Ramez Naam
It seems we will now be entering a phase of AI congressional hearings and legislation. There are many no-regret ways to increase AI safety. Mandatory transparency. Audits. An NTSB-like investigatory safety agency that identifies root cause of incidents. Proactive ways to use AI and other tools to bolster digital and biological security. Incentives and anti-trust exceptions to encourage lateral sharing on safety and alignment techniques between labs. Liability law to align lab incentives even further with society at large. There are also policies that would impede the positive developments of AI, or which would give governments undo influence on AI models, or dangerous new powers of surveillance and censorship. I want to register again that loss of control is not the AI risk that worries me most. It's over-control that leads to less freedom and a more authoritarian system of government that I see as the most plausible large scale harm of AI to society. As we have these debates over the coming weeks and months, I urge everyone to ask of any policy: How would a government use this for ill? How could this be used to surveil, suppress, manipulate, or control a populace? We have no hard evidence that AI poses an existential threat to humanity. That doesn't mean we shouldn't be cautious and embrace no-regret policies. But in doing so, let's not make the obvious and most historically likely threat - further rollback of democratic freedoms - more likely. Thanks.