Abstract:  This paper examines ‘open’ AI in the context of recent attention to open and open source AI systems. We find that the terms ‘open’ and ‘open source’ are used in confusing and diverse ways, often constituting more aspiration or marketing than technical descriptor, and frequently blending concepts from both open source software and open science. This complicates an already complex landscape, in which there is currently no agreed on definition of ‘open’ in the context of AI, and as such the term is being applied to widely divergent offerings with little reference to a stable descriptor.

So, what exactly is ‘open’ about ‘open’ AI, and what does ‘open’ AI enable? To better answer these questions we begin this paper by looking at the various resources required to create and deploy AI systems, alongside the components that comprise these systems. We do this with an eye to which of these can, or cannot, be made open to scrutiny, reuse, and extension. What does ‘open’ mean in practice, and what are its limits in the context of AI? We find that while a handful of maximally open AI systems exist, which offer intentional and extensive transparency, reusability, and extensibility– the resources needed to build AI from scratch, and to deploy large AI systems at scale, remain ‘closed’—available only to those with significant (almost always corporate) resources. From here, we zoom out and examine the history of open source, its cleave from free software in the mid 1990s, and the contested processes by which open source has been incorporated into, and instrumented by, large tech corporations. As a current day example of the overbroad and ill-defined use of the term by tech companies, we look at ‘open’ in the context of OpenAI the company. We trace its moves from a humanity-focused nonprofit to a for-profit partnered with Microsoft, and its shifting position on ‘open’ AI. Finally, we examine the current discourse around ‘open’ AI–looking at how the term and the (mis)understandings about what ‘open’ enables are being deployed to shape the public’s and policymakers’ understanding about AI, its capabilities, and the power of the AI industry. In particular, we examine the arguments being made for and against ‘open’ and open source AI, who’s making them, and how they are being deployed in the debate over AI regulation.

Taken together, we find that ‘open’ AI can, in its more maximal instantiations, provide transparency, reusability, and extensibility that can enable third parties to deploy and build on top of powerful off-the-shelf AI models. These maximalist forms of ‘open’ AI can also allow some forms of auditing and oversight. But even the most open of ‘open’ AI systems do not, on their own, ensure democratic access to or meaningful competition in AI, nor does openness alone solve the problem of oversight and scrutiny. While we recognize that there is a vibrant community of earnest contributors building and contributing to ‘open’ AI efforts in the name of expanding access and insight, we also find that marketing around openness and investment in (somewhat) open AI systems is being leveraged by powerful companies to bolster their positions in the face of growing interest in AI regulation. And that some companies have moved to embrace ‘open’ AI as a mechanism to entrench dominance, using the rhetoric of ‘open’ AI to expand market power while investing in ‘open’ AI efforts in ways that allow them to set standards of development while benefiting from the free labor of open source contributors.