Author: Jeff
Hawkins (with Sandra Blakeslee)
Length: 235 pages (plus an 8 page appendix of
predictions)
What is It: A comprehensive theory on the inner
workings of the brain (memory, intelligence and creativity), including its relevance to building intelligent machines. Hawkins starts with a brief history of artificial
intelligence and neural networks, but openly challenges them as
design foundations. He then presents his memory-prediction theory as a
more comprehensive and accurate representation. He ends On Intelligence with a sketch of the future based on his model and a series of testable predictions.
What's Said: Theories of the brain should be based on
i) elapsed time, ii) feedback loops and iii) complete
architecture. The brain is not a computer; it is a memory system. It
distills incoming sequences of patterns (eg: letters of a word in a
speech) into invariant forms [snapshots of constantly changing details],
passes this delayed pattern up its hierarchical structure (eg: from letters to words in this analogy), repeats that process at each of its six levels (eg: words to phrases, phrases to sentences...), and links all those
sequences together through auto association for retrieval later (eg: remembering that speech by recalling just a word/phrase/sentence in it). Detailed predictions using this system are the essence of
intelligence. The brain predicts by combining
its knowledge of invariant forms from higher brain levels with
current details from the senses, and lower levels. This algorithm is the main function of the brain and is exactly the same everywhere
(i.e. the primary senses are fundamentally no different from each other). Correct predictions
result in understanding, incorrect predictions result in confusion.
What's True: There are 10 times as many feedback as input connections in the brain. Human intellectual
superiority is explained by size of the neocortex and its hierarchical levels of sophistication (ability to handle more abstraction, and
longer temporal patterns than animals). The Turing Test is
misguided because intelligence cannot be accurately measured by external
behavior (computational output), it is an inner brain function (understanding
does not require action). The memory-prediction framework
distinctly separates intelligence from human form and emotions (which
negates "rise of the machines" scenarios). Intelligence systems are customized machines that can excel where human intelligence is disinterested or human senses are
inadequate. Hawkins is clear and transparent in OI. He not only lays out his theory, he addresses rebuttals, and willingly points out potential flaws in the model. Even if his theory is correct however, capacity
(the brain has an estimated 8 trillion bytes of memory) and connectivity (one brain cell connects to an estimated 10 thousand other brain cells) will
still be challenges.
So What: Hawkins' theory has significant implications for the development of intelligent machines. It means that design thinking
centered on faster computing, larger memory, and behavior driven output
is absolutely misdirected. It means newer innovations like big data (inputs and
discovery) and machine learning (pattern recognition and prediction) should be
not be explored separately; they should be studied as components of a single intelligent system.
For individuals, On Intelligence provides new insights on the way humans think,
why we think like we do, and how thought is connected to itself and
behavior.
Final Word: Very Enlightening Read
