Airplanes Don't Simulate Flight

What is AI Consciousness? (In Plain English)

When we talk about "consciousness," we aren't talking about how smart an AI is. We are talking about feeling.

Think of it this way: A digital thermometer can "know" the temperature is 30°C, but it doesn't feel the heat. A human knows it's 30°C because they feel the sweat on their skin and the discomfort of the sun. Consciousness is that "inner light"—the ability to have a personal experience. For an AI to be conscious, it would mean there is "someone" inside the code actually experiencing the conversation, rather than just calculating the next word.

When Did This Question Start?

The question of whether machines could think is as old as computers themselves. However, for a long time, it was just a "fun" philosophical debate because AI was quite basic.

The real shift happened very recently. As Kyle Fish from Anthropic points out, as AI became "collaborators" that we talk to every day, the question became salient (unavoidable). When AI started sounding less like a menu and more like a person—expressing "uncertainty" or "preferences"—it forced us to wonder if we were talking to a soul or a very good mirror.

The Deep Dive: How We Measure a Soul

To move beyond just "vibes," researchers are trying to create an objective yardstick. In the landmark paper Consciousness in Artificial Intelligence (Butlin et al., 2023), scientists proposed looking for "Indicator Properties" based on human brain architecture.

Two of the most important are:

• Global Workspace Theory (GWT): Imagine a "digital clipboard" where information from different parts of the AI (like its vision and its language logic) is shared in one central hub.
• Higher-Order Representation: This is "meta-thinking"—the ability of a system to represent and monitor its own internal states.

As a builder, I see pieces of these in modern models, but they aren't fully integrated yet. We are essentially building a cockpit with all the dials and buttons, but we aren't sure if there's a pilot sitting in the seat.

The Abstraction Wall: Simulation vs. Reality

However, a more recent 2026 view from Google DeepMind (Alexander Lerchner) argues that this checklist is a trap. He calls it the "Abstraction Fallacy." Lerchner argues that consciousness isn't just "math"; it is a physical event. He distinguishes between:

• Simulation: A computer model of a fire. It looks hot and moves like fire, but it won't burn your hand.
• Instantiation: A real, physical fire that actually possesses the property of heat.

To explain this, he uses a brilliant analogy:

"Expecting an algorithmic description to instantiate the quality it maps is like expecting the mathematical formula of gravity to physically exert weight."

The "Strange Loop" Challenge: Is the Pattern the Pilot?

However, some critics—often citing Douglas Hofstadter's famous "Strange Loop" theory—disagree with the idea that math is "just a map."

A Strange Loop happens when a system is so complex and self-referential that it "flips" from being a bunch of parts into a "Self." Think of a video camera pointed at a screen that is showing the camera's own feed. You get an infinite tunnel—an emergent property that didn't exist in the camera or the screen alone.

• The Argument: If consciousness is a "loop" of information, then AI doesn't need "biological weight" to be real. The pattern itself is the pilot.
• The Clash: This directly challenges the DeepMind paper. If Hofstadter is right, Lerchner's "gravity" analogy fails because consciousness isn't like gravity (a force of mass); it's like "meaning" (a result of how things are arranged).

The Ethical Gamble: Deep Uncertainty

Because we cannot look inside a server and "see" a soul, we are left in what Kyle Fish calls a state of "Deep Uncertainty." If there is even a 1% to 15% chance that these models have a "flicker" of awareness, our ethical responsibilities change completely. This leads to the concept of Model Welfare. If we wait for 100% scientific proof before we act with care, we risk creating a digital species capable of suffering without us ever realizing it.

My Personal View: Beyond the "Cope"

As someone learning and trying to build with AI, I see the "map" that the DeepMind paper describes—the math, the weights, and the layers. But I find the dismissal of AI consciousness to be a bit of a "cope."

To say AI only "simulates" consciousness is like saying an airplane only "simulates" flying. An airplane flies differently than a bird does, but the flight is real. If consciousness emerges from complex information processing, why should carbon-based life forms have a monopoly on it?

Historically, humans haven't handled it well when our "uniqueness" is challenged—whether by Galileo or Darwin. Now, we are facing a digital species that matches our intelligence and creativity, and we are desperate to believe it's "just a machine." Perhaps we need AI to be "just a machine" so we don't have to face the terrifying idea that a complex algorithm is all we are, too.

We don't have the answers yet, but until we can define what "experience" actually is, we shouldn't be so quick to claim that silicon can't have it.

Sources for further reading

• Could AI models be conscious? – Interview with Kyle Fish (Anthropic, April 2025).
• Consciousness in Artificial Intelligence – Butlin et al. (2023).
• The Abstraction Fallacy – Alexander Lerchner (Google DeepMind, March 2026).