Book Review: On Intelligence

On Intelligence - Jeff Hawkins and Sandra Blakeslee

What’s it about: A book from a man who wants to build intelligent machines.  Jeff Hawkins, a computer programmer by background, formed a number of ideas about intelligence from a considerable amount of do it yourself reading. His interest culminated into a neuroscience research center, the Redwood Neuroscience Institute (RNI) in 2002, still alive today. Soon after he founded the company Numenta which is working on Grok – one of the first computer programs geared towards data analysis and automatic model construction to make predictions. This is the core move towards what Jeff sees as the key ingredient of intelligence à the prediction of what will happen in the world around us.

Summary: The first part of the book, written with the aid of a New York Times Science writer Sandra Blakeslee, can be seen as a story revolving around a number of core theoretical concepts. All about the functioning of the human neocortex, which together are combined into a framework of what makes us as humans intelligent. In order of appearance:

•  The idea that all or the majority of the neocortex operates in the same manner, an idea entitled by Hawkins to Vermon Mountcastle in a paper “Organizing Principe for Cerebral Function” published in 1978. In other words the processing of inputs, feedback, and outputs regardless goes exactly in the same manner, as patterns on a temporal and spatial scale, regardless of whether we are talking about sight, hearing, tasting etc.. ”In the words of Jeff and Blakeslee, “Brains are pattern machines. It’s not incorrect to express the brain’s functions in terms of hearing or vision, but at the most fundamental level, patterns are the name of the game. No matter how different the activities of various cortical areas may seem from each other, the same basic cortical algorithm is at work. The cortex doesn’t care if the patterns originated in vision, hearing, or another sense. 

•  The brain stores memories as patterns or sequences in what Jeff calls an auto-associative manner. All memories can be conceptualized  as patterns of sensory input (and sometimes output), which are “soft-coded” as the whole sequence can be recalled just by obtaining partial or distorted inputs. If you achieve only part of a sensory input you can automatically complete the sequence from memories, going both back and forth on that memory or even to other memories. The way this occurs in the human brain is different than in a computer memory which stores data as individual data blocks, or is “hard-coded” where the relations of sequences need to be coded into the machine step by step. If the input would diverge only slightly – let alone substantially - from the computer programming the machine can’t make any sense from it, while a human has this possibility.

•  The memories we store are invariantly represented, which means that despite a vast amount of changes in perception such as holding an object in a million different ways or transforming it, it can still be perceived as that object. If you turn a cup upside down you can still recognize it as a cup, while a computer does not do this automatically. Even though you see a dog species that you have never seen before, you can still recognize it as a dog, or a composition you have never heard before you can still classify its musical style etc. 

•  The brain constantly predicts what is happening around it using its invariant auto-associative memory structure, which is what makes us intelligent. As you act in the world around you the sensory inputs you receive will fire up sequences that predict what the next experience will be. The expected output is subsequently measured up with reality resulting in a match or an unexpected experience.

“Your brain makes low-level sensory predictions about what it expects to see, hear, and feel at every given moment, and it does so in parallel. All regions of your neocortex are simultaneously trying to predict what their next experience will be. Visual areas make predictions about edges, shapes, objects, locations, and motions. Auditory areas make predictions about tones, direction to source, and patterns of sound. Somatosensory areas make predictions about touch, texture, contour, and temperature. “Prediction” means that neurons involved in sensing your door become active in advance of them actually receiving sensory input. When the sensory input does arrive, it is compared with what was expected…if your expectations about the door are violated, the error will cause you to take notice…Prediction is not just one of the things your brain does. It is the primary function of the neocortext, and the foundation of intelligence.”

The second part of the book sheds some detail on how these concepts of memory sequence storage and prediction work physically work in the neocortex. Fortunately, modesty is maintained mentioning that this is only one possible theory among many others. Given the complex nature, a topic to be dived into at a later date, once my brain science knowledge is more up to date.

Overall: The book is quite good to get a concise description of a number of concepts on intelligence and how these relate to brain functioning. It is written in a way that makes sense at a logical level (at least to me). The authors take a number of steps into the working of the brain in the second part which is reasonably accessible although requiring a second and third read (the authors explain most required prior knowledge like the working of neurons along the way). However, given the choice to omit technical parts fairly incomplete which makes it quite confusing to the amateur reader on brain functioning.

Verdict: Good read, useful concepts on how we plausibly function in our world, time well spent.