jeff hawkins on pattern and predictability

Hawkins (2004: 12) sets the scene for his theoretical framework by contrasting the fundamental principles on which computers and brains are built:

One is programmed, one is self-learning. One has to be perfect to work at all, one is naturally flexible and tolerant of failures.

He (2004: 18) stresses that:

Even today no computer can understand language as well as a three-year-old or see as well as a mouse.

To cut to the chase, Hawkins’ theory of intelligence emphasizes the primacy of pattern recognition in the neocortex. Coupled with the brain’s ability to recognize invariant sequences in real time, are elaborate feedback mechanisms that generate predictive patterns. Usually below the level of conscious awareness, the brain is in a constant state of anticipation. Hawkins (2004: 13) holds that a “fundamental flaw” of Artificial Intelligence is that it “fails to adequately address what intelligence is or what it means to understand something.” Hawkins reminds us that (2004: 31, 58) that “people learn practically everything, as a sequence of patterns.” Even vision itself is “more like a song than a painting.” For Hawkins (2004: 63) “all our knowledge of the world is a model based on patterns.”

Remember the brain is a dark quiet box with no knowledge of anything other than the time-flowing patterns on its input fibres. Your perception of the world is created from these patterns nothing else.

According to Hawkins (2004: 84, 104) “[m]aking predictions is the essence of intelligence.” “To know something means that you can make predictions about it.” Hawkins proposal encompasses the brains capacity to deal with fuzzy or incomplete information and our ability to generalize or make stereotypes. His theory also takes into account the anatomy of the neocortex and its rich connectivity with the primitive thalamus below, and the hippocampus above, its hierarchy. Hawkins (2004: 89) is uncompromising:

Prediction is not just one of the things your brain does. It is the primary function of the neocortex, and the foundation of intelligence. The cortex is an organ of prediction.

Strictly speaking Jeff Hawkins is a “gentleman amateur” in the field of neuroscience, who has popularized a powerful model for intelligence and the brain. His earlier claim to fame was the creation of the PalmPilot and some other hand held computer devices.

Hawkins’ impetus (2004: 10) came from Nobel laureate, Francis Crick, who, in the late seventies, had declared that, despite all the accumulated research data on the brain, “a broad framework of ideas” was “conspicuously lacking.” Hawkins saw this as an understated way of saying, “We don’t have a clue how this thing works.” For Hawkins (2004: 16) the brain remained the “impenetrable black box” of the behaviorists.

Three Golden Columns - The neocortex is organised into thousands of columns of neurons. Each column has a diameter of 0.5mm and contains 10,000 neurons. The neocortex is also organised into 6 layers and the golden neurons shown are the large output pyramidal neurons in the fifth layer. In the background are other neurons making up the neocortical column.

Image and verbatim caption from IBM/EPFL Blue Brain Project (permission application in progress)

Hawkins describes (2004: 40-43) the neocortex as a thin, mostly homogenous sheet of densely packed neural tissue enveloping most of the older parts of the brain. It is about 2mm thick, has 6 discernable layers and is about the size of “a large crumpled dinner napkin.” One cubic millimeter of neocortex contains about 100,000 neurons. A conservative estimate for the number of neurons in the entire structure is some thirty trillion. Since an individual cortex cell connects to 5 or 10 thousand others, the number of possible patterns between cells soon becomes hyperastronomical.

Hawkins’ framework rests on pioneering work in the seventies by neuroscientist, Vernon Mountcastle, who emphasized that the neocortex is “remarkably uniform in appearance and structure.’ Mountcastle’s singular insight was, given that neocortical regions, “all look the same, perhaps they are all performing the same basic operation.” The neocortex has a six-layered “columnar architecture.” Hawkins echoes Mountcastle (2004: 141) in assuming a “single cortical algorithm” at work across the repeating, six-layered, columnar architecture of the neocortex.

Hawkins defines (2004: 70, 171) four criteria for this algorithm. Specifically, the neocortex:

stores sequences of patterns

recalls patterns auto-associatively

stores patterns in an invariant form

stores patterns in a hierarchy

Hawkins hypothesizes that an elaborate process of generalization and classification is at play up and down the cortical hierarchy; with the hippocampus at the apex as the depository of “input that is truly new and unexpected.” Hawkins proposes (2004: 125) that:

all regions of the cortex form invariant representations of the world underneath them in the hierarchy. There is beauty in this.

The sheer quantity of interconnections within, and feeding in and out of, the neocortex is beyond any visualization. Although Hawkins himself does not dwell on this, it is worth mentioning that these connections are characterized by a certain repetition and redundancy. Rather than representing a deficiency, the inherent degeneracy may actually facilitate a whole catalog of brain features, including: its famous plasticity, its ability to “fill in” the missing parts of imperfect information, its penchant for generalization and classification, its uncanny ability to acquire language, its propensity for making analogies and loose associations, and its quirky and often erroneous retrieval of sequential information from its own shadowy memories.

Hawkins (2004: 147) envisages:

converging patterns going up the cortical hierarchy, diverging patterns going down the cortical hierarchy, and a delayed feedback through the thalamus…

With regard to the thalamus, Hawkins notes (2004: 84) that: connections going backward (toward the input) exceed the connections going forward by almost a factor of ten! For Hawkins (2004: 156) Every moment in your waking life, each region of your neocortex is comparing a set of expected columns driven from above with a set of observed columns driven from below. Where the two intersect is what we perceive… Furthermore (2004: 85), if a prediction is erroneous, “attention is immediately aroused.”

Hawkins asserts that this explains the difficulty we encounter “not looking at people with deformities.” He also points to our heightened appreciation of “highly creative works of art” which “violate our predictions.”

Hawkins, Jeff with Blakeslee, Sandra (2005) On Intelligence. Times Books/Henry Holt and Company, New York.

Vernon Mountcastle

Hawkins in full flow