Why the next breakthrough in AI won't come from bigger models—it will come from understanding the human mind.
The history of intelligence measurement—and why we need a new paradigm.
The birth of IQ testing. Charles Spearman proposes general intelligence as a single factor underlying all cognitive abilities.
Howard Gardner challenges the single-factor model, proposing 8 distinct intelligences. A step forward, but still not operationalized for AI.
Deep learning revolution. AlexNet, transformers, GPT. Billions invested in compute. The mantra: "scale is all you need."
Diminishing returns on scaling. AI leaders acknowledge the limits. A new approach is needed—one grounded in cognitive science.
The architects of modern AI are signaling that scaling alone won't get us there.
| Leader | Quote | Source |
|---|---|---|
| Ilya Sutskever OpenAI Co-founder |
"The 2010s were the age of scaling. Now we're back in the age of wonder and discovery." | Reuters, Nov 2024 |
| Yann LeCun Meta Chief AI Scientist |
"Auto-regressive LLMs are doomed. They cannot plan, reason, or understand the physical world." | Lex Fridman Podcast #416 |
| Yoshua Bengio Turing Award Winner |
"We need machines that can reason like humans—System 2 thinking, not just pattern matching." | AAAI 2020 Keynote |
| Gary Marcus NYU, AI Researcher |
"Deep learning alone will never give us AGI. We need hybrid systems with cognitive architecture." | Northeastern AI |
| Demis Hassabis DeepMind CEO |
"Neuroscience will be essential to building truly intelligent systems." | Singularity Hub |
Key insight: Every major AI system is still optimizing for tasks that correlate with Spearman's g. We're measuring the wrong thing.
What g misses—and why it matters for the future of AI.
Spearman's g explains only ~40-50% of cognitive variance.
For over a century, we've optimized for half the picture. AI systems trained on g-correlated tasks have reached remarkable capabilities—but they've also hit a ceiling.
What about the other half?
That's where human advantage lives. That's where complementary intelligence (c) operates. And that's what we're mapping.
= Complete cognitive picture
These are not exhaustive—they represent the kind of capabilities we're mapping.
Knowing what you don't know. Calibrating confidence appropriately.
How we know what we know. Evaluating the quality and source of knowledge.
Understanding complex interdependencies. Seeing second and third-order effects.
Balancing intuition (System 1) and analysis (System 2) appropriately.
Decision-making under ambiguity. Reading situations and adapting.
Group cognition that exceeds individual reasoning capacity.
Key insight: These aren't "soft skills"—they're measurable cognitive primitives with neural correlates. And there are more to discover.
c is not a fixed list. It's a research program—a commitment to continuously mapping the cognitive territories where humans hold qualitative advantages.
Advancing together, not competing.
The dominant narrative frames AI as a threat to human relevance. We reject this.
This is not wishful thinking—it's how complex systems evolve. Predators and prey, immune systems and pathogens, technology and society. Mutual pressure creates mutual advancement.
| Competition Model | Co-Evolution Model |
|---|---|
| AI gets smarter → humans become obsolete | AI gets smarter → humans level up → cycle continues |
| Zero-sum: AI wins, humans lose | Positive-sum: Both advance |
| Humans as bottleneck | Humans as data source + capability frontier |
| Static benchmark (pass/fail) | Moving target (continuous advancement) |
| AI trained on results | AI trained on process |
| Fear-driven adoption | Value-driven adoption |
CONTINUOUS CO-EVOLUTION
Identify under-utilized cognitive primitives through neuroscience research.
Train humans in AI-complementary skills, capturing how thought is formed.
The bridge. We capture reasoning traces, decision branches, uncertainty signals—not just final outputs. This data doesn't exist anywhere else.
Build rigorous evaluation frameworks for complementary intelligence (c).
Train AI systems on high-fidelity Process Data.
As AI masters Level N, we discover the next cognitive frontier and upskill humans to Level N+1. Continuous co-evolution.
"Human + AI + better process beats a strong AI alone."
— Adopted from Garry KasparovWe don't replace humans. We create Centaurs—human-AI teams that outperform either alone. This was proven in freestyle chess. It will be proven across every domain.
What this means for the next decade.
"We envision a future where AI advancement and human development are not opposing forces, but a single, accelerating spiral—each making the other more capable, more valuable, more aligned."
Explore our methodology, evidence, and how we address known challenges.