There is one key difference between dream thoughts and our thinking while awake. One of the lessons we learn in life is that certain actions, even thoughts, are not permissible in the real world. For example, we learn that we cannot immediately fulfill our desires. There are rules against grabbing the money in the cash register at a store, and constraints on interacting with a person to whom we may be physically attracted. We also learn that certain thoughts are not permissible because they are culturally forbidden. As we learn professional skills, we learn the ways of thinking that are recognized and rewarded in our professions, and thereby avoid patterns of thought that might betray the methods and norms of that profession. Many of these taboos are worthwhile, as they enforce social order and consolidate progress. However, they can also prevent progress by enforcing an unproductive orthodoxy. Such orthodoxy is precisely what Einstein left behind when he tried to ride a light beam with his thought experiments.
Cultural rules are enforced in the neocortex with help from the old brain, especially the amygdala. Every thought we have triggers other thoughts, and some of them will relate to associated dangers. We learn, for example, that breaking a cultural norm even in our private thoughts can lead to ostracism, which the neocortex realizes threatens our well-being. If we entertain such thoughts, the amygdala is triggered, and that generates fear, which generally leads to terminating that thought.
In dreams, however, these taboos are relaxed, and we will often dream about matters that are culturally, sexually, or professionally forbidden. It is as if our brain realizes that we are not an actual actor in the world while dreaming. Freud wrote about this phenomenon but also noted that we will disguise such dangerous thoughts, at least when we attempt to recall them, so that the awake brain continues to be protected from them.
Relaxing professional taboos turns out to be useful for creative problem solving. I use a mental technique each night in which I think about a particular problem before I go to sleep. This triggers sequences of thoughts that will continue into my dreams. Once I am dreaming, I can think—dream—about solutions to the problem without the burden of the professional restraints I carry during the day. I can then access these dream thoughts in the morning while in an in-between state of dreaming and being awake, sometimes referred to as “lucid dreaming.”5
Freud also famously wrote about the ability to gain insight into a person’s psychology by interpreting dreams. There is of course a vast literature on all aspects of this theory, but the fundamental notion of gaining insight into ourselves through examination of our dreams makes sense. Our dreams are created by our neocortex, and thus their substance can be revealing of the content and connections found there. The relaxation of the constraints on our thinking that exist while we are awake is also useful in revealing neocortical content that we otherwise would be unable to access directly. It is also reasonable to conclude that the patterns that end up in our dreams represent important matters to us and thereby clues in understanding our unresolved desires and fears.
The Roots of the Model
As I mentioned above, I led a team in the 1980s and 1990s that developed the technique of hierarchical hidden Markov models to recognize human speech and understand natural-language statements. This work was the predecessor to today’s widespread commercial systems that recognize and understand what we are trying to tell them (car navigation systems that you can talk to, Siri on the iPhone, Google Voice Search, and many others). The technique we developed had substantially all of the attributes that I describe in the PRTM. It included a hierarchy of patterns with each higher level being conceptually more abstract than the one below it. For example, in speech recognition the levels included basic patterns of sound frequency at the lowest level, then phonemes, then words and phrases (which were often recognized as if they were words). Some of our speech recognition systems could understand the meaning of natural-language commands, so yet higher levels included such structures as noun and verb phrases. Each pattern recognition module could recognize a linear sequence of patterns from a lower conceptual level. Each input had parameters for importance, size, and variability of size. There were “downward” signals indicating that a lower-level pattern was expected. I discuss this research in more detail in chapter 7.
In 2003 and 2004, PalmPilot inventor Jeff Hawkins and Dileep George developed a hierarchical cortical model called hierarchical temporal memory. With science writer Sandra Blakeslee, Hawkins described this model eloquently in their book On Intelligence. Hawkins provides a strong case for the uniformity of the cortical algorithm and its hierarchical and list-based organization. There are some important differences between the model presented in On Intelligence and what I present in this book. As the name implies, Hawkins is emphasizing the temporal (time-based) nature of the constituent lists. In other words, the direction of the lists is always forward in time. His explanation for how the features in a two-dimensional pattern such as the printed letter “A” have a direction in time is predicated on eye movement. He explains that we visualize images using saccades, which are very rapid movements of the eye of which we are unaware. The information reaching the neocortex is therefore not a two-dimensional set of features but rather a time-ordered list. While it is true that our eyes do make very rapid movements, the sequence in which they view the features of a pattern such as the letter “A” does not always occur in a consistent temporal order. (For example, eye saccades will not always register the top vertex in “A” before its bottom concavity.) Moreover, we can recognize a visual pattern that is presented for only a few tens of milliseconds, which is too short a period of time for eye saccades to scan it. It is true that the pattern recognizers in the neocortex store a pattern as a list and that the list is indeed ordered, but the order does not necessarily represent time. That is often indeed the case, but it may also represent a spatial or higher-level conceptual ordering as I discussed above.
The most important difference is the set of parameters that I have included for each input into the pattern recognition module, especially the size and size variability parameters. In the 1980s we actually tried to recognize human speech without this type of information. This was motivated by linguists’ telling us that the duration information was not especially important. This perspective is illustrated by dictionaries that write out the pronunciation of each word as a string of phonemes, for example the word “steep” as [s] [t] [E] [p], with no indication of how long each phoneme is expected to last. The implication is that if we create programs to recognize phonemes and then encounter this particular sequence of four phonemes (in a spoken utterance), we should be able to recognize that spoken word. The system we built using this approach worked to some extent but not well enough to deal with such attributes as a large vocabulary, multiple speakers, and words spoken continuously without pauses. When we used the technique of hierarchical hidden Markov models in order to incorporate the distribution of magnitudes of each input, performance soared.
CHAPTER 4
THE BIOLOGICAL NEOCORTEX
Because important things go in a case, you’ve got a skull for your brain, a plastic sleeve for your comb, and a wallet for your money.