Narrow, General, AGI, ASI: What We Mean and Why It Matters

The terminology ladder of AI capability is loaded. Clarify your definitions and you clarify your whole view of the field.

40 min · Reviewed 2026

Why Definitions Are Political

If a CEO says their product is AGI, they are simultaneously making a scientific claim, a marketing claim, and often a trigger for contractual clauses with investors. Definitions are not neutral. Pin yours down before you use the words.

The classical ladder

Level	Working definition
Narrow AI (ANI)	Performs a specific task, sometimes superhumanly
General AI (AGI)	Matches human-level performance across most cognitive tasks
Artificial Superintelligence (ASI)	Vastly exceeds human performance across all domains

Google DeepMind's Levels of AGI

A 2023 paper by Morris et al. at DeepMind proposed a two-axis framework: performance (from emerging to superhuman) and generality (narrow vs general). This escapes the binary trap and lets you ask, how general, how competent, and at what percentile of humans.

Level 0: No AI
Level 1: Emerging (equal to or better than unskilled human)
Level 2: Competent (50th percentile of skilled adults)
Level 3: Expert (90th percentile)
Level 4: Virtuoso (99th percentile)
Level 5: Superhuman (outperforms 100 percent)

Why the AGI term is contested

No agreed benchmark means anyone can claim it
The label can trigger contracts and IPO milestones
Different stakeholders prefer different goalposts
Anthropomorphizing leads to bad policy decisions
Emergent capabilities blur thresholds further

Alternative framings worth knowing

Framing	Core concept
Transformative AI	AI that triggers changes of industrial revolution scale
Human-level AI	Performs any cognitive task humans can do
Automated AI researcher	Can itself do the work of an AI researcher (recursive)
AGI as economic test	Can do 50 percent of remote work as cheaply as a human

Implications for your thinking

Whenever you see AGI, ask whose definition
Benchmarks should be paired with generality claims
Short timelines and long timelines both have credible defenders
Plan products for a range of outcomes, not a point estimate

AGI is the horizon — it keeps receding as you approach it.
— An AI researcher

The big idea: intelligence is multidimensional, and the AGI label collapses too much nuance. Use multi-axis frameworks and be explicit about what you are measuring.

End-of-lesson check

15 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-creators-narrow-general-agi-asi

What is the core idea behind "Narrow, General, AGI, ASI: What We Mean and Why It Matters"?
1. The terminology ladder of AI capability is loaded. Clarify your definitions and you clarify your whole view of the field.
2. MMLU
3. human feedback
4. RLHF
Which term best describes a foundational idea in "Narrow, General, AGI, ASI: What We Mean and Why It Matters"?
1. AGI
2. ANI
3. ASI
4. transformative AI
A learner studying Narrow, General, AGI, ASI: What We Mean and Why It Matters would need to understand which concept?
1. ANI
2. ASI
3. AGI
4. transformative AI
Which of these is directly relevant to Narrow, General, AGI, ASI: What We Mean and Why It Matters?
1. ANI
2. AGI
3. transformative AI
4. ASI
Which of the following is a key point about Narrow, General, AGI, ASI: What We Mean and Why It Matters?
1. Level 0: No AI
2. Level 1: Emerging (equal to or better than unskilled human)
3. Level 2: Competent (50th percentile of skilled adults)
4. Level 3: Expert (90th percentile)
Which of these does NOT belong in a discussion of Narrow, General, AGI, ASI: What We Mean and Why It Matters?
1. Level 2: Competent (50th percentile of skilled adults)
2. Level 1: Emerging (equal to or better than unskilled human)
3. MMLU
4. Level 0: No AI
Which statement is accurate regarding Narrow, General, AGI, ASI: What We Mean and Why It Matters?
1. The label can trigger contracts and IPO milestones
2. Different stakeholders prefer different goalposts
3. No agreed benchmark means anyone can claim it
4. Anthropomorphizing leads to bad policy decisions
Which of these does NOT belong in a discussion of Narrow, General, AGI, ASI: What We Mean and Why It Matters?
1. No agreed benchmark means anyone can claim it
2. Different stakeholders prefer different goalposts
3. The label can trigger contracts and IPO milestones
4. MMLU
What is the key insight about "Where 2026 models sit" in the context of Narrow, General, AGI, ASI: What We Mean and Why It Matters?
1. Today's frontier LLMs are roughly Level 2 on general text tasks and occasionally Level 3 or 4 on narrow tasks like compe…
2. MMLU
3. human feedback
4. RLHF
What is the key insight about "The capability–deployment gap" in the context of Narrow, General, AGI, ASI: What We Mean and Why It Matters?
1. MMLU
2. A model can exhibit AGI-like capability in a lab while not being deployed at that level in products.
3. human feedback
4. RLHF
What is the recommended tip about "Ground your practice in fundamentals" in the context of Narrow, General, AGI, ASI: What We Mean and Why It Matters?
1. MMLU
2. human feedback
3. Every AI capability has an underlying mechanism. Understanding that mechanism tells you where it'll fail — which is more…
4. RLHF
Which statement accurately describes an aspect of Narrow, General, AGI, ASI: What We Mean and Why It Matters?
1. MMLU
2. human feedback
3. RLHF
4. If a CEO says their product is AGI, they are simultaneously making a scientific claim, a marketing claim, and often a trigger for contractua…
What does working with Narrow, General, AGI, ASI: What We Mean and Why It Matters typically involve?
1. A 2023 paper by Morris et al. at DeepMind proposed a two-axis framework: performance (from emerging to superhuman) and generality (narrow vs…
2. MMLU
3. human feedback
4. RLHF
Which of the following is true about Narrow, General, AGI, ASI: What We Mean and Why It Matters?
1. MMLU
2. The big idea: intelligence is multidimensional, and the AGI label collapses too much nuance.
3. human feedback
4. RLHF
Which best describes the scope of "Narrow, General, AGI, ASI: What We Mean and Why It Matters"?
1. It is unrelated to foundations workflows
2. It applies only to the opposite beginner tier
3. It focuses on The terminology ladder of AI capability is loaded. Clarify your definitions and you clarify your who
4. It was deprecated in 2024 and no longer relevant

← Back to interactive lesson

Tendril · Creators · AI Foundations

Narrow, General, AGI, ASI: What We Mean and Why It Matters

The terminology ladder of AI capability is loaded. Clarify your definitions and you clarify your whole view of the field.

40 min · Reviewed 2026

Why Definitions Are Political

The classical ladder

Level	Working definition
Narrow AI (ANI)	Performs a specific task, sometimes superhumanly
General AI (AGI)	Matches human-level performance across most cognitive tasks
Artificial Superintelligence (ASI)	Vastly exceeds human performance across all domains

Google DeepMind's Levels of AGI

Level 0: No AI
Level 1: Emerging (equal to or better than unskilled human)
Level 2: Competent (50th percentile of skilled adults)
Level 3: Expert (90th percentile)
Level 4: Virtuoso (99th percentile)
Level 5: Superhuman (outperforms 100 percent)

Why the AGI term is contested

No agreed benchmark means anyone can claim it
The label can trigger contracts and IPO milestones
Different stakeholders prefer different goalposts
Anthropomorphizing leads to bad policy decisions
Emergent capabilities blur thresholds further

Alternative framings worth knowing

Framing	Core concept
Transformative AI	AI that triggers changes of industrial revolution scale
Human-level AI	Performs any cognitive task humans can do
Automated AI researcher	Can itself do the work of an AI researcher (recursive)
AGI as economic test	Can do 50 percent of remote work as cheaply as a human

Implications for your thinking

Whenever you see AGI, ask whose definition
Benchmarks should be paired with generality claims
Short timelines and long timelines both have credible defenders
Plan products for a range of outcomes, not a point estimate

AGI is the horizon — it keeps receding as you approach it.
— An AI researcher

The big idea: intelligence is multidimensional, and the AGI label collapses too much nuance. Use multi-axis frameworks and be explicit about what you are measuring.

End-of-lesson check

15 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-creators-narrow-general-agi-asi

What is the core idea behind "Narrow, General, AGI, ASI: What We Mean and Why It Matters"?
1. The terminology ladder of AI capability is loaded. Clarify your definitions and you clarify your whole view of the field.
2. MMLU
3. human feedback
4. RLHF
Which term best describes a foundational idea in "Narrow, General, AGI, ASI: What We Mean and Why It Matters"?
1. AGI
2. ANI
3. ASI
4. transformative AI
A learner studying Narrow, General, AGI, ASI: What We Mean and Why It Matters would need to understand which concept?
1. ANI
2. ASI
3. AGI
4. transformative AI
Which of these is directly relevant to Narrow, General, AGI, ASI: What We Mean and Why It Matters?
1. ANI
2. AGI
3. transformative AI
4. ASI
Which of the following is a key point about Narrow, General, AGI, ASI: What We Mean and Why It Matters?
1. Level 0: No AI
2. Level 1: Emerging (equal to or better than unskilled human)
3. Level 2: Competent (50th percentile of skilled adults)
4. Level 3: Expert (90th percentile)
Which of these does NOT belong in a discussion of Narrow, General, AGI, ASI: What We Mean and Why It Matters?
1. Level 2: Competent (50th percentile of skilled adults)
2. Level 1: Emerging (equal to or better than unskilled human)
3. MMLU
4. Level 0: No AI
Which statement is accurate regarding Narrow, General, AGI, ASI: What We Mean and Why It Matters?
1. The label can trigger contracts and IPO milestones
2. Different stakeholders prefer different goalposts
3. No agreed benchmark means anyone can claim it
4. Anthropomorphizing leads to bad policy decisions
Which of these does NOT belong in a discussion of Narrow, General, AGI, ASI: What We Mean and Why It Matters?
1. No agreed benchmark means anyone can claim it
2. Different stakeholders prefer different goalposts
3. The label can trigger contracts and IPO milestones
4. MMLU
What is the key insight about "Where 2026 models sit" in the context of Narrow, General, AGI, ASI: What We Mean and Why It Matters?
1. Today's frontier LLMs are roughly Level 2 on general text tasks and occasionally Level 3 or 4 on narrow tasks like compe…
2. MMLU
3. human feedback
4. RLHF
What is the key insight about "The capability–deployment gap" in the context of Narrow, General, AGI, ASI: What We Mean and Why It Matters?
1. MMLU
2. A model can exhibit AGI-like capability in a lab while not being deployed at that level in products.
3. human feedback
4. RLHF
What is the recommended tip about "Ground your practice in fundamentals" in the context of Narrow, General, AGI, ASI: What We Mean and Why It Matters?
1. MMLU
2. human feedback
3. Every AI capability has an underlying mechanism. Understanding that mechanism tells you where it'll fail — which is more…
4. RLHF
Which statement accurately describes an aspect of Narrow, General, AGI, ASI: What We Mean and Why It Matters?
1. MMLU
2. human feedback
3. RLHF
4. If a CEO says their product is AGI, they are simultaneously making a scientific claim, a marketing claim, and often a trigger for contractua…
What does working with Narrow, General, AGI, ASI: What We Mean and Why It Matters typically involve?
1. A 2023 paper by Morris et al. at DeepMind proposed a two-axis framework: performance (from emerging to superhuman) and generality (narrow vs…
2. MMLU
3. human feedback
4. RLHF
Which of the following is true about Narrow, General, AGI, ASI: What We Mean and Why It Matters?
1. MMLU
2. The big idea: intelligence is multidimensional, and the AGI label collapses too much nuance.
3. human feedback
4. RLHF
Which best describes the scope of "Narrow, General, AGI, ASI: What We Mean and Why It Matters"?
1. It is unrelated to foundations workflows
2. It applies only to the opposite beginner tier
3. It focuses on The terminology ladder of AI capability is loaded. Clarify your definitions and you clarify your who
4. It was deprecated in 2024 and no longer relevant

← Back to interactive lesson