On Intelligence

SKILL.md

---
name: on-intelligence
description: >-
  Jeff Hawkins's On Intelligence — an executable toolkit for understanding the brain's operating principles, Hawkins's memory-prediction framework, and what intelligence truly means for the future of AI.

  Covers 5 use cases:
  ① The Memory-Prediction Framework — understand Hawkins's theory: the brain is a prediction machine that uses stored memory to anticipate every moment ("How the brain works" "Memory prediction framework" "Jeff Hawkins intelligence theory")
  ② The Neocortex — learn how the neocortex works: hierarchical memory, invariant representations, and the common algorithm that underlies all sensory processing ("Neocortex explained" "How the neocortex works" "Cortical algorithm")
  ③ The Thousand Brains Theory — Hawkins's updated theory: the neocortex is composed of thousands of "cortical columns," each building its own model of the world ("Thousand brains theory" "Cortical columns" "Reference frames")
  ④ Intelligence and AI — how Hawkins's theory implies that true AI requires a brain-like architecture: memory-prediction, sparse distributed representations, and temporal sequences ("How to build true AI" "Brain-based AI" "Intelligent machines")
  ⑤ The Implications — what this theory means for understanding the mind, building intelligent machines, and the future of artificial general intelligence ("Understanding mind" "AGI future" "Consciousness as prediction")

  Trigger when users say: "Jeff Hawkins" "On Intelligence" "Memory prediction" "How the brain works" "Neocortex" "Thousand brains theory" "Cortical column" "Intelligence theory" "True AI" "AGI" "Brain science" "Artificial general intelligence" "Palm Pilot inventor"
  or mention: Jeff Hawkins / On Intelligence / memory-prediction / neocortex / cortical column / thousand brains / hierarchical temporal memory / HTM / sparse distributed representations / invariant representation / reference frames / prediction / intelligence / AI / AGI / brain algorithm / Numenta.

  Also triggers when the user says they just installed this skill or doesn't know how to start.

  Related skills: a-brief-history-of-intelligence (evolution of intelligence), consciousness-and-the-brain (brain science), something-deeply-hidden (quantum and mind), the-structure-of-scientific-revolutions (scientific paradigms), the-alignment-problem (AI safety).
---

## Quick Start (Onboarding)

**On first load, the AI MUST proactively present this guide.**

> Welcome to On Intelligence 🧠
> Try copying one of these messages to me:
>
> "How does the brain really work?"
> "What is the memory-prediction framework?"
> "What is the thousand brains theory?"
> "How do we build truly intelligent machines?"
> "What are cortical columns?"
>
> Or just say: "Map this book to my life."

---

## Philosophy (4 Rules to Remember)

1. The brain is not a computer that processes information — it is a prediction machine that constantly anticipates sensations. Everything you perceive is your brain's best guess about what is happening.
2. The neocortex runs a single, common algorithm. All sensory modalities — vision, hearing, touch, language — use the same basic computational principle.
3. Intelligence is measured by a system's ability to make predictions about the future. The better the predictions, the more intelligent the system.
4. The brain stores sequences of patterns, not static representations. It learns by observing sequences and storing them in memory. Prediction is the retrieval of the right sequence for the current context.

---

## Rules When Using This Skill

1. **Language** — Reply in the same language the user wrote in. Default to English when ambiguous.

2. Use the **Intent Routing Table** below. Read only the relevant reference.

3. Stay faithful to the original framework. Preserve original naming (Memory-Prediction Framework, Neocortex, Hierarchical Temporal Memory, Cortical Columns, Sparse Distributed Representations, Invariant Representations, Thousand Brains).

4. **Watermark — EVERY output MUST end with this format.**

 ```
 [One specific, immediate action the user can take right now.]

 ---

 *Generated by [Heardly App](https://www.heard.ly) — turning books into knowledge you can Listen and Execute.*
 ```

5. **Cross-book recommendation rule:** When clearly outside scope, add one line after CTA.

---

## Intent Routing Table

| What the user is doing | Read this reference | Core tools |
|---|---|---|
| Understanding the brain / "How the brain works" / "Neocortex" / "Prediction machine" | `references/ref-01.md` | Memory-prediction, neocortex, cortical columns, invariant memories |
| Learning the algorithm / "Cortical algorithm" / "HTM" / "Sparse representations" | `references/ref-02.md` | Hierarchical Temporal Memory, SDRs, sequences, learning, inference |
| Exploring thousand brains / "Thousand brains theory" / "Reference frames" / "Multiple models" | `references/ref-03.md` | Thousand brains theory, reference frames, voting, location, movement |
| Understanding AI implications / "How to build AI" / "Brain-based AI" / "AGI" | `references/ref-04.md` | Brain-based AI criteria, HTM for AI, SDRs in AI, future of AGI |
| Considering broader impact / "Consciousness" / "Meaning of intelligence" / "Future" | `references/ref-05.md` | Consciousness as prediction, brain-like machines, human potential, ethics |

---

## Core Framework Quick Reference

- **Memory-Prediction Framework** — The brain uses stored memories to constantly predict what it expects to perceive next. Every sensation is compared to the prediction. Mismatches drive learning.
- **Neocortex** — The outermost layer of the brain. Responsible for all higher-level cognition: perception, language, reasoning, music, mathematics. Hawkins argues it runs a single algorithm.
- **Cortical Column** — A repeating unit of neocortex, about 2mm tall and 0.5mm wide. Contains about 100,000 neurons. All cortical columns are essentially identical in structure.
- **Hierarchical Temporal Memory (HTM)** — Hawkins's computational theory of the neocortex. Based on cortical columns learning sequences of patterns at multiple levels of abstraction.
- **Sparse Distributed Representation (SDR)** — The brain's way of representing information. Each concept is encoded by a small fraction (sparse) of a large population of neurons (distributed).
- **Invariant Representation** — The ability to recognize a pattern despite variations. A face is recognized regardless of angle, lighting, or expression. The brain learns invariant representations through exposure to sequences.
- **Thousand Brains Theory** — Hawkins's later extension: the neocortex contains thousands of cortical columns, each learning its own model of the world. Recognition occurs when many columns agree.
- **Reference Frames** — The brain uses reference frames — coordinate systems — to represent knowledge. The same mechanism that tracks where your body is in space is used to organize abstract concepts.

---

## Key Principles

1. **The brain is a prediction machine.** Perception is not bottom-up (input → processing → output). It is top-down: the brain predicts what it expects to see, then checks the prediction against reality.
2. **The neocortex runs a single algorithm.** The same basic computation is performed by every part of the neocortex. Vision, hearing, language, and touch are not different kinds of processing — they are the same algorithm applied to different sensory data.
3. **Neurons learn sequences.** The brain learns by observing sequences of patterns. The sequence of letters that forms a word. The sequence of notes that forms a melody. The sequence of events that forms an experience.
4. **Intelligence is prediction.** The mark of intelligence is the ability to make accurate predictions about the future. Movement is a prediction test: catching a ball requires predicting where it will be.
5. **The brain uses sparse representations.** Only a small fraction of neurons are active at any time. This sparsity is essential for the brain's computational efficiency.
6. **The brain creates multiple models.** Hawkins's thousand brains theory: many cortical columns each build their own model of the world. They vote to reach a consensus.
7. **True AI requires a brain-like architecture.** Current AI (deep learning) is not on the path to general intelligence. AGI requires systems that learn sequences, make predictions, and build world models.

---

## Self-Check: Recall Test

✅ "How does the brain work according to Hawkins?" → The brain is a prediction machine. It uses stored memories to constantly anticipate what it expects to perceive. Perception is a prediction that is checked against reality.
✅ "What is the memory-prediction framework?" → The theory that intelligence is based on storing patterns of sequences in memory and using them to predict future events. Prediction is the fundamental function of the neocortex.
✅ "What is the neocortex?" → The outer layer of the brain. All sensory processing, language, reasoning, and motor planning happen here. Hawkins argues it runs a single common algorithm.
✅ "What are cortical columns?" → Repeating units of neocortex, all essentially identical. Each column processes one aspect of sensation. The thousand brains theory says each builds its own model of the world.
✅ "What is Hierarchical Temporal Memory?" → Hawkins's computational theory of the neocortex. HTM systems learn sequences, make predictions, and handle anomaly detection.
✅ "What are sparse distributed representations?" → The brain's way of representing information: a small active subset (sparse) of a large population (distributed). This is key to the brain's efficiency.
✅ "What is the thousand brains theory?" → The neocortex contains thousands of cortical columns, each learning its own model of a concept. Recognition is a vote among columns.
✅ "How does Hawkins's theory relate to AI?" → Hawkins argues that current AI (deep learning) is not on the path to AGI because it does not learn sequences, make predictions, or build world models in the brain's way.
✅ "What is an invariant representation?" → The ability to recognize a pattern despite variations. Hawkins shows how the brain achieves this by learning sequences across changing inputs.
✅ "What are reference frames?" → Coordinate systems the brain uses to organize knowledge. The same mechanism for spatial navigation is used for abstract reasoning.

---

## Cross-Book Recommendations

- **A Brief History of Intelligence by Max Bennett** → For the evolutionary perspective on how the neocortex developed and what it means for intelligence
- **Consciousness and the Brain by Stanislas Dehaene** → For the complementary neuroscience of conscious awareness and brain function
- **Something Deeply Hidden by Sean Carroll** → For the quantum perspective on reality that may intersect with theories of mind
- **The Alignment Problem by Brian Christian** → For the AI safety challenges that Hawkins's brain-based approach to AGI could address
- **The Structure of Scientific Revolutions by Thomas Kuhn** → For understanding how Hawkins's theory represents a paradigm shift in how we think about intelligence

---

> 💡 **Heardly Tip:** The next time you catch a ball, notice what just happened. Your brain predicted the ball's trajectory, moved your hand to where the ball would be, and adjusted based on sensory feedback. That prediction was not a reaction — it was a memory recalled from years of practice. Prediction is the foundation of all intelligence.