Free and Open Source Software (FOSS) refers to software that is freely available to use, modify and distribute. Unlike proprietary software, which restricts access to its source code, FOSS promotes transparency and collaboration by granting users access to the underlying code. This freedom is essential for fostering innovation, as it lets developers worldwide improve upon the software and adapt it to meet specific needs.
What is FOSS?
Free and Open Source Software (FOSS) refers to software that is freely available to use, modify and distribute. Unlike proprietary software, which restricts access to its source code, FOSS promotes transparency and collaboration by granting users access to the underlying code. This freedom is essential for fostering innovation, as it lets developers worldwide improve upon the software and adapt it to meet specific needs.
FOSS licences govern the use, modification and distribution of the software and generally fall into two main categories: ‘copyleft’ licenses which require derivative works to remain open-source, and ‘permissive licenses’ which impose minimal restrictions and often allow commercial use.
What is the relationship between FOSS and AI?
The rapid development and use of AI tools, which often build on previous work, naturally intersect with FOSS. AI tools like GitHub Copilot and OpenAI Codex leverage FOSS to train their models by using machine learning algorithms to analyse patterns within publicly available code databases, generating or optimising code automatically. This use of FOSS in training AI models reduces software development costs and fosters innovation by allowing developers to build on existing code.
Beyond its technical advantages, the use of FOSS in AI development has sparked widespread debate regarding its business implications and ethical considerations. Notably, Meta's Chief AI Scientist, Yann LeCun, highlighted the advantages of open-source AI models, stating, “The key takeaway is that open-source AI models are surpassing proprietary ones.”
Meta's CEO, Mark Zuckerberg, articulated the broader implications of open-source AI, asserting, “Open-source AI is good for the world and therefore a platform that will be around for the long term.”
A recent Business Insider article goes on to explain that “those who support open-source say it allows technology to develop rapidly and democratically since anyone can modify and redistribute the code. On the other hand, advocates for closed-source models argue that they're more secure because the code is kept private”.
However, as AI models increasingly rely on FOSS, questions around legal compliance and intellectual property rights have become more pressing. The FOSS code used for training these models may have licences that limit how the models are distributed, potentially requiring the AI model's code to be freely released under the same licence terms. This could limit the AI model provider from obtaining and enforcing its intellectual property (IP) rights on the particular embodiment that includes the FOSS code. Additionally, if an AI model trained on FOSS code generates code in response to a user's prompt, there's a risk that the AI-generated code includes FOSS under a copyleft licence, necessitating that the entire codebase be open-sourced.
What’s next?
To address these risks, organisations are implementing measures such as automated code scanning tools, clear policies on FOSS usage, and training for developers on licensing compliance. Regulatory developments are also on the horizon. The EU's AI Act, for example, mandates that providers of general-purpose AI models (GPAI) must observe intellectual property rights, including by drafting and publicly disclosing a sufficiently detailed summary of the content used for training these AI models (a requirement that is considered onerous). Given that the Act’s exceptions for open-source systems are limited, the full impact on FOSS development remains to be seen.
What is the Responsible AI License? The Responsible AI License (RAIL) is a type of open-source license specifically designed for Artificial Intelligence (AI) projects. It aims to address the ethical and legal complexities of deploying AI systems by introducing restrictions and guidelines around their usage. While traditional open-source licences focus primarily on granting freedom to use, modify, and distribute software, RAIL introduces responsibility clauses to ensure that AI technologies are used in ways that align with ethical standards and avoid harm. |
Additionally, the creation of AI-specific open-source licenses, such as the Responsible AI License (RAIL), may help address the ethical and legal complexities of AI-generated code.
Looking ahead, the landscape of AI and FOSS is expected to evolve. Developers face choices between proprietary models, which offer polished tools in closed ecosystems, and open-source frameworks like Meta’s LLaMA, which provide transparency and customisation opportunities but still have limitations.
Finally, as courts and regulators continue to engage with these issues, the future of FOSS and AI will hinge on finding a balance between fostering innovation, maintaining transparency, and ensuring compliance with legal and ethical standards.