Every few decades, humanity invents something powerful, stares at it for a while, and concludes it’s the end of the world. The printing press would destroy memory. The telephone would destroy conversation. Television would rot our brains. The internet would end privacy, bookstores, and possibly civilization.
Now it’s AI’s turn in the panic barrel. And I get it — this one feels different. It talks. It writes. It codes. It can pass exams and paint pictures and argue philosophy at 2 a.m. When something feels that close to thinking, it’s natural to wonder what happens when it starts thinking for itself.
But here’s my prediction: we’re going to look back on the AI panic the way we look back on the panic over the locomotive — remember when people genuinely believed traveling over 30 miles per hour would suffocate passengers? — and wonder why we were so nervous.
Not because AI doesn’t matter. It matters enormously. But because the fear is built on a confusion about what AI actually is, what consciousness actually is, and how human beings actually handle new technology. Let’s take those one at a time.
We’ve been here before
Roy Amara, the American scientist, nailed it decades ago: we tend to overestimate the effect of a technology in the short run and underestimate it in the long run. That’s Amara’s Law, and it explains every tech panic on record.
In the short run, we imagine the most dramatic version of the future — the robots rise, the jobs vanish overnight, the machines decide we’re obsolete. In the long run, the real transformation happens quietly, looks nothing like the panic, and mostly involves humans adapting the way we always do.
The Luddites weren’t wrong that mechanized looms destroyed their livelihoods. They were wrong that it meant the end of prosperity. The jobs transformed. New categories appeared that nobody in 1811 could have imagined. Living standards rose beyond anything the pre-industrial world could conceive.
Disruption is real. Extinction isn’t. Those are very different predictions, and we keep confusing them.
Nobody can agree on what consciousness is
Here’s the awkward foundation the whole AI doom debate is built on: the people who study consciousness for a living can’t agree on what it is.
Ask a philosopher what consciousness means and you’ll get different answers depending on who you ask. The philosopher Ned Block drew a famous distinction between two completely different things we call “consciousness”: phenomenal consciousness — what it feels like to be you, the raw inner experience of seeing red or tasting coffee — and access consciousness — information being available for reasoning, reporting, and guiding action. A system can have the second without the first, and we’d have no way to tell from the outside.
Thomas Nagel made the point uncomfortably vivid in 1974: we can’t know what it is like to be a bat. Not really. We can study its brain, map its sonar, predict its behavior — but the inner experience, if there is one, is sealed off from us. Now replace the bat with a chatbot. Same problem, except we built the thing and we’re still guessing.
Then there’s David Chalmers’ “hard problem”: why does brain activity feel like anything at all? We can map every neuron firing when you see the color blue, and still have no explanation for why there’s an experience of blueness attached. It’s the deepest unsolved problem in science, and the experts are split into camps — materialists, panpsychists, illusionists — who barely agree on the terms of the debate.
John Searle’s Chinese Room argument put it even more bluntly: a system can manipulate symbols perfectly — producing flawless Chinese responses — without understanding a word of Chinese. Syntax isn’t semantics. Fluency isn’t comprehension.
So when someone tells you with total confidence that AI is “about to become conscious” or “already is,” ask them a simple question: conscious in which sense? They usually can’t answer, because the field can’t answer. We’re building confident doomsday scenarios on top of a concept nobody has managed to define.
What AI actually is right now
Strip away the marketing and the science fiction, and today’s AI systems are prediction machines. They were trained on vast oceans of human text, and they’ve become extraordinarily good at predicting what comes next — so good that the predictions read like thought.
But fluency is not a mind. A team of researchers put it memorably in 2021: these systems are “stochastic parrots,” stitching together linguistic patterns without understanding what the words point to. That’s not an insult — it’s a description of the mechanism. Parrots can recite Shakespeare. It doesn’t mean they’ve read him.
This matters because the doomsday scenarios quietly assume something we have no evidence for: that these systems have goals, desires, or a drive for self-preservation. The classic superintelligence nightmare — the AI that turns the world into paperclips because it was told to make paperclips — assumes an agent with persistent drives pursuing them across contexts. What we actually have are systems trained to predict text, with no inner life we can detect, no desires we can identify, and no evidence of the thing philosophers can’t even define.
Could that change someday? Maybe. “Someday” is doing an enormous amount of work in every AI panic, and it’s worth noticing how quickly the conversation slides from “this system writes good emails” to “this system will inevitably want to survive.” Those are different claims separated by a chasm of missing evidence.
The real risks are boring (and that’s good news)
None of this means AI is harmless. It means the real dangers look nothing like the movies.
The actual risks are human problems wearing a technology costume: bad actors using AI to scam, manipulate, and deceive. Power concentrating in the handful of companies and countries that control the best systems. Deepfakes eroding our shared sense of what’s real. Biases in training data getting laundered into “objective” decisions.
These are serious. But notice what they have in common: they’re problems of people misusing tools, and people have a long track record of eventually building laws, norms, and institutions to handle exactly that. We did it with the printing press (libel law), with the automobile (traffic codes, licensing, insurance), with the internet (still messy, still working on it, but working on it).
Mundane risks have mundane solutions: regulation, liability, transparency, competition. Existential risks have no solutions, which is precisely why they’re such compelling stories. But a compelling story isn’t evidence.
Why everything is going to be okay
Here’s the part the panic merchants leave out: human beings are absurdly good at adapting to new technology. It’s our signature move. Fire, agriculture, writing, electricity, antibiotics, the internet — every single one rearranged civilization, and every single one left us, on balance, better off than we were.
AI will take jobs. It will also create categories of work we can’t yet name — just as the internet destroyed the travel agent and created the app developer, the content moderator, and about ten million livelihoods nobody predicted in 1995. The transition will be bumpy and unfair in places, and we should be honest about that. But “bumpy transition” and “human extinction” are not the same forecast.
And on the other side of the transition is something worth getting excited about: the most powerful problem-solving tool our species has ever built. Cures we couldn’t find. Science we couldn’t do. Creativity amplified for everyone, not just the credentialed few. The printing press democratized knowledge; AI is democratizing capability.
The view from the future
Someday — maybe twenty years from now, maybe fifty — our grandchildren will read about the Great AI Panic of the 2020s the way we read about the panic over the telephone. They’ll find the op-eds, the open letters, the solemn warnings from very serious people, and they’ll feel the way we feel reading Victorian doctors warning that trains would unsettle the uterus or whatever the concern of the week was.
They won’t think we were stupid. They’ll recognize the pattern, because by then they’ll be panicking about something new — some technology we can’t yet imagine, doing something that feels uncomfortably close to magic.
And someone, sitting where I’m sitting now, will write an essay explaining why it’s going to be okay. I hope they get half as much comfort from writing it as I got from writing this.
We’ll be fine. We always are, eventually. The future isn’t the absence of problems — it’s problems we’re capable of handling, met by people who refuse to stop adapting.
Further reading, for the curious:
Thomas Nagel, “What Is It Like to Be a Bat?” (1974) — the classic essay on why inner experience can’t be observed from outside.
John Searle, “Minds, Brains, and Programs” (1980) — the Chinese Room argument: perfect symbol-manipulation without understanding.
Daniel Dennett, Consciousness Explained (1991) — the materialist counter: consciousness as explainable machinery, no magic required.
David Chalmers, “Facing Up to the Problem of Consciousness” (1995) and The Conscious Mind (1996) — the hard problem: why brain activity feels like anything at all.
Ned Block, “On a Confusion about a Function of Consciousness” (1995) — the distinction between phenomenal and access consciousness.
Giulio Tononi’s Integrated Information Theory and Bernard Baars’ Global Workspace Theory (see also Stanislas Dehaene, Consciousness and the Brain, 2014) — the two leading scientific theories of consciousness, which disagree with each other.
Nick Bostrom, Superintelligence (2014) — the book that launched the modern AI doom debate; worth reading to understand what you’re disagreeing with.
Stuart Russell, Human Compatible (2019) — a calmer, more grounded take on AI risk from one of the field’s founders.
Melanie Mitchell, Artificial Intelligence: A Guide for Thinking Humans (2019) — what AI can and can’t actually do, from a working researcher.
Kate Crawford, Atlas of AI (2021) — the human, material, and political costs behind the technology.
Emily M. Bender, Timnit Gebru, and colleagues, “On the Dangers of Stochastic Parrots” (2021) — the paper that named what large language models actually are.
Future of Life Institute, “Pause Giant AI Experiments” open letter (2023) — the most famous artifact of the panic itself; useful as a primary source.
Leave a Reply