. Now, backed by tech leaders and philanthropies like the Patrick J. McGovern Foundation, it’s taking off as an independent effort to build the public infrastructure AI desperately needs. ROOST is like the Apache or Linux foundations for the AI era. It is starting with obvious near-term risks in a practical and robust way and will become the critical infrastructure as huge waves of business and society start using GenAI.
Q: There is a lot of talk at the summit about governments using AI. What about business?
MS: For the past 20 years many large companies have been powered by open source. Whether you are a government, a bank or some other kind of big company your network is likely to be sitting on top of Linux, Apache, and an open source database. Open source is the plumbing of the digital world. A Harvard Business School study found that many companies build their businesses on open source software, code that would cost firms $8.8 trillion to create from scratch if it weren’t freely available. The $8.8 trillion number represents the demand-side value of open source software—if it did not exist at all, and every company that used it had to rewrite that software from scratch. From the supply side, according to this research it would cost some $4.2 billion to build those software packages—if open software existed, but the most widely used packages were deleted and needed to be rewritten.
Now interest in open source AI is growing. A recent survey of more than 700 technology leaders and senior developers across 41 countries by the Mozilla Foundation, McKinsey and the Patrick J. McGovern Foundation provides the largest and most detailed analysis of how enterprises are thinking about and using open source AI. The research shows that enterprises are using open source more than one might expect and some 76% of respondents expect their organizations to increase use of open source AI technologies over the next several years.
Across several areas of the AI technology stack more than 50% of respondents’ organizations report using open source AI technologies, often alongside proprietary tools from players such as OpenAI, Google, and Anthropic. The research shows that organizations placing a high priority on AI are most likely to use open source technologies and those that view AI as important to their competitive advantage are more than 40% more likely to be using open source AI models and tools than respondents from other organizations.
Q: There is some confusion around the word open. What does open AI actually mean?
MS: The Open Source Initiative [ a California based authority recognized globally by individuals, companies and by public institutions] defines open source AI as having three things: one is weight, which means ability to configure the model. The second is the data used to train the model. It needs to be transparent which means that if you wanted to you could retrain the model using the same data. The third has to do with tooling used to frame the model: you need to be able to tune it for the model to be truly open source. Very few players fully conform. Llama and DeepSeek, for example, offer open weight, but their offer is like a street hotdog: you know it is a hotdog, you just don’t know what’s in it. An example of fully open stack AI models are those coming out of the Allen Institute for AI in Seattle. Last year it released a new multimodal model called Molmo and the year before it released Open Language Model or OLMO as part of a larger effort to bring more transparency to AI models. These models are good enough to be used by business and fully open. We think that ultimately Meta will go fully open source and will have to tell people what is in the hot dog and clean up the list of ingredients. The competitive advantage is not in the base model. It is in performance. That is the special sauce. The rest can be open source.
Q: What is Mozilla’s role in all this?
MS: We have been trying out AI features on Firefox very slowly so as not to violate our principles. The other thing we are doing, which is still very experimental, is a chatbot sidebar that allows people to choose between Gemini, Mistral, ChatGPT and Hugging Face – models that are open source and closed source – and jump between models. We have also created and spun out a company called Mozilla.ai to make open source AI easier to adopt. It is already doing some basic things. If offers, for example, something called Lumigator which helps developers to evaluate, refine, and select models that best fit their project needs. The bigger vision is helping to make it easier to pull the whole stack together. The barrier to open source AI adoption right now is not availability of open source components, it is ease of adoption, and we think we can play a role there. My job is to help Mozilla find new centers of gravity, so we have also launched a venture fund which is mostly AI focused.
Q: What do you hope will come out of the AI Action Summit?
MS: I really hope that companies and countries outside of the U.S. and China find common cause around open source and build their own industries and infrastructure on top. Governments should really lean into the open science community to work together to build some of the core infrastructure at the software and data level. Sovereignty comes from collaboration. [French President Emmanuel] Macron just announced $109 billion in AI infrastructure. If similar investments were made in the rest of the EU countries, you would end up with a number as big as or surpass the investments being made in the U.S. Countries need to get the message that you are not going to be able to go it alone and achieve sovereignty. If you pull resources together and out invest the bigger AI players on the underlying infrastructure you can build sovereignty on top of that. Countries want their AI models to be trained on local data in local language, but this is late stage finetuning and doesn’t require countries to build all the pieces themselves. About 20% of the underlying infrastructure can be pooled. It will give countries a robust head start, and then they can build their own sovereignty from there. This is where the opportunity lies. Sovereignty through collaboration is how to achieve both cultural and economic success, otherwise countries are going to end up as branch plants of the West Coast of America and China.
What I hope is here will be another summit that will go even further and include even more small countries. I’d like countries to work in common on open source AI as a way to get the economic positioning and growth that they need. Ideally, we will end up with a third AI bloc banding together.
This article is content that would normally only be available to subscribers. Sign up for a four-week free trial to see what you have been missing.
