An independent research project developing new approaches to evaluating cognitive capabilities in artificial intelligence systems.
Current AI benchmarks measure knowledge recall and pattern matching — not intelligence. MHIL measures the cognitive capabilities that actually matter.
A rigorous, science-grounded framework that captures what intelligence actually is — across 8 measurable cognitive dimensions.
Comprehensive multi-turn conversational assessments that probe genuine understanding, not memorized responses. Evaluation framework being designed.
Independent certification and public rankings calibrated against human cognitive development stages. Certification methodology in development.
AI systems consistently score 90%+ on traditional benchmarks while failing basic reasoning tasks that any teenager could handle. The industry's measurement tools are broken.
Hypothetical illustration: how a process-based evaluation might score systems differently from knowledge-recall benchmarks.
This compares MHIL's design approach to existing benchmark frameworks.
| Capability | MHIL | MMLU | HumanEval | HELM |
|---|---|---|---|---|
| Measures cognitive process | ✓ | ✗ | ✗ | ✗ |
| Multi-turn conversational | ✓ | ✗ | ✗ | Partial |
| Anti-gaming design | ✓ | ✗ | ✗ | ✗ |
| Human-calibrated scale | ✓ | ✗ | ✗ | ✗ |
| Dimensional profiling | ✓ | ✗ | ✗ | Partial |
| Dynamic test generation | ✓ | ✗ | ✗ | ✗ |
| Independent evaluation | ✓ | ✗ | ✗ | ✓ |
| Metacognition assessment | ✓ | ✗ | ✗ | ✗ |
Explore the cognitive framework or get in touch if you're interested in early collaboration.