Has artificial intelligence finally arrived? This was the question that most people were asking after the book On Intelligence by Jeff Hawkins and Sandra Blakeslee was published and Hawkins' theories on the functioning of the neocortex were mathematically modeled by Dileep George, a Stanford graduate. The neocortex was modeled as a Bayesian network. The model (hierarchical temporal memory, or HTM) developed by the team at their company, Numenta, has proven to be very useful, particularly in visual pattern recognition tasks.
The question remains unanswered. Is this a great step towards making artificial intelligence a reality? Let me give you my ideas.
Problem analysis and decision making form the two most important aspects without which intelligence cannot be defined. Consciousness forms the backbone of these aspects, the existence of which in humans is responsible for their intelligent behavior. And talking about consciousness, the region in the brain that is responsible for it is the thalamus. The thalamus is the center of consciousness. All the sensory information from the sense organs (except the nose) reaches the neocortex only after passing through the thalamus. It has been suggested that the thalamus acts like an active blackboard on which the neocortex writes and reads data to analyze a problem and make a hypothesis. Feedback between the thalamic nuclei and the cortical neurons is responsible for this.
The cortex is just the data processing center of the brain that processes sensory data much efficiently, whereas the thalamus is the center of intelligence since it establishes consciousness. This vital region and its functionalities have not been considered at all in the book and the model developed by Numenta. So, artificial intelligence remains unlit yet.
There is one more issue. The cortical algorithm is believed to be universal and sense independent. That is, the algorithm that the cortex uses to process visual data is the same as that it uses to process auditory data. I highly doubt that HTM could do this. One can argue that the thalamus does the preprocessing and de-noising of the data, but the thalamus is a single unit that does this to all forms of data.
The functions of the thalamus will comprise a most important part of artificial intelligence and it cannot be ignored. Except for the issues discussed above, the book provides many interesting facts about the human brain and its complexity. The data processing architecture proposed by Numenta is quite good in performance with visual pattern recognition tasks.