What Is "Intelligence", Really?
A word we thought was clear—but never actually defined. The next time you debate "is AI really intelligent," read this first.
If you’ve spent any time online discussing AI, you’ve seen people argue “GPT-4 isn’t real intelligence” and others claim “AI has surpassed humans.” What are they actually arguing about?
They’re arguing about what “intelligence” means—and there has never been a definition humans agreed on. Let’s look at the major definitions; each gives a different verdict on whether AI is “intelligent.”
Definition 1: Pass the Turing Test = Intelligent
The oldest standard, proposed by Turing in 1950. If a machine can fool humans in conversation, making them unable to tell whether they’re talking to a human or a machine—then it has intelligence.
By this standard: ChatGPT passed long ago.
But this standard is increasingly questioned—it tests “the ability to act like a human,” not “the ability to think.”
Definition 2: Solve Problems = Intelligent
The engineer’s view: intelligence is problem-solving capability. Beating Go masters, recognizing images, translating text, writing code—all count.
By this standard: AI has surpassed humans in many narrow domains.
But critics say: this is just “tool strength,” not “mind.” A calculator computes 1+1 lightning fast, but we don’t call it smart.
Definition 3: Learn and Adapt to New Environments = Intelligent
The cognitive science view. Emphasis on “new environments”—doing well on what you’ve seen is unimpressive; handling things you haven’t seen is intelligence.
By this standard: humans dwarf AI. A 3-year-old, seeing an unfamiliar animal, immediately judges “is this animal? dangerous? playable?” AI’s abilities outside training data often collapse.
Definition 4: Has Consciousness = Intelligent
The philosophical view. Intelligence must be accompanied by “subjective experience”—no matter how capable a machine is, if it’s “dark inside” with no feelings, it isn’t intelligent.
By this standard: we can’t prove AI has or doesn’t have consciousness. This debate may rage for centuries.
“What is it like to be a bat?” Even if we fully understand bat neurology, we don’t know what being a bat feels like—the hard problem of consciousness. AI faces the same.
Definition 5: Generality = True Intelligence (AGI)
The current AI field’s pragmatic standard: AGI (Artificial General Intelligence).
Roughly: “achieves human expert performance on most economically valuable tasks”—OpenAI’s working definition.
By this standard:
- Today’s AI is “narrow AI”
- AGI is “general AI”
- ASI is “superintelligence”
Many people’s real question isn’t “is AI intelligent” but “how far is AGI?” Optimists say 2027-2030; pessimists say 50 years; realists say “we don’t even agree on the standard, how can we judge arrival?”
An Interesting View: Intelligence Is a “Spectrum”
Some recent AI researchers (Yann LeCun, Demis Hassabis) lean toward:
Intelligence isn’t on/off. It’s a multi-dimensional spectrum.
You can evaluate a system on these dimensions:
| Dimension | Description | Current AI |
|---|---|---|
| Perception | See, hear, read | Strong |
| Language | Understand and generate | Strong |
| Reasoning | Multi-step logic, causality | Medium, improving fast |
| Planning | Long-term goals, sub-task decomposition | Weak → Medium |
| Learning | Adapt from experience | Strong in training, weak in inference |
| Common sense | Basic physical world knowledge | Medium |
| Creativity | Truly “new” things | TBD |
| Consciousness | Subjective experience | Unknown |
| Emotion | Real feelings | Unknown (can simulate) |
By this map, AI doesn’t “think” or “not think”—it surpasses you on some dimensions and lags far on others.
Common Misconceptions
Misconception 1: AI isn’t real intelligence, it just “predicts next words”
This argument may apply to human brains too. When you write a sentence, the next word also comes from your “model” with some probability—just that your model is billions of years of evolved neurons, vs an AI’s months of artificial neurons.
Who can rigorously prove they’re fundamentally different?
Misconception 2: AI lacks consciousness, so it can’t be intelligent
Consciousness and intelligence may be separate things. A conscious infant isn’t more “intelligent” than an unconscious calculator. Conversely, an unconscious system might exhibit intelligent behavior.
Misconception 3: When AI passes XX test it’ll be intelligent
Throughout history, every time AI breaks a benchmark (Turing test, Go, ImageNet, math olympiad), someone says “well that’s not real intelligence.”
This is called “AI Effect”: once machines can do something, humans immediately remove it from the “intelligent” list.
By this logic, AI can never be “really intelligent”—not because it can’t, but because the goalpost keeps moving.
One-line Summary
“Intelligence” is a word we haven’t defined for ourselves, so “is AI intelligent” is essentially a language game.
The meaningful question is: what can it do, what can’t it do, what will it look like in 5 years.
That’s what the L0 path is for.
If an AI perfectly simulated a human—could chat, problem-solve, create, even “fake sadness”—but you knew it had “nothing inside.” Would you say it’s intelligent?
Next: “Your First AI Chat in 5 Minutes” — theory’s done, time for hands-on.