The Case for Deleting 'AI' From Our Vocabulary

The short version

A debate is swirling around whether we should even use the term “Artificial Intelligence.” The core argument, championed by thinkers like Jaron Lanier, is that the name itself is a misleading label that creates confusion, hype, and fear. Using more accurate, if boring, terms would help us all have more productive conversations about what this technology actually is.

What sparked this whole conversation?

I’ll admit it: I have AI news fatigue. Every single day brings a new model, a new demo, a new claim that changes everything. It’s a firehose of information that’s impossible to keep up with, and most of it feels designed to generate hype or terror. So when I stumbled across a blog post by computer scientist Philip Wadler titled simply “There is no AI,” it stopped me in my tracks. What a statement.

The post itself is short, mostly quoting the influential computer scientist and VR pioneer Jaron Lanier. Lanier has been arguing for years that the term “AI” is a philosophical trap. Wadler’s post brought this argument to the forefront again, and a recent Hacker News discussion about it absolutely exploded. The thread was a fascinating mix of engineers debating semantics, philosophers arguing about consciousness, and regular users just trying to make sense of it all. It felt like a release valve for a tension that’s been building for a while.

We’ve all been using this one, two-letter acronym to describe everything from the algorithm that recommends shows on Netflix to the large language models that can write poetry. But what if the term itself is the problem? What if it’s actively hurting our ability to understand what we’re building and using?

So what’s the big deal with the name ‘AI’?

The problem isn’t the “artificial” part. It’s the “intelligence.” That single word carries so much weight. It implies a mind. It suggests consciousness, intention, and a subjective experience of the world. When we say something is intelligent, we instinctively map our own human experience onto it. We think of a being that understands, reasons, and feels.

Large language models do not do that. They are not thinking. They are not sentient beings pondering the nature of existence. They are extraordinarily complex statistical machines. That’s it. They are trained on unfathomable amounts of text and image data created by humans, and they learn the patterns and relationships between words, phrases, and pixels. When you ask ChatGPT a question, it is not understanding your query. It is calculating, on a mind-boggling scale, a sequence of words that is statistically likely to follow the words you gave it.

It’s a form of very sophisticated autocomplete. A prediction engine. Calling it “intelligence” is like calling a calculator an “Artificial Mathematician.” A calculator is incredibly fast and perfectly accurate at arithmetic, far better than any human. But it has zero understanding of what the number ‘7’ actually means. It’s just executing a programmed function. LLMs are a far more advanced version of this principle, operating on language instead of just numbers. The output can look like understanding. It can even look like creativity or empathy. But it’s an illusion—a masterful echo of the human intelligence embedded in its training data.

But don’t these systems seem intelligent?

Oh, absolutely. I get it. I use these tools all day, every day. I use them to brainstorm ideas, write code, summarize dense articles, and draft emails. The output is often so good it feels like magic. It’s easy, in the flow of a conversation with a chatbot, to forget you’re not talking to something that gets you. The performance is incredible. The illusion is powerful.

But then the illusion shatters. It always does.

Just last week, I was experimenting with an image generator. I gave it what I thought was a simple, concrete prompt: “A close-up photo of a human hand holding exactly four blue marbles.” It’s a straightforward instruction. It gave me five images back. One image showed a hand with seven fingers, clutching a pile of what looked like blue pebbles. Another had a perfectly normal hand, but it was holding only three marbles. A third showed four marbles floating vaguely near a hand that seemed to be melting into the background. None of them got it right.

A five-year-old child would understand that request instantly. The model does not understand the concept of a “hand” or the quantity “four.” It just knows the statistical association of pixels that tend to appear when those words are in a prompt. It’s a system for generating plausible imagery, not a system for reasoning about the world. When it fails, it fails in ways that are profoundly inhuman, revealing the statistical wiring under the hood. It’s not thinking; it’s pattern-matching on a planetary scale.

What should we call it instead?

This is the hard part. If we ditch “AI,” what do we replace it with? Jaron Lanier’s preferred term is “applied statistics,” which is technically accurate but probably too academic to ever catch on in the mainstream. It lacks sizzle.

The Hacker News thread offered a bunch of other alternatives. “Computational statistics.” “Complex algorithms.” “Automated pattern-matching systems.” They all suffer from the same problem. They are descriptive and accurate, but they’re also a mouthful.

Personally, I think the best approach is the simplest one: be specific. Stop using a catch-all term and just say what the tool is. I’m not using “AI” to help write this article. I’m using a “large language model.” I’m not using “AI” to create a header image. I’m using a “diffusion model” or an “image generator.” When I’m talking about my Roomba, I’ll call it a robot vacuum, not an AI-powered cleaning agent.

This act of being specific does something important. It demystifies the technology. It grounds it. It takes it out of the realm of science fiction and places it firmly in the category of “tool.” It’s less sexy, and that is precisely the point. It’s about clarity over marketing.

Why does changing the name actually matter?

This isn’t just an argument about words for the sake of it. The language we use has massive, real-world consequences. It shapes our thinking, our investments, our regulations, and our fears. The “AI” framing is dangerous because it encourages magical thinking and pushes us toward two unproductive extremes: utopianism and doom.

The utopian view sees AGI—Artificial General Intelligence—as a near-future savior that will solve climate change, cure disease, and usher in an age of abundance. The doomer view sees it as Skynet-in-waiting, an existential threat that will inevitably escape our control and destroy us. Both of these narratives, fueled by the anthropomorphic idea of “intelligence,” distract us from the very real, very urgent, and very practical problems we need to address right now.

If we reframe the technology, the questions change. When a company announces a new “AI hiring tool,” the public might imagine a wise, objective robot judge. If they instead called it a “resume pattern-matcher trained on our last 10 years of hiring data,” the questions become much more pointed. Questions like: “Does your past hiring data reflect a bias toward a certain gender or race?” Or, “Is this system just going to perpetuate the same homogenous culture you already have?” Suddenly, we’re having a useful conversation about bias, fairness, and labor, not a sci-fi debate.

When we talk about LLMs as “applied statistics” instead of “intelligence,” we start asking better questions. Instead of “Will it become conscious?” we ask, “Whose data was used to train it, and were they compensated for their labor?” Instead of “Could it take over the world?” we ask, “How can we mitigate the risk of this tool generating harmful misinformation at scale?” These are the boring, practical conversations we need to be having.

What’s my take on all this?

I’m completely on board. I’m making a conscious effort to scrub the generic “AI” acronym from my vocabulary and be more specific. It’s difficult. The term is a reflex, a convenient shorthand baked into every headline and product announcement. But forcing myself to be more precise has been a clarifying exercise.

It helps me think more clearly about what these tools can and cannot do. It serves as a constant reminder that these are not my colleagues. They are not collaborators. They are incredibly powerful, deeply strange, and often flawed instruments.

Thinking of them as complex statistical systems rather than nascent minds puts the responsibility back where it belongs: on us. The people who build them, train them, and choose how to use them. They are not alien intelligences arriving from the ether. They are mirrors, reflecting the vast, messy, brilliant, and biased dataset of human culture we fed them. And grappling with that reflection is a much more interesting and urgent task than waiting for the robots to wake up.

FAQ

Is ‘AI’ really that bad of a term? It’s a useful shorthand, but it comes with a lot of philosophical baggage. The word “intelligence” implies a mind or consciousness, which can mislead people about what the technology actually does and lead to unnecessary hype and fear.

Who is Jaron Lanier? Jaron Lanier is a computer scientist, musician, and writer who is often called the “father of virtual reality.” He’s known for his critical and humanistic perspective on technology and has been warning about the misleading nature of the term “AI” for many years.

What’s a better term for AI? There’s no single perfect replacement. Suggestions include “applied statistics,” “machine learning,” or, most practically, just describing the specific tool you’re using, like “large language model” or “image generator.” The goal is accuracy over buzz.

Will people actually stop calling it AI? Probably not anytime soon. The term “AI” is deeply embedded in our language and culture. However, challenging the term is a valuable exercise that helps us all think more critically about these powerful tools and their real-world impact.