Sunday, January 24, 2010

On Consciousness

In On Intelligence (Times Books, October 2004), Jeff Hawkins sets forth a new theory of intelligence based on a concept called hierarchical temporal memory, or HTM. While HTM itself is not new, Hawkins’ application of HTM theory to the brain moves us significantly forward in understanding the human mind.

HTM is a method for recognizing, storing, and responding to inputs that works by organizing those inputs into repetitive hierarchical patterns. It requires only that those inputs be both hierarchical and temporal in nature. Hawkins attributes to HTM the ability of the cortex to learn and adapt to a wide range of very complex inputs, and mounts a very strong argument to support this idea.

If you are interested in the brain and mind and have not yet read On Intelligence you should do so. The basic concept of HTM is remarkably simple, so much so that Hawkins has already moved beyond theory to implementation in the form of a new software venture, Numenta, Inc., which he founded in 2005 with partners Dileep George and Donna Dubinsky. Already in active use by a number of development partners, Numenta’s HTM tools and technology are nothing short of revolutionary. They have already begun to redefine how we think of computer software, particularly in the realm of artificial intelligence. Both in On Intelligence and in more recent conference presentations, available on YouTube, Hawkins says that his interests and efforts are now focused mostly on HTM. His hope is that such efforts will ultimately yield a full understanding of mind, as a matter of incremental achievement.

Here I disagree. Hawkins posits HTM as a repetitive feedback loop where essentially the only divergent activity is due to the appearance of novel inputs, or, “Is it new?” What’s missing is the concept of importance, or, “Is it important?” Why importance? The answer is definitional: unlike a software program, which exists at the whim of its developer, a living creature is an independent complex system that exists separately from its environment, an environment with which it must interact in order to survive. Novelty alone is not enough in this case; the very existence of the organism is dependent upon its ability to categorize and associate inputs and input patterns in accordance with their relative importance to its survival.

Like HTM itself this idea is an excellent fit with the biology of the brain. The hypothalamus, common to all vertebrate brains, is the likely main repository of this determination, acting in coordination with the cortex via complex pathways that flows mainly through the thalamus. HTM is at work here, I believe, and most likely originated in this area, but the precise method whereby the activities of the hypothalamus and the cortex are combined and coordinated involve other mechanisms as well, including the sleep cycle common to all advanced brains.

It also fits well with our understanding of brain evolution. “Is it important?” comes first, as in the very simple tropic behaviors of the simplest of living organisms. This is the province of the older, more primitive brainstem and mid-brain region. “Is it new?” represents a quantum leap forward, as it signals the emergence of learned behavior, a significant advance in the ability of living organisms to interact with their environment in increasingly diverse and complex ways, an advance that culminates in the development of human behavior. This is the province of the cortex.

But human culture is based on more than just learned behavior. It is also based on the concepts of self and self-consciousness. Humans don’t just think, they think about the fact that they can think. This isn’t something that seems likely to emerge from HTM alone, but it does seem likely to emerge from the combination of “Is it new?” and “Is it important?” described here.

Where the cortex is responsible for thinking and behavior, the mid-brain is responsible for expressing and satisfying the organism’s basic needs, needs such as hunger and thirst, pain and pleasure, and the need to reproduce.  These are all basic attributes of selfness. Their existence defines the self, and the combination of HTM and the mid-brain/cortex relationship described here make the emergence of consciousness not only possible but likely.

Like HTM this concept should be susceptible to development as software. Unlike HTM alone, however, where questions of self and self-consciousness are irrelevant, these questions will soon loom large. At one extreme:  efforts to create software based as closely as possible on the human brain, built around the same basic needs as human beings; at the other, software formulated around entirely different basic needs, with results we can barely imagine.

5 comments:

  1. excellent post! Very few people have taken this view till now. I too have been working on such a project.

    ReplyDelete
  2. You are right! me too working on similar project.

    ReplyDelete
  3. There has been an awakening on Earth for the last 40 years. The Gulf Oil Mistake is the worst ever & I fear now for the animals, birds and oh yeah too many of us now. 2012 is approaching...communicating...vibrationally & verbally.

    ReplyDelete
  4. I applaud your conclusion that the HTM algorithm is lacking the capacity for subjective analysis. While it does to some extent determine the importance of stimuli to recognize specific features and patterns, it stands as a platform of almost pure observation. I would however like to direct your attention to the amygdala rather than the hypothalamus when searching for a way to analyze importance. Its primary function is to process emotional reactions to stimuli, most notably fear. I believe that modeling the behaviors of the amygdala and how it interfaces with the attention, reward, and emotional systems will lead to a better understanding of how to integrate subjective analysis into HTM networks.

    ReplyDelete
  5. I think the ultimate goal goal of Jeff Hawking is to build truly intelligent machines. So his theory of HTM (if it is right) is enough. I don't think understanding all the other brain structures would result in a big improvement: first, it is very very complicated. So it would take a long time to understand. And as a application designer, I prefer to have intelligent rather that self-conscient algorithms. For practical and ethical reasons.

    ReplyDelete