Governance can prevent AI from undermining democracy. But only if it has teeth

Link: In Geneva, the World Can Anchor AI Governance in Free Expression, by Isabelle Anzabi in Tech Policy Press

I like that this is happening, will appreciate the recommendations that arise from it, and know with complete certainty that no major AI vendor will adhere to them unless they are forced:

“The United Nations is convening its Global Dialogue on Artificial Intelligence Governance in Geneva July 6-7, and the stakes are high: what rights will anchor the future of how the world governs AI? Generative AI systems mediate, filter, and generate the information we encounter every day. They are, at their core, a technology of expression and access to information. How we govern them will have downstream effects on what people can say, seek, and know. Geneva is where we can get that right.”

A previous UNESCO meeting recommended governance that “bars AI systems from being used for social scoring or mass surveillance, and  requires member states to ensure that AI actors respect rights in the AI lifecycle”. As we all know, that has definitely happened. Both AI vendors and the two countries that predominantly house them, the United States and China, are famously against surveillance and social scoring, and in favor of maintaining rights. Certainly none of them have, for example, used AI with the most sensitive personal information government has access to, or worked with the private sector to create a surveillance apparatus for policing that includes the broad detainment of immigrants. Neither of them is famous for social scoring — in China by government or in the United States through private enterprise. Job done!

Sarcasm aside, it’s important to have these meetings, and it’s important to continue to add pressure to protect rights and prevent abuses of this technology. But we should be clear-eyed about the fact that any recommendations almost certainly won’t actually be followed unless major nations that represent real dollars to these vendors enshrine them in law and hold the line when countries like the US apply pressure to undermine them.

It can’t just come down to individual governments. The linked article argues that we should also protect freedom of expression, in particular from government (although I’d argue we should also worry about the influence of wealthy private entities). Because these are black box systems, it’s vital that we prevent governments (or anyone with power) from opaquely tweaking their answers and outputs to benefit their agendas. They should not be allowed to be opaque systems; full transparency and auditability should be requirements. The danger is that, in the wrong hands, AI can be used to cement centralized, undemocratic power. To prevent this, ideally, organizations like the UN should apply real sanctions to nations that don’t obey transparency rules and tweak AI systems in service of undemocratic goals — but the UN’s history of doing this effectively is not strong.

Part of the problem is the “AI is the information technology of the future” framing used here. If you believe that AI is the future of information systems, you are also more likely to believe that your nation will miss out if you don’t embrace it completely. (That’s been the marketing: if you don’t jump into AI, you will be left behind.) But the reality is, of course, far more nuanced. AI will absolutely change industries like software engineering, and has already made an impact there, although it’s far from an existential transformation. I’m less receptive to the idea that it will transform entire nations. AI vendors need global markets more than global markets need AI. Governments should understand the power they have, and use it.