The prospect of artificial superintelligence (ASI) raises a fundamental question: who controls it, and what happens to human freedom? One proposal gaining attention is democratizing ASI, ensuring many people have access to this powerful technology to preserve civil liberties. But is this plan viable, or does it introduce risks as grave as the problems it aims to solve?
The Case for Democratization
A unifying driver behind many catastrophic risks that remain even if we solve the alignment problem is that power is relatively evenly distributed across people today. The most powerful person in the world doesn't have an extraordinary advantage over the 10th-most-powerful person. ASI could change that.[reference:12]
If people still have civil liberties post-ASI, that will only be because the controllers of ASI allow us to have them.[reference:13] One line of thinking suggests that if everyone had their own personal AI, we could each use it to protect our own interests, and things will turn out okay.[reference:14]
Why the Obvious Approach Fails
The straightforward plan for democratizing ASI doesn't work for two fundamental reasons. First, someone has to build it and then voluntarily give it to everyone. What if they decide not to do that? Second, offense/defense balance issues: if attacking is easier than defending, then bad actors can use their personal ASI to do catastrophic damage.[reference:15]
A Gradualist Alternative
A gradualist approach has some chance of sidestepping the first problem, though it has maybe a 1-10% chance of working.[reference:16] The idea involves making marginal improvements to AI capabilities so that the next-generation AI isn't enough of an improvement to defeat all the previous-generation AIs combined.[reference:17] If the leading AI developer decides to keep the new AI for themselves, they risk a conflict with the rest of the world, and they can't win that conflict.[reference:18]
But even this plan requires international cooperation. Major countries should agree not to build dangerous AI.[reference:19] You still need an international binding agreement not to build AI except under certain narrow conditions, and not to use recursive self-improvement. You still need enforcement mechanisms like transparency and GPU monitoring.[reference:20]
Beyond Human Liberalism
Even if we somehow succeed at preserving personal liberties by democratizing AI, that still doesn't get us a good outcome. Most sentient beings alive today are non-human animals. Today's liberalism has failed to help them; almost all animals live lives of misery. Democratizing AI would not straightforwardly improve on the status quo.[reference:21]
Beings who cannot assert their rights do not automatically benefit; other measures are required to ensure their well-being. Having many ASI agents with different goals could cause tremendous suffering via retributivism, war, or agents threatening torture to incentivize other agents to do what they want.[reference:22]
The Hard Truth
Democratizing AI is not a great plan. But it's less bad than a lot of other plans.[reference:23] For it to be a less bad plan, it has to be implemented with particular care. Otherwise, it's just a bad plan.[reference:24]
What we should really do is pursue international binding agreements. The plan to democratize AI is riskier, but at the same time, in order to preserve the balance of power, you need AI capabilities to advance slowly, and you need some way to enforce that.[reference:25]
The sobering reality is that there is no easy path. Preserving civil liberties in a world with ASI requires difficult choices, international cooperation, and a willingness to confront uncomfortable tradeoffs. The window for action is closing, and the stakes could not be higher.