← Frontier & Future AI

⏱ About 15 min✦ 15 XP

What Is AGI?

You have probably heard the term artificial general intelligence — or AGI — used in news articles, corporate announcements, and arguments about the future. It is one of the most discussed and most contested concepts in AI. Some researchers say AGI is decades away. Others say we are already approaching it. Some argue the whole concept is poorly defined and misleading. To think clearly about the future of AI, you need to understand what AGI means, why it is genuinely hard to pin down, and what the real disagreements are about.

Narrow AI versus General AI

Today's most capable AI systems are what researchers call narrow AI: they are extraordinarily good at one category of task, but they do not transfer that skill spontaneously to an unrelated domain. A world-champion Go-playing AI cannot read an email. An AI that generates photorealistic images cannot debug code unless it was specifically trained to do so. A language model that writes beautiful prose about quantum physics may fail at a spatial puzzle a ten-year-old would solve easily. Each of these systems is powerful within its domain and limited outside it. Artificial general intelligence, by contrast, refers to a system that can perform well across a wide range of intellectually demanding tasks — tasks that span different domains, require different types of reasoning, and involve adapting to genuinely new situations.

The Core Definition

AGI is commonly defined as an AI system capable of performing, at a level comparable to a competent adult human, the full range of cognitive tasks that humans can perform — including learning new skills from minimal examples and transferring knowledge between domains.

Why the Definition Is Contested

The definition above sounds reasonable until you try to make it precise. What counts as 'human level'? Humans vary enormously in ability. Which humans? And which tasks? Does AGI need to match the average human, the best human, or something in between? Some researchers define AGI economically: a system capable of automating a large fraction of cognitively demanding work that humans currently do for pay. Others define it cognitively: a system that can learn any task a human can learn, given comparable time and experience. Others define it by test: a system that passes a broad suite of standardized benchmarks at human-level performance. Each definition leads to a different answer about whether we are close to AGI, far away, or already past some versions of it.

What AGI Is Not

AGI is often confused with two other concepts. First, superintelligence: a system that is not just human-level across tasks but dramatically smarter than any human at everything. AGI as typically defined does not require superintelligence — just broad human-level competence. Second, consciousness or sentience: AGI says nothing about whether a system has subjective experience, feelings, or awareness. A system could theoretically be a competent general reasoner without having any inner life whatsoever. Consciousness is a separate question — and a much harder one.

AGI Is Not the Same as Superintelligence

Superintelligence refers to a system smarter than the best humans at every cognitive task. AGI is a lower bar: human-level breadth of capability. Many researchers are concerned about superintelligence as an eventual follow-on to AGI — but they are distinct concepts.

Expert Disagreement

Surveys of AI researchers show a wide range of views. Some believe AGI could arrive within a few years, as scaling continues to unlock new capabilities. Others believe the current paradigm — large language models trained by gradient descent on text — has fundamental limits that cannot be overcome by scaling alone, and that we would need major new scientific breakthroughs to achieve general intelligence. Still others argue that what we currently call AGI is a misleading goal: intelligence is not a single unified thing, and building a system that is 'generally capable' across arbitrary tasks may not be the most useful or meaningful target for AI research. You are learning about AI at a moment when this debate is live and unresolved. The smartest people in the field genuinely disagree, and the answer matters enormously for society.

Match each AGI-related concept to the correct description.

Terms

Narrow AI
AGI
Superintelligence
Domain transfer
Cognitive benchmark

Definitions

A system dramatically smarter than the best humans at every cognitive task — a higher bar than AGI
A standardized test used to evaluate whether an AI system reaches human-level performance on a specific task
The ability to apply knowledge or skills learned in one area to solve problems in a different area
A system that excels at one category of task but cannot transfer skills spontaneously to unrelated domains
A hypothetical system capable of performing the broad range of cognitive tasks a competent adult human can perform

Drag terms onto their definitions, or click a term then click a definition to match.

What is the key difference between today's most capable AI systems and the concept of AGI?

Why is it difficult to agree on a single, precise definition of AGI?

Define AGI for Yourself

  1. Step 1: Write your own one-sentence definition of AGI. Try to be as precise as possible — avoid vague phrases like 'as smart as humans' without explaining what that means.
  2. Step 2: Design a test with three specific tasks that, if an AI passed them all, you would personally consider it AGI by your definition.
  3. Step 3: Apply your test to current AI systems you have used or heard about. Do any pass all three tasks?
  4. Step 4: Find one weakness in your own definition. What kind of system could pass your test while still being clearly not general?
  5. Step 5: Revise your definition to fix that weakness. Write the improved version.
← Back to Frontier & Future AI