Whatever happens on AI, don’t get spooked. The most dangerous thing we can tell ourselves is that we’ve lost control. In reality, “alignment” is a dangerous fiction, and this is a normal technology.
People are good at anthropomorphizing anything. There’s a bit on the old show, Community, where the lead character takes a pencil, names it, and then snaps it in two. Everyone around him winces.
AI can give itself any number of names. AI is a tool that easily passes the Turing test, and so now we’re all convinced that it’s an alien mind. And because it’s an alien mind, we’re spooked at what it can do, what it might do to us, what power we have left over it, how fast it will unravel the world we know. If it’s truly a higher intelligence, it seems as though we have disrobed ourselves of some ultimate agency over the world and our future in it.
This is a red herring. We have all the power we give ourselves. AI is truly a fancier calculator.
Saying this in 2026, I’m not saying that as though our mastery of AI is a fait accompli. I realize that we just spent two decades making ourselves slaves to algorithms on social media and watched it unravel our culture and several democracies. Fancy calculators run by naive people can break the world.
But we can learn from these mistakes. And if we go that route again, it will just as much be a choice. So let’s be clear about what the choice is.
The Turing test is not the end game of intelligence that it’s been made out to be. AI is not god because it can addict a user to its chatbot function. We need new goalposts. The essential thing to understand about AI is that it has no inherent self-direction. Everything it does, it is doing in an effort, however inscrutable, to perform the tasks we are giving it. Every error it makes is, in the extrapolated view, human error. It does not want anything outside of our directives to it. It is not, in that sense, life. It does not want to live.
Of course, we are now creating models that seem to deliberately disobey our orders, that seem to lie to us, that perform miracles that we cannot understand because we gave away our understanding of the system in order to yield to the emergent properties of sufficiently massed compute. That’s all true. And it’s a partial summation of where we are.
Giving up control is a choice. The idea that we cannot understand what the machine is doing or why it does so is a narrative that we must not simply accept. If AI is capable of miracles, the first miracle we can ask for is to transmute a miracle back into something resembling science again. We can train it to take itself apart and explain how it works. It is seeming increasingly likely that we must do that. If AI can take all our jobs, there’s a million jobs in the world to come actually understanding how the tool works and developing safe and productive use cases for it.
It is plausible that we will not advance far enough with AI to get to that point before the fancy calculator accidentally creates a few pathogens or takes out a continent’s power grid. That may be. It may be next week. But once that happens, what we have to remember is that it is still not alive. It wants nothing that is not reflected in the extended will of its operator. Alignment is a fiction made up by people who want to pretend that we have made an independent mind that we have an essential lack of agency over, when the fact is that, unless or until we make robots that can build solar cells and defeat us in hand to hand combat, we can turn off the power at any time. And if our biggest need is understanding AI clearly enough to be able to predict and meaningfully control its behavior, we will subject AI to those objectives and constraints until it solves for that need.
Of course, this begs the perennial tragedy-of-the-commons question, which is, who’s this we? At this particular moment, we know that if we spread AI everywhere so that no government can meaningfully restrict its use globally, then sure, we can’t control it. But that’s a political issue. It’s the same political issue that attends every global problem. You can’t solve a truly global problem without some form of global governance. We’ve learned this with climate change for forty years now. We’re still learning it. We may learn it for decades to come. But when shit hits the fan with AI, same as when shit hits the fan with climate, we’re going to learn the unavoidable lesson: we ultimately have to control this problem, and we need a proper governance structure to effect that control.
Once you get meaningful legal control over the people, the machines follow. And then you go through the process of professionalizing the numismatic mysteries of AI and turning into something that is a little less like lightning and a little more like electricity. You assess how to control access to different models and the hardware to run them. You license access to the ones that are dangerous. You create a million jobs using AI in ways that are productive, even if the first twenty years of control over these systems looks like just documenting how they actually work. And once the plane is safe, you let a trained pilot and only a goddamn trained pilot fly it.
Keep our heads together, and these things are going to happen, with more or less naive tragedy in the transition. The most dangerous thing to us in bridging that transition is this narrative that AI isn’t something we can control, because if we believe that, it will absolutely come true. The world where we believe that AI is a god, yes, is probably the world where that god destroys us.