A new paper argues that the uneven, "jagged" abilities of modern AI systems stem from a missing training signal the authors call "cognitive dark matter" (CDM)—brain functions that shape behavior but are nearly invisible to behavioral observation alone.
The Research
Published on arXiv on March 3, 2026 (revised July 31, 2026), the paper is authored by Patrick J. Mineault, Thomas L. Griffiths, and Sean Escola. Griffiths is a professor of psychology and computer science at Princeton University and a leading researcher in computational cognitive science; Mineault and Escola work at the intersection of neuroscience and machine learning.
The authors define cognitive dark matter as brain functions that meaningfully shape behavior yet are hard to infer from behavior alone. They identify six key CDM domains:
- Metacognition (thinking about your own thinking)
- Cognitive flexibility (adapting to new rules or contexts)
- Lifelong learning (accumulating knowledge over time)
- Abductive reasoning (inferring the best explanation from incomplete information)
- Social and common-sense reasoning
- Emotional intelligence
The team presents evidence that these CDM-loaded functions are largely unmeasured in current AI benchmarks and that the large-scale neuroscience training datasets needed to instill them do not yet exist. They outline a research program built on three complementary data types designed to surface CDM for model training: (i) latent variables from large-scale cognitive models, (ii) process-tracing data such as eye-tracking and think-aloud protocols, and (iii) paired neural-behavioral data.
"These data will enable AI training on cognitive process rather than behavioral outcome alone, producing models with more general, less jagged intelligence," the authors write.
Why It Matters
The same blind spots that make AI benchmarks incomplete likely apply to how we measure human intelligence, too. Standard IQ tests and most cognitive assessments focus on outcomes—correct answers, reaction times, scores—rather than the processes behind them. If CDM functions like metacognition and cognitive flexibility are as important as this paper suggests, then understanding and training them could be key to improving your own learning, decision-making, and adaptability. The researchers note a "dual benefit": the same data that could improve AI would also advance our understanding of human intelligence itself.
What You Can Do
- Practice metacognition: after a task, ask yourself what strategy you used and what you'd change next time.
- Train cognitive flexibility by switching between different types of puzzles or tasks in one session.
- Engage in think-aloud problem solving—verbalizing your reasoning can surface hidden thought patterns.
- Challenge yourself with social and emotional reasoning tasks, such as perspective-taking exercises.
Source: arXiv q-bio.NC
Curious about your own brain? Take our free adaptive IQ test or try 306 brain training levels.