Discussion 9C: The River of the Mind

The tremendous amount of introductory material that the reader has had to endure throughout this dialog should give testimony to the fractured and fragmented state of neuroscientific research as it exists today. It seems within the academic community as if there are a dozen theories for every research topic, and there are so very few settled issues.

By starting with fundamentals and first principles, the intention of the Organon Sutra dialog is to establish a concrete foundation which supports an engineered path to an artificially intelligent agent. But there is still a lot of open territory to be covered between those first principles and that unified approach to artificial intelligence, and so the interested reader, having demonstrated heroic endurance to this point, must remain patient but engaged.

At the same time the dialog will be asking the reader to change perspectives regarding much of what is considered to be “intelligent behavior” as it maps that new world of artificial intelligence the Organon Sutra is designed to convey. This changed perspective will require leaving the familiar surroundings of known landmarks by crossing the far horizon along new paths and navigational references.

In The Society of Mind, Marvin Minsky attempted to bridge our understanding of the observed behavior of the human animal to the known neurophysiological organization of the brain with a brilliantly conceived model, which he describes as a society of agents, who like their human counterparts, all possess individual agendas and influences on other agents, but unlike their human counterparts, possess no intelligence on their own.

Although strikingly intuitive, this meme was but a single bridge, lacking the entire map that details the enormous unknown gray area between the relatively high level behavior of his mindless agents and the low level behavior of individual neurons. The formalisms of massively asynchronous assemblies were developed to entirely bridge this gray unknown, but in contrast to the intuitive nature of Minskys’ gifted essay, these formalisms are not so intuitive.

The formalisms of MAA’s begin at a conceptual level below that which Minsky has characterized in his Society of Mind, and so we will consider this a conceptual Community of Mind. These formalisms give us a framework to establish models which describe the behaviors of massively asynchronous assemblies, from the unit level and building up to a successive complexity on the order of the overall behavior of our (intelligently adapting) agent.

In introducing these models, the dialog will develop two new terms (among a multitude of others) which will shape the entire conversation on massively asynchronous assemblies. Those two terms are unit interconnectivity and unit intraconnectivity, and it is the contention of this dialog that Nature herself has discovered that the first secret to the emergent nature of intelligence lies in understanding the confluence between interconnected and intraconnected patterns of units in massively asynchronous assemblies, whether they be implemented as biological neurons, synthetic neural networks, or digital computer code.

Now, to see how these two concepts express themselves in a self-organizing entity, we need to learn how to visualize the activities of our (massive) asynchronous assemblies in a manner that is somewhat different than the conventional perspective used in academia and many research circles. The traditional outlook, where the conversation revolves around the various modalities in the excitation being exchanged between the connections of a clutch of neurons, or in many cases, the connection from a single axon branch to an individual post-synaptic neuron, draws the focus away from the true behavior of biological neurons and MAA’s. These new formalisms are intended to boresight the focus back to those intrinsic behaviors which form the foundation of artificially intelligent agents.

Certainly, the fundamentals of the singular synapse are important in understanding neural behavior, but overall, these vagaries influence the aggregate behavior of larger populations of neurons in abstractable ways, and it is the visualization of this predictable, aggregate behavior that can enlighten the student of massively asynchronous assemblies and bottom-up design. This will allow for a comprehension of the first and second order dynamics that occur throughout an entire MAA or biological brain, which we ultimately desire to develop into the emergent behaviors we are seeking.

So in order to get beyond traditional hierarchical thinking and adopt a truly bottom-up design methodology, we must redefine our very concept of “processing”, at the very bottom of our design.

The key to this, from the very beginning, is to abandon the concept of an individual neuron or asynchronous unit as conducting the “processing” of a bit of information. Instead, we must adopt a conceptualization envisioning that, at the singular unit level, all synaptic activity is seen as merely the signaling of a participation in some processing that is occurring at a multi-unit level. If we bring our conceptualization of the “bit of information” down to the synaptic level, we will become forever lost in the circular maze of hierarchical illusions. This “new-think” in bottom-up design is crucially important. This sentiment was aptly characterized by Konrad Lorenz, in his book Behind the Mirror, when he stated that “the isolated report of the sensory cell is, in principle, always ambiguous. One cannot identify an individual star that shines through a small gap in the clouds; only when there is a larger area of clear sky visible with a number of stars in it is one in a position to compare the pattern one sees with the stellar pattern one knows.”

The definitive “bits of information” which are being “processed” in an MAA are actually to be found in the ever shifting dynamic patterns of activation which the Organon Sutra will define as the locus of processing. There is no hierarchy of top level and lower level patterns of activation. It is at the level of a locus of processing that the concepts of interconnected and intraconnected patterns of activation will provide the needed perspective when defining the source and target of this conceptual “processing”. But alas, as has been the lament throughout this entire dialog, more definitions will need to be introduced.

This is a very fundamental change in engineering perspective, one which must be adopted from the onset, but it will put the bottom-up engineer in a better frame of thinking to visualize the ever-increasingly complex patterns of behavior that will be part of the tapestry of massively asynchronous assemblies. Top down hierarchical thinking forces one to insist that there be some “information” contained in the activity of a single synapse, which in the case of much of the yet-to-be-objectified signaling in a MAA or biological brain, puts the cart before the horse.

As the design principles of the Organon Sutra unfold (there is so very much yet to be introduced), the interested reader will come to understand that if we force the activity at the synaptic level to be defined in terms of “information”, which by its nature implies the prior abstraction of context, there will be no way for our artificial agent to perform the abstraction by itself, in its effort to accomplish the necessary self-programming which will ignite intelligence.

The desire here is to yank the thinking of the bottom-up engineer totally away from the alluringly seductive mindset of the brain as a specialized computer system, organized along the design paradigms of today’s digital systems. This dialog has already demonstrated the inability of massively parallel digital systems to behave as their biological neural counterparts, and there are very few specific aspects of digital computer systems that truly apply to the actual behavior of MAA’s and biological nervous systems.

So allowing even just a few snippets of this computer metaphor to creep into your design-think will corrupt and hopelessly straightjacket any bottom-up methodology.

Going back to our biological example, at the level of individual signaling units, cortical activation occurs as the result of the learned response to synaptic activity delivered at the dendritic sites of neurons.  This delivered synaptic activity is itself the product of synaptic activity delivered to the dendrites of the efferent neurons preceding these neurons, in an endless (backward) progression. There is no beginning or end to this progression, and since a neuron is rarely triggered by the reception of a singular input, what we observe when studying neural activity is the progression of whole patterns of synaptic activations.

Traditionally, it has been difficult to interpret these activations without a clearer understanding of the varied interconnectivity between neurons, the jungle of the mind, but even then knowledge of the connections between neurons will not reveal the various signaling mechanisms employed by neural cells once these synaptic connections are established. Although neuroscientists are developing a better picture of the processes by which individual impulses (called action potentials) are exchanged throughout synaptic junctions, when we begin to examine the principles involved in the generation of action potentials themselves, the picture becomes murkier.

So how did Nature find her way out of the endless circular maze?

Let us use our new-found perspective as bottom-up engineers to peer at the behavior of biological neurons and asynchronous units in a new light, by introducing those clarifying definitions that have been needed.

We have already used the term ‘massively asynchronous assemblies’ many times, and that term shall refer to the entirety of a collection of asynchronous units that are placed in an environment within an agent mechanism capable of interacting with that environment.

We will subsequently refer to the term ‘aggregate’ as any unspecified collection of asynchronous units, from the concerted activity of just two individual units up to an entire assembly. (Although, because the term ‘assembly’ has necessarily been used in many contexts throughout this dialog up to this point, be kind if the terms ‘assembly’ and ‘aggregate’ are sometimes used interchangeably).

Further, we will refer to the term ‘unit ensemble’ as a collection of asynchronous units characterized by their mutual passive connectivity. Under this heading, there will be subgroupings such as feedback ensembles, reflexive ensembles, reciprocal ensembles, and others, all of which will be described as their functionality becomes relevant.

Similar to the definition of unit ensemble, we will define the term ‘unit coalition’ as aggregates whose groupings are characterized by mutual activation, collective units whose “post-synaptic” logic has been tuned to respond to certain patterns of activation preceding it, aggregates whose connectivity does not have to be common.

We must make these distinctions very early on in our thinking because of the fundamental hyper-connectedness of MAAs. In the cerebral cortex, the most abundant type of neuron is called the pyramidal cell, and it is estimated that a single pyramidal neuron might on average receive some 10,000 to 12,000 afferents from other neurons, and send on average over a thousand axonal branches to other neurons. If we look at the interconnectedness of large populations of perhaps billions of units, we can see that, for the sub-populations that have direct connections, at any given time not all of the connections will be active simultaneously (simultaneous activation is meant to mean neurons which fire within a rough window of cumulative refractory periods), and not all aggregates that are simultaneously active will have common connections. We must be able to discern the patterns of these selective activations within such large populations of hyper-connectedness.

Along the lines of the definitions for ensemble and coalition aggregates, this dialog will develop the paradigm that the connections in unit ensembles represents the static “river bed”, and the activity occurring in unit coalitions represents the actual flowing “river of the mind”, a paradigm that will be returned to repeatedly in the dialog, and a paradigm that will be essential in visualizing the genesis of gestalt abstraction.

Making the distinction between the mutual behavior based on unit connectivity and the behavior observed by collective activation will allow the dialog to develop concepts of first and second order assembly dynamics, thus providing a foundation for the formalisms of MAAs that will be developed, and a basic understanding of the confluence between interconnected and intraconnected patterns of units, two more terms which have yet to be defined. Geez, there is so much Mom never told us about neuroscience…

It is important to understand this separation in the locus of processing from the actual elements doing the processing, a conceptualization of the “information” being separate from the physicality of the processing units themselves, as we develop the concepts of first and second order patterns of activation in our models of MAAs. This conceptualization is weakly expressed in in neural networks, whose distributed processing separates “information” from their specific computing devices, however, neural networks do not exhibit first and second order patterns of activation. These orders of activation will be defined in due course, but the Organon Sutra cannot over-emphasize the necessity of divorcing the concept of “processing” from the behavior of individual connections between units in an MAA. Processing in its effective sense occurs only as the result of many activations within a coalition, and it is the recognition of patterns in these varying multiple activations impinging within the MAA unit that should be the focus for “learning” at the neural and MAA unit level.

The locus of processing is a conceptualization which sees a “bit of information” as comprising a singular nexus in a much larger web of knowledge being woven in the progression of multiple patterns of activation throughout a global assembly. The entire woven fabric is a dynamic thing, much as waves roll across the ocean, and to tie a single locus of processing, a single “bit of information”, to an individual asynchronous element (or neuron) is to tear the fabric of that much larger web of knowledge being woven.

(Note that the dialog has not as yet given a definitive definition of either biological neurons in the context of the Organon Sutra or asynchronous units themselves. Nor has a description been given to the functionality in the connections between units. A concise definition for our asynchronous units and their connectivity will have to wait for the discussions on the remaining fundamental precepts. Yikes! So much introductory stuff!)

So, in order to form a basic understanding of the Yin and Yang surrounding unit connectivity and unit activation in populations of assemblies, we must develop and choreograph models which synthesize those particulars and constraints that allow a visualization of the dynamics in our assembly as a whole.

If properly constructed, these models will reflect the actual tradeoffs and compromises that Nature herself discovered when evolving natural intelligence. And to develop these models, the dialog will continually turn to the one existing model of natural intelligence, the human cerebral cortex, which was introduced in discussion 9A. And although the dialog has yet to provide a complete description of the basic unit elements in our model (a very necessary step, but as has been said, a complete outline must await important contextual definitions), it is time to define some of those contextual definitions which actually characterize the “processing” of “information” being conducted at the unit level. It is so unfortunate that the computer metaphor of “information bits” has crept into the lexicon of neuroscience, because as we shall see, it is wholly uncharacteristic of true neural behavior.

The disparity in characterization between digital logic systems and biological nervous systems stems from a disparity in functionality between the basic units comprising these systems.

Digital logic devices are state devices.

Biological neuron cells are signaling devices.

Digital logic devices have states which represent an already abstracted “bit of information”, an abstraction that has been pre-determined by the logic systems’ (hierarchical) designer.

The signaling that a biological neuron demonstrates carries no pre-determined abstraction, no pre-designed “information”.

What they do convey will be the over-riding subject of the remainder of this dialog.



Copyright © 2019 All rights reserved