The Memory Prediction Framework

On Intelligence by Jeff Hawkins is the best book I’ve read in, oh, 5 or 10 years. It introduces a new unifying theory of how the human brain works and extrapolates it into a vision of how we can build smart machines (ones that unlike AI and neural networks exhibit true intelligence). I’m no expert in AI or neurobiology, but I found the book very exciting and very readable. It lays out how little we know about the workings of human intelligence, proposes a simple and elegant new theory to explain it and then steps through example after example of how the theory makes sense for all kinds of hard to grasp concepts like intelligence, perception, creativity and consciousness. The new theory, called the memory prediction framework, describes how the cortex, the outermost layer of our brain, acts as a giant hierarchical memory bank. We use it to store all the events we experience around us as abstract patterns. At the same time, we use previously stored patterns to constantly predict what to expect next. This leads to the book’s core stipulation that “intelligence is the capacity of the brain to predict the future by analogy to the past”.

You can read the prologue here. I highly recommend this book.