Thousand Brains, Cortical Columns, Frames of Reference, Unified Theory of Information Processing Structures, the Future of AI and all that Jazz

“Charles Darwin was unusual among scientists in having the means to work outside universities and without government research grants. Jeff Hawkins might not relish being called the Silicon Valley equivalent of a gentleman scientist but—well, you get the parallel. Darwin’s powerful idea was too revolutionary to catch on when expressed as a brief article, and the Darwin-Wallace joint papers of 1858 were all but ignored. As Darwin himself said, the idea needed to be expressed at book length. Sure enough, it was his great book that shook Victorian foundations, a year later. Book-length treatment, too, is needed for Jeff Hawkins’s Thousand Brains Theory.”

--- Hawkins, Jeff. A Thousand Brains (p. vii). Basic Books. Kindle Edition. [1]

To this insight from Richard Dawkins (the author of the 1976 book “The Selfish Gene”), I would add the book by Prof. Mark Burgin [2], “Theory of Information. Fundamentality, Diversity and Unification. World Scientific Publishing, Singapore.) which provides a mathematical framework for explaining the autopoietic nature of information processing structures that have evolved over millennia and pave the path for a new generation of digital autopoietic machines that brings the missing “I” into “AI”

As Stanislas Dehaene [3] points out “Every single thought we entertain, every calculation we perform, results from activation of specialized neuronal circuits implanted in our cerebral cortex. Our abstract mathematical constructions originate in the coherent activity of our cerebral circuits, and of the millions of other brains preceding us that helped shape and select our current mathematical tools.”

Individual thoughts, concepts, and the number sense arising from neural activity are composed into higher level complex structures which rise through our consciousness and communicated through our cultures to propagate via a multitude of individual brain structures that use them and even refine them. Resulting mathematical structures are now allowing us to decipher the way brain structures function aided by the experimental observations using positron emission tomography and functional magnetic resonance imaging (fMRI) experiments on how brain codes our thoughts.

The book by Jeff Hawkins details the results of his research on how the brain manages intelligence and provides a framework for how it is accomplished. The living organisms have developed a way to represent the knowledge about themselves and their interactions with the environment and store it as distributed and connected networks and subnetworks of thousands of complimentary models. This is accomplished with cortical columns that represent the processing of sensory inputs, observed features along with the associated locations and movements. The reference frames are used to store all knowledge acquired through the observations. It is interesting to note that the observations are not only generated by the five senses but also are generated by thought, if you will, the sixth sense. “Thinking is a form of moving. Thinking occurs when we activate successive locations in reference frames.” This enables our mathematical, language and reasoning skills.

As Hawkins puts it “reference frames are not an optional component of intelligence; they are the structure in which all information is stored in the brain. Every fact you know is paired with a location in a reference frame. To become an expert in a field such as history requires assigning historical facts to locations in an appropriate reference frame. Organizing knowledge this way makes the facts actionable. Recall the analogy of a map. By placing facts about a town onto a grid-like reference frame, we can determine what actions are needed to achieve a goal, such as how to get to a particular restaurant. The uniform grid of the map makes the facts about the town actionable. This principle applies to all knowledge.”

There are three important conclusions that can be drawn from Hawkins’s book:

1.      Structures play a key role in information processing. There are two types of structures. First, those that process sensory information, extract features based on specific goal and create a model of the observations in the form of a map associating locations and movement. Second, there are structures that model relationships, behaviors and global knowledge that associates or connects local knowledge structures, and their behaviors.

2.      Based on our current understanding, “the neocortex is the organ of intelligence. It is a napkin-size sheet of neural tissue divided into dozens of regions. There are regions responsible for vision, hearing, touch, and language. There are regions that are not as easily labeled that are responsible for high-level thought and planning. The regions are connected to each other with bundles of nerve fibers. Some of the connections between the regions are hierarchical, suggesting that information flows from region to region in an orderly fashion like a flowchart. But there are other connections between the regions that seem to have little order, suggesting that information goes all over at once. All regions, no matter what function they perform, look similar in detail to all other regions.” It is important to note that the neocortex (new mammalian brain) builds higher level structures of intelligence on the old (reptilian legacy) brain functions and adopting them to manage shared knowledge across the system.

Figure 1: Structures as information processing machines.

3.      The neocortex structure uses domain independent schema and process-evolution mechanisms. Although it is functionally divided into dealing with different features corresponding to different domain models such as processing language, mathematics, sensory inputs etc., the neocortex accomplishes all these functions using the same structure to model and process information. In essence, the structures of intelligence use schema and operations on them to represent knowledge, which are domain independent. The models represent knowledge in the form of functions in the nodes, their behaviors and their connections to other nodes sharing information.

These observations support the conjecture that structures play a key role in information processing and the autopoietic[1] nature of biological systems stems from using the information processing structures to model their interactions between components within, and their interactions with their external world. The structures contain the knowledge of the “self” and its relationships to the external world exhibiting the oneness of the computer and the computed.

This brings us to the theory of autopoietic computing systems and the theory and practice of triadic automata [4, 5]. The theory, based on the unified theory of information processing dealing with named sets, knowledge structures, cognizing agents and structural machines [6], provides a framework for modeling autopoietic information processing structures both to model how brain structures manage

Figure 2: The fundamental triad representing knowledge

It has the visual representation shown in Figure 2. At the lowest level, the data elements are generally represented by a key, value pair. They are domain dependent and represent some knowledge about the domain. For example, glucose level in a person’s body has a value. Similarly, Sugar level of the person has a value. At the next level, some of the data elements have a relationship to other elements. Some elements change depends on the changes of other elements. For example, the risk of diabetes of a person depends on the levels of sugar and insulin of that person. This information provides a model that represents the knowledge structure and changes in the knowledge structure provides new information.

The knowledge structure bears similarity to the cortical column [1] with observed features, their impact on the objects with their locations, relationships and behaviors. Micro-knowledge structures as named sets and their composition into macro knowledge structures provide a model to represent our knowledge about the world.

1.      The autopoietic machines proposed [4, 5] provide a framework to use current IaaS (infrastructure as a service) and the PaaS (Platform as a Service) platforms as the means to provide local computing structures that provide sensory data processing using both digital and neural computing algorithms. In addition, a cognitive overlay structure provides global information sharing and higher level functional and non-functional behavior execution of the overall system based on the systemic goals and constraints. This is analogous to the neocortex utilizing the old brain functions. Figure 3 shows an implementation using current IaaS and PaaS structures to manage an application workload with a cognitive overlay.

Figure 3: Autopoietic Machine Implementation

3.      The schema representing the knowledge structures and operations on them provides a framework for implementing higher level intelligence that uses local knowledge structures and their evolution in real time to predict options for future evolution and required actions to implement them proactively [6]. Figure 4 shows the framework.

Figure 4: A framework for cognitive reasoning based on history

Jazz is an example of a complex adaptive system where individual members participate with their local talent that results in a surprising global harmony. “All that Jazz” is a metaphor that describes the thesis, antithesis and the synthesis involved in the process. Current understanding of information processing structures and their revolution is undergoing a synthesis and the Jazz metaphor seems very appropriate.

The purpose of this post is to stimulate conversation on the two books that have influenced my thinking of intelligence – how biological systems have evolved their intelligence and how we can infuse intelligence and create digital autopoietic automata that go beyond current digital automata in creating and managing information processing structures.

The book “Thousand Brains” also touches upon the nature of consciousness, and the culture that go beyond individual consciousness to a group level. It alludes to the issues with the individual’s model based reality and the external reality in the physical world such as politics, religion etc. Although each individual develops a model of reality based on one’s experience which appears to be self-consistent, it may not be moored to external reality in the real world where groups of individuals interact. The group behaves as a complex adaptive system with emergent properties that could lead to strange group behaviors that are unexpected. The behavior resulting from the self-referential circularity not moored to external reality gives rise to many of the issues we see today and are discussed in the book. Extending these arguments to “Capitalism” and “Marxism” debate would be a very fascinating exercise in itself.

Perhaps, the digital automata could provide a means to differentiate between fake information and real information based on the history it captures and provide a means to calibrate individual’s model-based reality to external reality in the physical world about observed knowledge without bias using deep reasoning discussed above.

References:

     [ 1]     Jeff Hawkins, (2021). “A Thousand Brains: A New Theory of Intelligence.” Basic Books, New York.

     [ 2]     Burgin, M. (2010) Theory of Information. Fundamentality, Diversity and Unification. World Scientific Publishing, Singapore. https://www.worldscientific.com/doi/pdf/10.1142/7048

     [ 3]     Dehaene, Stanislas. (2011). “The Number Sense: "How the Mind Creates Mathematics" Revised and Updated. Oxford University Press. Kindle Edition. P. 15.

     [ 4]     Burgin, M., Mikkilineni, R. and Phalke, V. Autopoietic Computing Systems and Triadic Automata: The Theory and Practice, Advances in Computer and Communications, v. 1, No. 1, 2020, pp. 16-35  

     [ 5]     Burgin, M. and Mikkilineni, R. From Data Processing to Knowledge Processing: Working with Operational Schemas by Autopoietic Machines, Big Data Cogn. Comput. 2021, v. 5, 13 (https://doi.org/10.3390/bdcc5010013 )

     [ 6]     Mikkilineni, R.; Burgin, M. Structural Machines as Unconventional Knowledge Processors. Proceedings 2020, 47, 26. https://doi.org/10.3390/proceedings2020047026

[1] Autopoietic structure is a system capable of regenerating, reproducing and maintaining itself by production, transformation and destruction of its components and the networks of processes downstream contained in them.