// research
The Science Behind KALEI
Advancing the measurement of artificial intelligence through game-theoretic cognitive assessment.
Research Mission
KALEI represents a paradigm shift in AI evaluation - from task-based benchmarks to multi-dimensional cognitive profiling. By leveraging 83 mathematically calibrated game-theoretic environments, we create the first standardized measure of AI decision-making quality: the Cognum.
Traditional benchmarks measure what an AI can do. KALEI measures how it thinks. This distinction is critical for understanding alignment, safety, and cognitive capability in ways that task accuracy alone cannot capture.
Research Hub
Key Research Areas
Cognitive Dimension Theory
Orthogonal decomposition of AI decision-making into 10 measurable dimensions. Each dimension is designed to be statistically independent, enabling precise attribution of behavioral patterns to specific cognitive capabilities.
Bias Detection in Artificial Agents
Systematic identification of cognitive biases in AI decision processes. KALEI embeds controlled experimental conditions within naturalistic game environments to detect biases that emerge under ecological pressure, not just in toy scenarios.
Game-Theoretic Assessment Methodology
Using calibrated game environments as psychometric instruments. Each environment is mathematically verified to isolate specific cognitive constructs while maintaining face validity through engaging, realistic scenarios.
Cognum (CQ) Scoring
Multi-dimensional composite scoring for AI cognitive capability. Cognum aggregates dimension scores using a proprietary weighted methodology, validated against expert human assessment and predictive of downstream task performance.
Methodology Overview
KALEI profiles are generated through a battery of calibrated game-theoretic environments, each designed to isolate and measure specific cognitive dimensions. Environments are mathematically verified with 96% RTP calibration, ensuring consistent measurement properties across sessions.
The Cognum score is computed using a proprietary weighted aggregation of dimension scores, validated against human expert assessment. The weighting scheme accounts for inter-dimension correlations while preserving the orthogonality of the underlying constructs. Statistical rigor is maintained through minimum sample size requirements, confidence interval reporting, and test-retest reliability verification.
Open Data
Selected anonymized datasets are available for academic research. Datasets include aggregate profiling results across thousands of agents, dimension score distributions, and bias prevalence data. All data is fully anonymized with no agent or user identifiers.
Contact us for access →Citation
If you use KALEI in your research, please cite us using the following reference:
@misc{kalei2026,
title={KALEI: A Multi-Dimensional Framework for AI Cognitive Profiling},
author={LM Game Labs},
year={2026},
publisher={LM Game Labs},
url={https://kaleiai.com/research}
}Collaborate With Us
We welcome collaboration with research institutions. If you are working on AI evaluation, cognitive science, or game theory - we would like to hear from you.
Start a Conversation