Where Do You Make the Cut?

What are you? A person, with agency, or a tool, to be used?

This question is about to become uncomfortably relevant because of the rapid rise of AI—but maybe not for the reasons you think.

I’ve seen some hand-wringing online about our ceding too many of our decisions to automated systems. The concern is that we (and particularly, our kids) are offloading thought to ChatGPT for the sake of convenience and efficiency. And it’s true that it may be time to decide whether you want everything you do to be easy; but there’s a deeper problem.

It’s not just the surrender of decision-making to AI that you have to worry about. The bigger problem is that we may be starting to let systems that we don’t control define what counts as real. —What’s real, what’s a person, and what’s a thing. When that happens, the game is fully rigged: by owning the terms of the argument, the house (the AI and the power structure backing it) wins every time.

To head off this dystopia, you first have to recognize that it’s happening, and then design interventions to prevent it. Here’s how.

We have legislative, legal, and ethical ways to handle situations that we’ve encountered before. If AI just accelerated processes that already exist, all we’d have to do is keep up, or ideally stay one step ahead. That’s not what’s happening. Genuinely new conditions are starting to arise. Our existing systems aren’t able to manage them, not because of the accelerating pace of change, but because they fall outside the categories—the grammar, if you will—of our working models.

I’ve talked about this before, using Alain Badiou’s idea of the Event. You can follow the link to catch up, but here I’ll provide two concrete examples. These Events are examples of situations where our existing legal language, governance frameworks, and ethical intuitions simply stopped working—not so much because they were wrong but because something had occurred that they weren’t designed to handle. Such occurrences are (sometimes) what Badiou calls Events.

An Event isn’t a typical crisis. Crises fit inside our existing categories, even when they’re severe. A crisis is a fire in a building: terrible, but we know what fires are. An Event is something that arrives without the right word attached to it. You reach for your vocabulary and come up empty. You reach for the law and find it wasn’t written for this. The Event doesn’t just challenge your existing answers—you don’t even know what questions to ask.

The proliferation of AI-driven decision-making is an Event in this sense. We have inherited categories—employee, applicant, citizen, person, data—that were built for a world where humans made consequential decisions about other humans. Now those decisions are being made by systems that aren’t “employees” or “people” or even “tools,” and the legal architecture we have wasn’t built for what’s actually happening.

The EU AI Act is, among other things, an acknowledgment of this. It’s a legislative body trying to construct new categories in realtime—to respond to an Event that’s still happening. The Act entered into force in August 2024. It defines four levels of risk; prohibited practices became enforceable in February 2025, and rules for high-risk AI systems embedded in regulated products have a transition period until August 2028. But what it defines is a process for evaluating hazards, not what those specific risks are. It’s not so much legislation as a protocol for dealing with surprises arising because of AI.

Similar in crucial ways is the movement for Indigenous data sovereignty.

Indigenous communities have governed their relationship to land, species, and knowledge through frameworks developed over centuries—frameworks that understood data about the salmon, or about the language, or about time, as inseparable from the community that held it. Then Western science arrives with its instruments, gathers the data, puts it in a database, trains a model on it, and produces outputs that claim authority over the very systems those communities have managed for generations. The category it uses is “data”—neutral and ownerless, exemplifying the famous “view from nowhere” of objective science. The category the communities use, on the other hand, is more like “relationship,” which cannot be extracted at all, only participated in.

These two frameworks don’t just disagree. They can’t even argue properly, because they’re not speaking the same language. I’m kind of mangling Badiou here, and I apologize, but you could say that each side of this dispute experiences it as an Event: the moment when “this is our data and you need our permission” meets “data is by definition open to measurement” and neither side has the vocabulary to negotiate with the other. Institutions that rely on the view from nowhere, such as UNDRIP, copyright law, and even research ethics boards—were not built for this collision. New instruments are being invented on the fly, and none of them fully work yet.

These might seem distant from your own concerns, but remember that the World Economic Forum is claiming that AI will take 92 million jobs by 2030. (They also claim that it’ll create 78 million, but that’s still a net loss.) Whether you’re white collar or blue collar, automation may be coming for your job, whether on the desktop or the worksite.

Is it the efficiency of corporate performance that matters, or the welfare of the workers? Is it the availability of human knowledge in the form of helpful chatbots, or human critical thinking and investigative skill? Job loss may be the crisis, but the Event is that all these occurrences involve decisions about what counts: what is to be considered an agent in a given situation, and what counts as the thing that an agent acts on. Each AI-related crisis hinges on where we make that cut when we create a new grammar to describe a new situation. In other words, who gets to decide what’s real when we respond to an Event?

Share

Okay, so we can look past crises to see whether they land firmly in the realm of what we know how to deal with, or are Events. What do we do with that knowledge?

The answer is that we learn how things come to matter—how the cut is made that separates agent from acted-upon. Then, hopefully, we can ensure that the cut’s made in our favour.

To start with, what do I mean by making the cut?