Defining Intelligence

What Intelligence
Actually Means

Before measuring machine intelligence, we must define it rigorously — grounded in cognitive science, not marketing.

1

Define

A scientific, testable framework.

2

Reject

Common myths and false proxies.

3

Operationalize

Turn theory into measurable dimensions.

4

Measure

Score and profile against human baselines.

Misconceptions

What Intelligence Is Not

Intelligence ≠ Knowledge

A hard drive with Wikipedia is not intelligent. Storage is not cognition.

Intelligence ≠ Speed

Calculators are fast. They are not intelligent.

Intelligence ≠ Task Performance

Excelling at one task does not demonstrate general cognitive capability.

Intelligence ≠ Pattern Matching

Autocomplete predicts the next token. Understanding requires more.

Intelligence ≠ Mimicry

Sounding intelligent is not the same as being intelligent.

Working Definition

The MHIL Definition

"The capacity to learn from experience, reason about novel situations, solve problems creatively, adapt to changing environments, and understand one's own cognitive processes."

Learn from Experience

Generalize principles, update beliefs, transfer knowledge across domains.

Reason About Novelty

Handle situations absent from training data.

Solve Creatively

Generate new solutions. Break assumptions when appropriate.

Adapt

Notice failure, change strategies, manage resources effectively.

Metacognition

Know what you don't know. Calibrate confidence accurately.

The Scale

The MHIL Scale

Scores from 0 to 230+, calibrated to human cognitive development (σ = 15). MHIL-100 represents average adult baseline.

0–35
Minimal Cognitive Function
Extremely limited reasoning. Heavy scaffolding required.
Failed / very weak
35–85
Child to Adolescent Range
Improving abstraction. Weak metacognition and adaptability.
← All current AI
85–115
Adult Normative Functioning
Human baseline = 100. Full abstract reasoning capability.
Target zone
115–145
Superior to Gifted Capacity
Exceptional reasoning, creativity, and metacognitive awareness.
Not yet achieved
160+
Beyond Individual Humans
Consistent genius-level performance across all dimensions.
AGI territory
200+
Beyond Human Species Scale
Novel forms of reasoning. Measurement uncertainty rises sharply.
Early ASI
0
Adult Baseline
±15
1 Standard Deviation
0
Core Dimensions
Cognitive Profiles

Why a Single Score Is Not Enough

MHIL reports both an overall score and a multi-dimensional profile. Two systems can share the same score but have radically different capabilities and risks.

Example: Leading AI System (2026)
Overall: MHIL-52 — Child-level cognition with spiky profile
SAMPLE REPORT
Logical Reasoning68
Social Intelligence61
Working Memory54
Critical Thinking51
Causal Understanding43
Adaptive Behavior39
Creative Problem-Solving35
Metacognitive Awareness24
⚠ Child-level overall — strongest in pattern-based logic
✗ Cannot reliably self-monitor or recognize own errors
✗ Not suitable for autonomous decision-making

Explore the Eight Dimensions

See how we test each cognitive capability.

View Benchmarks →