Claims of 'open' AIs are often open lies, research argues

Rhetoric around "open" AI concentrates power in the AI sector rather than making it more open to competition and scrutiny, according to a research paper published in Nature.

The research sets out how the often-promoted concept of "open" AI can mislead the public and policy-makers, leading to a false sense of security.

"Claims posited around openness often lack precision, frequently focused on only one stage in the development-to-deployment life cycle of AI systems, often neglecting substantial industry concentration in large-scale AI development and deployment and thus warping common-sense understandings of openness," the paper from David Widder, Cornell University post-doctoral fellow, states.

The research compares concepts of free and open source software with that of open AI, looking at IBM's history with Linux, Google with Android, Amazon with MongoDB, and Meta with PyTorch.

"From the promise that open source democratizes software development, that many eyes on open code could ensure its integrity and security, or that open source levels the playing field and allows the innovative to triumph, open source software did many of these things, to varying degrees," the paper says.

Leaving Reg readers to debate precisely how varied those outcomes were, the point is open AI does not work in the same way as open source software.

"At present, powerful actors are seeking to shape policy using claims that 'open' AI is either beneficial to innovation and democracy, on the one hand, or detrimental to safety, on the other. When policy is being shaped, definitions matter," it adds.

The research analyzes what AI is and what "openness" amounts to in the AI world. It looks at models, data, labor, frameworks, and computational power.

"Just as many traditional open source software projects were co-opted in various ways by large technology companies, we show how rhetoric around 'open' AI is frequently wielded in ways that exacerbate rather than reduce concentration of power in the AI sector," the paper says.

The research finds open AI systems can "offer transparency, reusability and extensibility: they can be scrutinized, reused and built 'on top of,' to varying degrees."

The paper singles out Meta's LLaMA-3 as lacking openness in the common sense of the word, since it offers "little more than an API or the ability to download a model subject to distinctly non-open use restrictions ... In these cases, this is 'openwashing' systems that are better understood as closed."

• We can clone you wholesale: Boffins build ML agents that respond like specific people
• One thing AI can't generate at the moment – compelling reasons to use it for work
• AI Jesus is ready to dispense advice from a booth in historic Swiss church
• Amazon bets another $4B on Anthropic

The Register has offered Meta the opportunity to comment.

At the other end of the scale, EleutherAI's Pythia is the most open, "offering access to the source code, underlying training data and full documentation, as well as licensing the AI model for wide reuse under terms aligned with the Open Source Initiative's longstanding definition of open source."

However, even the most open interpretation in AI is unlikely to counter the vested interests of tech giants in the AI space because the data, development time, and computational power required to build substantial models create barriers to entering the market.

In the case of data, for example, the labor required to build contemporary AI systems "presents another barrier to democratic and open access to the resources required to create and deploy large AI models."

The research argues that on its own, "open AI" is not enough to get a more "diverse, accountable or democratized" industry.

"We also see that, as in the past, big tech companies vying for AI advantage are making use of open AI to consolidate market advantage while deploying the rhetorical wand of openness to deflect from accusations of AI monopoly and attendant regulation," the paper states.

As a consequence, other measures, including antitrust enforcement and data privacy protections, are necessary to create a more even playing field.

"Pinning our hopes on 'open' AI in isolation will not lead us to that world, and – in many respects – could make things worse, as policymakers and the public put their hope and momentum behind open, assuming that it will deliver benefits that it cannot offer in the context of concentrated corporate power," the researcher concludes. ®