Summary: "On Intelligence"

web layout web layout  

This article summarizes a number of key concepts that are found in Jeff Hawkins's book "On Intelligence". In this book, Hawkins develops a powerfull theory of how the human brain works and what intelligence is. He focuses on the neocortex because he thinks all essential aspects of intelligence occur in the neocortex, but he does agree that other brain structures are important to the functioning of the neocortex.

According to Hawkins the neocortex is a memory prediction engine. It predicts spatial- and temporal-patterns. It does not matter where the patterns are coming from: eyes, ears, touch, muscle or other parts of the neocortex.

The following are interesting citations from the book, at least to me, grouped by chapter.

Chapter 1 - Artificial Intelligence

  • Understanding occures when reading a story (whithout any outside signals), not when aswering questions about it.
    (page 20)

Chapter 2 - Neural Networks

  • There are ten times as many connections feeding information back to the thalamus as forward to the neocortex.
    (page 25)

  • Auto-associative memory with a time delay to the feedback can replay a sequence of patterns from a few partial patterns. People learn everything as a sequence of patterns.
    (page 30/31)

Chapter 3 - The Human Brain

  • You can lead a pretty normal life without the cerebellum, the brain part with the largest number of cells.
    (page 41)

  • The cortical sheet comprises 6 layers is 2 mm thick and roughly the size of a large dinner napkin.
    (page 42)

  • Daniel Fellman and David van Essen made a detailed map of the monkey neocortex.
    (page 45)

  • The functional regions/areas in the neocortex are hierachical organized with lateral connections. Information is flowing in both direcions.
    (page 45-47)

  • The neuroscientist Vernon Mountcastle points out in his paper titled "An Organizing Principle for Cerebral Function" that the neocortex is remarkable uniform in appearance and structure.
    (page 50)

  • The neocortex processes signals from the ear the same as the signals from the eyes, e.i. the neocortex is the same for all sences.
    (page 51)

  • A devoted area for written letters and digits tell us that the neocortex is still dividing (organizing) itself into task specific area's long into childhood.
    (page 53)

  • All the information that enters your mind (vision, hearing, touch or other) comes in as spatial and temporal patterns.
    (page 57)

Chapter 4 - Memory

  • The brain is not a parallel-computer. The brain recognizes an image in about half a second, or a chain of a hundred neurons long. One hundred computer instructions are not enough to answer a difficult problem.
    (page 66/67)

  • The brain retrieves the answer to problems from memory. The entire neocortex is a memory system. It isn't a computer at all.
    (page 68)

  • There are four attributes of neocortical memory that are different from computer memory:
    • The neocortex stores sequences of patterns.
    • The neocortex recalls patterns auto-associative.
    • The neocortex stores patterns in an invariant form.
    • The neocortex stores patterns in a hiearchy.

    (page 70)

  • To make a specific prediction the brain combines the knowledge of the invariant structure with the most recent details.
    (page 83)

Chapter 5 - A New Framework for Intelligence

  • Prediction is so pervasive that what we "percieve" - that is, how the world appears to us - does not come solely from our senses. What we percieve is a combination of what we sense and what our brains' memory-derived predictions.
    (page 87)

  • Prediction is the primary function of the neocortex, and the foundation of intelligence. The cortex is an organ of prediction.
    (page 89)

  • Behavior is best understood as a by product of prediction.
    (page 89)

  • A computer-driven robot would fall over, not realizing that anything was amiss, while a human would know as soon as the foot continues for even a fraction beyond the spot where the brain had expected it to stop.
    (page 91)

  • Te human brain is more intelligient than that of other animals because it can make predictions about more abstract kinds of patterns and longer temporal pattern sequences.
    (page 96)

  • The neocortex evolved to make more efficient use of existing behaviors, not to create new behaviors.
    (page 98)

Chapter 6 - How the Cortex Works

  • Scientist have been ignoring the feedback connections, but the feedback is needed to maake predictions. Prediction requires a comparison between what is happening and what you expect to happen.
    (page 113)

  • Hearing and feeling require a flow of sensorary patterns for recognition. A single pattern is not enough. The same is true for vision, which researchers generally ignored.
    (page 116)

  • All predictions are learned by experience.
    (page 119)

  • All cortical areas, including the visual area V1, should be simular, that is form invariant representations and receive converging inputs from two or more lower areas. This means that V1 should be considered as many small regions only connected indirectly, not as a single large area.
    (page 122/123)

  • A association area does not need to know where the inputs are comming from: vision, sound, touch or a combination. Rather, the job of any cortical area is to find out how its inputs are related, to memorize the sequence of correlations between them, and to use this memory to predict how the inputs will behave in the future.
    (page 123)

  • The brain can be said to store sequences of sequences.
    (page 129)

  • By collapsing predictable sequences into "named objects" at each region in our hierarchy we achieve more and more stability the higher we go. This creates invariant representatins. The oposite happens as a pattern moves down the hierarchy: stable patters get "unfolded" in sequences.
    (page 130)

  • The thalamus is responsible for the delayed feedback that lets the neocortex learn sequences, just like the auto-associative memory models.
    (page 146)

  • If recognition does not occur, an unexpected pattern will keep propagating up the cortical hierarchy until some higher reqion can interpret it as part of its normal sequence of events.
    (page 159)

  • The hippocampus occupies the peak of the neocortical pyramid. It temporarily stores the pattern that are unexplained and unanticipated.
    (page 170/171)

  • There is a second path up (not down) the hierarchy via de thalamus. The path is mostly shutdown in the thalamus or else the pattern is directly passed on. This makes it possible to focus on details.
    (page 172/173)

Chapter 7 - Consciousness and Creativity

  • All mammals, from rats to cats to humans, have a neocortex. They are all intelligent, but to differing degrees.
    (page 180)

  • Intelligence can be traced over three epochs, each using memory and prediction. The first would be when species used DNA as medium for memory. This include single-cell organisms and plants. The second epoch began when nature invented modifiable nervous systems that could quickly form memories. This includes the creation and expansion of the neocortex. The third epoch is unique to humans and began with invention of language and the expansion of our large neocortex.
    (page 182)

  • Creativity can be defined as making predictions by analogy, something that occurs everywhere in cortex. Creativity occurs along a continuum.
    (page 183)

  • Consciousness is the process of the neocortex forming memories.
    (page 199)

  • To imagine something, you merely let your predictions turn around and become inputs.
    (page 200)
(C) Vincent Kessels (, V2.21 - 1999-2008