I recently listened to an audiobook format of Jeff Hawkins’ 2004 book “On Intelligence.” In this essay I will try to review some of the key concepts from the book. In a future essay I will relate these concepts directly to holistic tracking and awareness.
Jeff Hawkins has focused much of his life on understanding computers, the human brain, and searching for an overarching theory of intelligence. In his book titled “On Intelligence,” he presents such a theory, supports it with research, and explains how this theory can help design truly intelligent machines. Hawkins does a great job of explaining the fundamental differences between computers and brains and the reasons why the most advanced computers of today still cannot do certain tasks that a three year old human does with ease.
Some of the key arguments that Hawkins makes can be summarized as follows:
1. Behavior is not the best way to measure intelligence.
2. The type of intelligence that he is looking at is an aspect of the neocortex (the most recently evolved part of the brain).
3. The main function of the neocortex and the basis of intelligence are memory and prediction.
4. Intelligent machines should be built by using the principles of the neocortex otherwise it will be difficult to make them truly intelligent.
Hawkins’ first point is that intelligence should not be measured by behavior such as the classic test of artificial intelligence called the Turing Test. Many aspects of having intelligence do not require action and behavior by itself can be very misleading. Instead, Hawkins’ is interested in the type of intelligence that humans take for granted but that has been very difficult to achieve in computer programs and robots. He is interested in how humans understand the world around us, learn patterns, and make meaningful predictions effortlessly without getting bogged down in millions of calculations.
Hawkins focuses almost exclusively on the most recent part of the brain known as the neocortex. While I might normally criticize this reductionism I feel that he justifies his focus and it works with his narrow definition of intelligence. One of the most interesting points he makes about the neocortex is the uniformity of it in appearance and structure. Hawkins’ cites the work of neuroscientist Vernon Mountcastle as pointing to the simple yet overlooked possibility that all of the neocortex is performing the same basic operation. He calls this the “single cortical algorithm.” This structure is seemingly redundant yet extremely flexible and functional. Unlike computers, it allows us to take incomplete information, recognize patterns, classify experience hierarchically, think creatively through analogy, learn language, and make useful predictions. Unlike a pre-programmed computer or robot this cortical architecture is specifically set up to learn, adapt, and function in an unknown and novel environment.
The single cortical algorithm does four main things:
1. It stores sequences of patterns
2. It recalls patterns in an auto-associative way
3. It stores patterns in invariant forms
4. It organizes stored patterns in a hierarchy
“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.” I must admit I was very stoked with how Jeff Hawkin’s emphasized the primacy of pattern recognition, memory, and prediction in his book. These are three attributes central to much of what I do and this framework has a lot of useful applications especially in tracking and awareness. In our daily life our brain is constantly making predictions about the world around us. These predictions are based upon patterns that we have learned through experience and store in our memory. As a child you learn how different objects respond to your touch, you store these experiences as patterns, and by the time you are an adult you unconsciously make predictions about how much force is necessary to push open a door. When everything goes according to subconscious predictions the tendency is not to notice. However, if something does not meet your prediction it is quickly brought to your conscious attention. If the door is rigged to be heavier you will notice because your prediction about how much force is necessary to open it will not match up.
In the last part of his book Jeff Hawkins explains the steps necessary to create intelligent machines based on the single neocortical algorithm and he delves into some of the ethics surrounding this field. As a non-computer scientist I still found this part of the book easy to understand and extremely interesting. His arguments were compelling and gave me a much better understanding of the hurdles faced by artificial intelligence. It also helped me broaden my conceptions of AI beyond the influence of Hollywood and popular media. It is amazing how the popular imagination and the worldview of the industrial age have had such a huge influence on the trajectory of AI research.
Overall, I truly enjoyed this book. It was thought-provoking, well-argued, and I look forward to synthesizing the key ideas and incorporating them into my approach.
For more information on Jeff Hawkins: