Reality Bats Last (Part 1 of 2)

Baudrillard’s prophecy, Habermas’s last problem, and why AI might be the worst thing that ever happened to constructivism

“The world is not ready for what’s coming this year... reality doesn’t care what you think.”

I saw a lot of sentiment like this in the last week. There’s something bracing about all of it—a kind of epistemic hardball that cuts through the noise. This tweet-length idea is not saying we’re unprepared in the usual sense (bad policy, weak institutions, insufficient leadership); instead, it’s saying reality operates on its own logic, indifferent to our readiness, our theories, our interpretations, our sense of control.

It’s a realist claim. Maybe the most fundamental one.

And here’s the thing that won’t leave me alone I guess: that claim is being validated right now, not by some dusty philosophy seminar, but by the technology that should have murdered realism for good. AI generates hallucinated citations, deepfakes that fool facial recognition, text produced by systems that understand nothing. It looks like the apotheosis of Baudrillard’s prediction—a world where simulation has so thoroughly replaced reality that maps generate their own territories, copies float free of originals, and the distinction between genuine and constructed has collapsed into irrelevance.

The other very uncomfortable part is how much of this is just unfolding. Rob Reid’s After On captured it back in 2017—the first-to-AGI-wins logic, the scale of what happens when power asymmetries open up before institutions notice. The novel was informed by people doing actual AI work, and it aged better than most futurism does. Reid saw the curve early. And Jürgen Habermas died this month at 96—the philosopher who spent his career trying to hold together rationality and critique, who believed the better argument could still win. I’ll get to why that matters below, but the timing feels like it means something.

But the technology that most looks like the triumph of the simulacrum is actually an empirical vindication of realism. Reality is winning, and the institutions built on denying that fact are scrambling.

The Prophecy Fulfilled

Jean Baudrillard predicted a world of copies without originals. Simulacra and Simulation (1981) painted a picture: simulations that precede reality, maps that generate their own territories, hyperreality where the distinction between authentic and constructed has become operationally meaningless. We live in a hall of mirrors, he said. Signs relate only to other signs. The signified has evaporated.

AI looks like proof he was right. (Several scholars have noticed this—Al Mashai’s “The Postmodern Crisis, Metamodernism, and the Emergence of a Fragmented World Order” is probably the sharpest.)

Generative models produce images that never existed. Text-to-speech systems generate voices without speakers. LLMs hallucinate citations, confabulate sources, synthesize plausible but fabricated arguments. The prophecy fulfilled. Baudrillard got to watch his vision instantiated as operational technology—simulacra became software. The map precedes the territory. The copy has entirely unmoored itself from the original.

But there’s a problem with the prophecy. Actually, with the whole philosophical tradition that descended from it.

The Refutation Hidden Inside the Prophecy

If Baudrillard is right—if we’re living in a world where simulation has wholly replaced reality, where there are no originals to fall back on, where everything is construction layering on construction—then why does anyone care about hallucinations?

Why would an AI error matter?

You can only hallucinate if there’s a reality to hallucinate about. The moment someone says an LLM “made something up,” they’re admitting something: there are facts that exist independently of the system’s construction. There are things that are true that the model got wrong. Reality as external standard. Reality as constraint.

Listen closely to the critics: they say an LLM fabricates sources that don’t exist. They say training data are biased—meaning it diverges from some objective fact about the world. They say the system is unreliable because it fails to track what’s actually the case. Notice the structure underneath? All of it presupposes that there’s a reality out there being violated. A constructivist can’t actually say that. If everything is construction, then “bias” is just another construction, and “accuracy” is just a preference. But nobody actually talks that way, because nobody really believes it.

The very vocabulary we use to criticize AI—hallucination, confabulation, misinformation, error—presupposes that there’s a reality being violated. (There’s a sharp piece in AI & SOCIETY called “Between Fact and Fairy” that traces how the hallucination metaphor itself smuggles realism back into the conversation.) The stronger the constructivist critique of AI, the more it admits that reality exists and constrains what counts as knowledge.

This is where it gets uncomfortable for a whole tradition. What should have been simulation’s final victory is actually the return of the repressed—the return of the empirical, the material, the stubborn fact that minds can’t construct away.

AlphaFold predicted protein structures that no human had ever determined. When we synthesized those proteins, they folded exactly as the model predicted. Atmospheric physics doesn’t care about social construction. DeepMind’s GenCast weather model beats traditional physics-based forecasts because it learned something real about how air moves. Reality pushes back. The model works or it doesn’t work.

Think about the paradox: in a genuinely constructed world—a world where everything is interpretation, and interpretation is all there is—there would be no such thing as “epistemic collapse” worth worrying about. You can’t “lose your grip on reality” if reality is just another construction. But everyone treats epistemic collapse as a genuine threat. Which means everyone believes, on some level, that there’s a reality to lose your grip on.

The strongest case for realism isn’t coming from analytical philosophers in seminar rooms. It’s coming from the engineering labs. AI systems trained on patterns in data—without lived experience, without social positioning, without any of the apparatus constructivists said was necessary for knowledge—produce outputs that reliably correspond to how things actually are. The proteins fold. The weather predictions land. The arguments survive scrutiny. Something is doing the work, and it isn’t standpoint.

The Hostility Has Roots

This didn’t arrive out of nowhere. The intensity of the reaction to AI in academia can’t be explained by legitimate pedagogical concerns or even by anxiety about technology. It’s the latest flash point in something deeper—a fifty-year tension between two incompatible views of how knowledge actually works.

One side says: reality has structure independent of our minds. Formalization captures something genuine about that structure. A system can learn to map reality without being socially positioned, without lived experience, without political reflexivity. Objectivity is possible.

The other side says: knowledge is always constructed through social positioning. Claims to neutrality mask power. Quantification and formalization are tools of domination disguised as objectivity. Be suspicious.

Both of these positions have roots. The first line descends from analytic philosophy, empiricism, and the scientific tradition. The second comes out of mid-20th century continental thought—Heidegger’s “Question Concerning Technology” (1954), Horkheimer and Adorno’s Dialectic of Enlightenment (1947), postmodern skepticism toward grand narratives. By the 1980s and 1990s, the second view had become institutional orthodoxy in humanities departments. Not presented as one perspective among others, but as the enlightened position. To claim that data might tell you something true about the world became naive, even complicit.

The thing is: AI forces a reckoning. It’s not that the critics are stupid—I’ve written about this before—the resistance is motivated reasoning protecting professional legitimacy, and it’s rational given the stakes. But a technology has arrived that empirically challenges the epistemological foundation these institutions are built on. And institutions don’t usually handle that gracefully.

The Last Refuge

Back to Habermas. He spent his career defending a position that was supposed to split the difference—a way to take seriously the critique of instrumental rationality while still believing that objectivity and shared understanding were possible. Communicative rationality, he called it in The Theory of Communicative Action (1981). Democratic legitimacy rests on validity claims that participants can in principle adjudicate together. Truth, rightness, sincerity. The better argument can win.

That was the appeal. You could embrace critical theory, understand how power shapes knowledge, and still think evidence mattered. Habermas was the constructivist who refused to abandon the possibility of truth.

But something strange is happening: an LLM can generate arguments that survive deliberative scrutiny. Produce validity claims you’d have a hard time dismissing in a seminar. The system has no genuine orientation toward understanding. It isn’t vulnerable to counterargument. It has no stake in the outcome. Yet the arguments are good.

Habermas’s framework was normative—it was about what democratic legitimacy requires, not about what can produce a valid-sounding sentence. He knew that. But he built the normative case on an empirical assumption: that producing good arguments is hard, that it requires the discipline of mutual accountability, that the deliberative process does the filtering. What happens when argument production becomes cheap? When systems with no communicative orientation can flood the deliberative space with validity-passing claims? The sorting mechanism Habermas relied on—the sheer difficulty of constructing the better argument—starts to degrade.

That’s not a refutation of communicative rationality. It’s a problem Habermas never had to face. And it’s a problem that only makes sense if the arguments are tracking something real—if there’s a structure underneath that the system is latching onto, independent of anyone’s orientation toward understanding.

What Happens Next

Here’s what I don’t know: whether the institutions can rebuild on honest ground.

There was a sensible position, back when all this started. The 1960s and 1970s version. Power shapes knowledge. Perspectives matter. Formalization has limits. Democracy depends on deliberation that’s accountable to the people affected. All of that was right.

What went wrong was the mutation into certainty. The claim that objectivity itself is a myth. The insistence that all knowledge is construction. The position that formalization is inherently dominating. These became institutional doctrine, enforced through hiring and publishing and language. But they’re stronger claims than evidence supports. And AI is now challenging them in ways that get harder to ignore.

The uncomfortable truth is that AI doesn’t resolve this philosophically. A smart pragmatist can look at AlphaFold and say “it found functional mappings within our epistemic categories” rather than “it discovered the structure of reality.” Both explanations predict the same experimental results. The philosophical question—whether successful prediction requires correspondence to a mind-independent world or just reliable patterns within our frameworks—isn’t the kind of thing a technology settles.

What AI does settle is the institutional question. For fifty years, the default posture in much of the academy has been that objectivity claims are suspect and formalization is inherently limited. AI makes that default untenable. Not because it proves metaphysical realism with a capital R—but because it shifts the burden of proof. The people claiming that knowledge requires social positioning now have to explain why a system without any positioning keeps generating knowledge that works. The argument is practical, not philosophical. A reckoning, not a proof.

And the practical reckoning is what matters for institutions. Reality has structure that constrains what counts as knowledge, whether or not you’re socially positioned to see it. But our access to that reality is partial, shaped by who we are and where we stand. Objective patterns exist and our access to those patterns is shaped by what we choose to attend to. Formalization captures something real and power shapes what gets formalized and who gets to do the formalizing. Luciano Floridi’s informational realism gets closest to this landing—reality has structure, our models approximate it, and the approximation is improvable. Call it what you want — I don’t think it’s constructivism, and it’s definitely not naive realism. It’s what the evidence actually supports.

I’ve called this metamodern before, and I keep coming back to it—but I want to be honest about what that actually means. Vermeulen and van den Akker’s original framing wasn’t a tidy synthesis. It was an oscillation. You swing between the modernist commitment to truth and the postmodern awareness that your access to truth is partial, situated, shaped by power. The swinging itself is the honest position. You don’t get to land.

So when I say “reality bats last,” I’m not claiming the oscillation is over. I’m saying AI has shifted its center of gravity. The realist swing has gotten harder to resist. The institutions that built fifty years of programming on the other swing—the one that says objectivity is a myth and formalization is suspect—are the ones feeling the G-forces right now. You still swing. But you swing from realism now, not toward it as an act of rebellion.

The question isn’t whether constructivism or realism wins. It’s whether institutions can be honest about having overcorrected—and whether they can acknowledge that without swinging all the way back.

That’s a harder conversation than it looks.

Next Time

There’s a fifty-year backstory to all this—the actual genealogy of how we got to a place where defending the reality of objective patterns sounds like heresy in half the academy. Understanding that history explains why the current panic feels so disproportionate. It also clarifies what’s actually at stake in the reconstruction work ahead.

Reality does indeed bat last. The institutions built on denying that are going to have to figure out what they’re for now.