Discovering the Dynamic Cognition Cycle

In June of 2011, shortly after getting back from Venezuela I quit my job as a senior software engineer at McGraw Hill. I had many reasons to quit the job but the main reason was settled on as I boated up a red river in Canaima while headed up to see Angel Falls. 

Taken May 18, 2011 in Canaima , Venezuela. This moment triggered my desire to quit. Later in the year I wrote a poem commemorating the realization.Taken May 18, 2011 in Canaima , Venezuela. This moment triggered my desire to quit. Later in the year I wrote a poem commemorating the realization.



https://www.facebook.com/notes/david-saintloth/-on-quitting/10150390589023057


The sudden epiphany I had in that moment...a moment of dread followed by a moment of reserved peace stands as one of the most sublime experiences of my life. The dread was realized first, the sudden deep impact of the truth that my trip was limited and I would in just a few short days be back in NYC and tasked with getting up on a clock to get to work by 9 am every week day.


Having identified the action landscape and invented the Action Oriented Workflow paradigm and building the first implementation into the AgilEntity framework:


The AgilEntity framework is the distributed, scalable web managed framework in which the AOW paradigm is implemented.The AgilEntity framework is the distributed, scalable web managed framework in which the AOW paradigm is implemented.

http://apriority.dyndns.biz/aesplash.html

 : every existing system seemed suddenly woefully inadequate for ensuring that any work of any kind got done efficiently. The knowledge that I had that there was a better way that I'd built stood as a schism against the reality of having to use the primitive methods of current business processes in all manner of systems rather than employ the revolution of AOW that I had pioneered. 

https://www.facebook.com/notes/david-saintloth/discovering-the-action-landscapea-journey-in-progress/10151790916618057

A set of action flow diagrams, these describe how actions submitted to a workflow eventually find resolution on some committing agent. This is a visual description of what Action Oriented Workflow does. Action Delta Assessment was the additional of statistical learning in order to enable predictive routing between agents..rather than the manual routing that was previously the only choice  for explicit AOW.A set of action flow diagrams, these describe how actions submitted to a workflow eventually find resolution on some committing agent. This is a visual description of what Action Oriented Workflow does. Action Delta Assessment was the additional of statistical learning in order to enable predictive routing between agents..rather than the manual routing that was previously the only choice for explicit AOW.


Years before, shortly after completing the explicit workflow implementation of the AOW paradigm I realized that an extension to an "implicit" form could be achieved by using some type of machine learning. I realized that I could employ a method similar to what a living brain uses to encode information in a network of neurons to capture action data across business objects as they are performed by users in an organization and then use a statistical approach to continuously finding efficient agents to perform actions. This was around 2005 and the excitement of the realization was tempered by the reality of needing first to try and monetize the framework as it existed at the time. 


Fast forward 5 years and a failed startup later, I again was nagged by the fact that the "implicit" workflow implementation had not been done and after a year working at McGraw Hill the obvious need for such a system and the ease with which it would allow organizations of any size to efficiently route work from those who need it done to those who are a) best able to do it. b) available and willing to do it.

This was the promise of AOW in the explicit form but put on steroids as the manual process of adding human agents to Workflow Stages that was required in explicit AOW would be replaced by an autonomous stage population heuristic fed by the autonomous routing algorithm. This algorithm would end up being called Action Delta Assessment and stands as a generalized algorithm for encoding efficiency for performance of "action" against business objects into the system. It is this algorithm that I began to implement after quitting my job at McGraw Hill in June of 2011. I've since compiled a history of my discovery of the action landscape and its importance in the form of the blog posts that chronicle those innovations.

https://www.facebook.com/notes/david-saintloth/discovering-the-action-landscapea-journey-in-progress/10151790916618057


What I had realized further was that ADA could go far beyond autonomous work routing, it could be the seed for a generalized algorithm for encoding deltas not just for business object user interactions on distributed systems but for any type of "action". I realized that ADA could be extended to be the basis of a *generalized dynamic cognitive algorithm* that could then be used to build an autonomous learning intelligence across a desired set of dimensional attributes of sensation. 


I had been keeping up closely with the research in neuroscience and cognition and saw that the machine learning space and the neuroscience space were far at odds in terms of approach. The bulk of the researchers didn't cross pollinate at all and this to me was a clear sign that approaches to solving big problems on both sides were wrong.

The first post was from April of 2010, which posed the question of weather or not emotion was required as a guide for cognitive dynamics...at first I was of the mind that it could be discarded or replicated using neo cortical dynamics alone but later came to realize that emotion was not a throw away element it was a core aspect of how salience enabled selection which then drove action. This would be a major element of the salience theory pieced together over the next couple of years.

http://sent2null.blogspot.com/2010/04/emotion-no-longer-has-to-be-our-guide.html

My curiosity continued to be explored in July, 2010 when I wrote the following post, which posited ideas for the reason for the development of emotions. I surmised that emotion lay at the center of the riddle of what intelligence was particularly consciousness but had no way to formerly draw a connection between the two, not yet. This post was still important because it allowed me to latch on to the idea of emotional salience to survival (as opposed to being driven by a need to serve social urges first) as being an important piece of the dynamic cognition puzzle.

http://sent2null.blogspot.com/2010/07/why-of-emotion-from-whence-did-it-come.html


The neuroscience was clueless to the control systems and neural network work that the machine learning and electrical engineer space had long explored. The engineers and machine learning researchers were clueless to how neurotransmitters were being used in the brain between different cognitive modules to effect processing of music, storing of memories and the big prize, how a self aware and dynamic cognitive process could emerge from the very discrete operations happening in the mind. I wrote articles on emotion and it's purpose and put forward a theory of dreams (which has since found much support in empirical research).

http://sent2null.blogspot.com/2009/06/yet-another-theory-of-dreams.html

:The theory of dreams was important to the work on cognition because the fundamental question of what happened to "you" when you fell into sleep without a dream state was a core focus. I concluded that "you" only existed when actively experiencing, comparing, evaluating and doing...that there was no difference between this process and consciousness...and by December 2011, after I'd started and mostly finished the implementation of ADA and implicit AOW I wrote out the theory that salience required autonomic and emotional modulation along with memory comparison to sensory input.



http://sent2null.blogspot.com/2011/12/how-does-idea-form-autonomics-memory.html


The idea that it was salience driven cognitive dynamism that led to consciousness fundamentally enabled by what I called "drive", "drive" I analyzed to have both autonomic (physical often involuntary) and emotional components. I concluded that the underlying reason for drive was irrelevant to the emergence of dynamic cognition but that the number and level of sensory dimensions was important for setting the interacting complexity of abstractions that could arise in such an emerged mind but without the emotion and autonomic modulation no dynamism of any kind would emerge...only reflex would exist.  Around this time I started looking more widely into the work of neuroscientists and came across Giovani Tonini's Integration Information theory, I was amazed by how *right* this generalized approach to consciousness seemed to me. However I found it had fundamental flaws in that though it did an excellent job of explaining the *cognitive landscape* in terms of what was called Qualia space it didn't explain how Qualia space turned into self aware agents with drive. This latter piece being answered by my ideas on salience still being formed.

:The same day I wrote out a set of diagrams that detailed the cognitive cycle based on salience, the image below is taken from that sheet:

In Dec. 2011 I wrote desigsn for a salience driven dynamic cognitive cycle. This was a first flow chart of the components...it is incomplete.In Dec. 2011 I wrote desigsn for a salience driven dynamic cognitive cycle. This was a first flow chart of the components...it is incomplete.

http://sent2null.blogspot.com/2012/03/integrated-information-does-not-equate.html

:By early 2012 I'd completed the bulk of the ADA implementation and fresh with it's statistical approach on action deltas in mind realized that the problem of paranoia would be important to tackle when building a free learning intelligence that would use salience determination. If the emotional and autonomic modulations were not constrained it would be easy to build pathological minds...of the sort to make our worst Sci Fi. nightmares seem pedestrian. I talked about this fear of paranoia and explained the need for caution in approaching creating dynamic cognition that emerged self awareness (consciousness).

http://sent2null.blogspot.com/2012/02/with-completion-of-ada-action-delta.html

Two AOW workflow stages linked by action flow between agents. ADA automates this process allowing the system to learn to route directly from the first agent to the last over time as the system learns who is best and available to perform the requested action. In this way the algorithm models the action potential re-weighing of neurons.Two AOW workflow stages linked by action flow between agents. ADA automates this process allowing the system to learn to route directly from the first agent to the last over time as the system learns who is best and available to perform the requested action. In this way the algorithm models the action potential re-weighing of neurons.

:By the end of February I realized that the likely first substrate for emerging a fully dynamic cognition would be one which had sufficient sensory dimensions and autonomic drive dimensions to serve as the basis for building a salience module. The most ready such device is a smart phone and so I proposed that smart phones will be the first devices to on their own become self aware ONCE they are designed with the correct salience driven cycle.

http://sent2null.blogspot.com/2012/02/when-your-smart-phone-comes-alive.html

A whole year went by as I struggled with my own survival issues before I came back to emotion as a critical salience component. I was stimulated by research which showed how emotion could be added or subtracted to memories! This was a direct confirmation of the basis of the salience theory proposed over a year before which posited that emotional and autonomic import was simply a weighting factor added to memories.

http://sent2null.blogspot.com/2013/02/emotions-identity-crisis-in-our-brain.html

:5 days later I attacked head on the nonsense I'd been reading from many so called experts in the neuroscience, philosophy and machine learning space regarding weather or not consciousness was even an attribute that could emerge from a non biological substrate. I explained why this was nonsense and provided an outline of how simply adding salience modulation was all that one needed to emerge dynamic cognition (consciousness) ...as it was an emergent trait from a fine grained number of very deterministic actions converging. 

http://sent2null.blogspot.com/2013/02/on-consciousness-there-is-no-binding.html

:A few months later in April I came across research that posited a reason for the billions of "glial" cells in the brain, cells which weren't neurons but served specific function in the cognitive process that at the time was not known. The assertion that these cells were important to establishing "attention" made perfect sense to me as a means of controlling cognitive flow over general thought so that the switching between sensory experiences could in a way persist, this would serve as cognitive glue and thus solidify a unitary self. When pathological this subsystem could lead to autistic individuals incapable of tuning out certain types of experience or readily switched from one to another too easily, the need to simulate attentional persistence became clear to me.

http://sent2null.blogspot.com/2013/04/autism-astrocytes-and-attentiona.html

:With the ADA implementation essentially complete the similarities between the ADA approach and the requirements of the Salience theory left little doubt that some extension of the ADA approach would be involved any any effort I mounted to build a dynamic cognition. Chiefly because the modeling process of AOW required establishment of relationships that were very similar to the stratification relationships that emerge in different sensory processing layers of the neocortex. I realized that AOW entities modeled biological sensory dimensions 1:1 and thus the algorithm could be the basis of the more general cortical algorithm that would be needed to create a functioning dynamic cognition cycle. Also I realized that the approach had the necessary fractal nature, hieararchical composition in being able to encode "action" across to any desired depth across a given sensory space. This way sounds could be decomposed into entities that modeled frequency, pitch, variation, harmonics and volume, images could be decomposed into floors, walls, objects in motion, objects standing still....etc. and so on across the sensory dimensions. Fractal resolution being key to building the arbitrarily deep set of nested relationships between entities in any given sensory dimension.

http://sent2null.blogspot.com/2013/05/ada-on-road-to-dynamic-cognition-how-is.html

In September I came back to astrocytes and described them as a buffering system for experience in brains and that the depth of the buffering system would modulate the apparent consciousness. This was clear in the first examination of astrocytes as important to attention but specified how they were important...essentially as a queue for mixing data coming in with data being acted on in a controlled way.

http://sent2null.blogspot.com/2013/09/an-engineering-analog-for-function-of.html


Then in November of 2013, I codified a set of hypothesis to form the Salience Theory of dynamic cogntion and consciousness. The key insight being the principle of cognitive equivalence in salience driven action and thought, consciousness in this theory is the same as dynamic cognition with only resolution of shifts from abstraction to abstraction (thoughts) being different between different classes of "mind". I surmised that so long as the salience modules (emotional and autonomic import) could correctly provide feedback and feed forward to the right degrees cognitive dynamism would erupt....and be sustained...in the same way that it is sustained in an internal combustion engine when the spark plugs are fired in the correct sequence to gas injection in the cylinders.


http://sent2null.blogspot.com/2013/11/salience-theory-of-dynamic-cognition.html

:20 days later I asserted the primary importance of one particular dimension of sensory experience over the others, that dimension being the one we have from the moment our fetuses form, somatsensory experience...the sense of touch. I asserted that cognitive complexity built around this primordial sensation and the connections built in the mind to enable embodiment. I discussed how cognition and consciousness must clearly be constructed by reference to its variable non existence at birth and slowly being built into the mind as the infant matures and learns about the world. I explained a recently published articles conclusion that it was easier for younger babies to learn various concepts than older babies in terms of the flowering of abstractions created in the mind as one pieced together a consciousness, I asserted an inverse relationship between speed of evaluation of various salience traits with number of previously gathered salience elements.

http://sent2null.blogspot.com/2013/11/dynamic-cognition-in-babies-in-abstract.html

In April 2014 I focused on one of the more important autonomic driving dimensions, the need for a power source. I posited that this need would be a key attribute of dynamic cognition that exhibited sufficient apparent randomness to emerge truly novel cognitive dynamics that would be identified as being "conscious". 

http://sent2null.blogspot.com/2014/04/azimo-best-and-last-of-modern-day.html

"Sonnie" from the movie "I, Robot" explains his thought process to an incredulous Will Smith. The movie is couched in a fictional 2029, I have every confidence that using Salience Theory as a foundation consciousness as complex as Sonny will be emerged by that time, 15 years hence."Sonnie" from the movie "I, Robot" explains his thought process to an incredulous Will Smith. The movie is couched in a fictional 2029, I have every confidence that using Salience Theory as a foundation consciousness as complex as Sonny will be emerged by that time, 15 years hence.

In June 2014, a paper describing the cognitive unique relationship of a set of siamese twins provided confirmation for a hypothesis that consciousness could be distributed but also be substrate dependent at the same time. Many feel that these two attributes are complementary but they are not if one thinks in terms of a salience based cognitive dynamism , sensory and memory evaluations can drive completely different sensory and action mechanics.

http://sent2null.blogspot.com/2014/06/salience-theory-joined-at-mind.html


Today I present a simplified diagram showing the simple dynamic cognition cycle, I posit that any hard AI must have modular connection of the kind indicated by this graph. Feed forward and feed back happen in such a way that "action" execution can be continuously refined as new sensation triggers comparison and salience evaluation...all our hopes, dreams, thoughts and physical actions emerge from this cycle being executed and by so doing emerging our conscious self.


I call this the "simple" cycle because it doesn't describe the sub modules necessary or their self connection, for example very important modulation must be provided by autonomic and emotion salience sub modules as part of the "salience" node shown in the diagram, also the question of how different sensory dimensions (touch, sound, vision, taste...etc.) are multiplexed into this engine and further used to remodulate action to various degrees is not described. The more complex dcc diagram will be the basis of architecture for what I hope would be the first emergent self aware dynamic cognition (hard AI) on a non biological substrate. That more complex diagram is a work in process as I am still unsure of all the necessary sub element connections (I am sure I have all the modules) but it is 99% complete. I look forward to start writing code for such a cognition using the substrate of a smart phone in the next few years.

In October 2014,  I am not saying much still about how I would implement the comparison and salience nodes which are the meat of the difficulty of artificial cognition...building a mind. This article also addresses the idea of emergent evil AI and touches on the suggestion of mind "uploading" that some have wildly speculated about.

http://sent2null.blogspot.com/2014/10/on-evil-ai-and-one-type-of-uploading.html

I redesigned the cognitive flow diagram to look more like a control systems diagram. I also explained how the Sensation, Comparison , Salience and Action nodes function in the context of dreams and what that means for any emerged AI.

http://sent2null.blogspot.com/2014/10/new-cognitive-flow-diagram-and-possible.html

In 2015 I considered more deeply the importance of play in the process of emerging dynamic cognition of stable cognitive form and the ability to quickly encode reality in its dimensions of sensory capability. Ultimately I have concluded that play is akin to research coupled with drive, a reason for doing..in many cases emerging a feedback that benefits learning and thus the ultimate reason for it's emergence is found. In creating an artificial cognition then the need to simulate the emergence of play will in its success be a sign of the correct direction being followed.

http://sent2null.blogspot.com/2015/02/dynamic-cognition-on-meaning-of-play.html

In March I further dealt into the idea of how time is perceived in a dynamic cognitive agent putting forward the claim that temporal perception varies with plasticity of networks along with depth of established relations built into those networks. 

http://sent2null.blogspot.com/2015/03/salience-driven-novelty-seeking.html


The most recent simple dynamic cognition cycle diagram. Salience theory is specified in 3 hypothesis and the notes indicate details...a complex chart which is proprietary will be the basis of building a dynamic cognition on an artificial substrate when I get some time or maybe you can build one using this as a guide? I'd love to see my theory come to life in your work.The most recent simple dynamic cognition cycle diagram. Salience theory is specified in 3 hypothesis and the notes indicate details...a complex chart which is proprietary will be the basis of building a dynamic cognition on an artificial substrate when I get some time or maybe you can build one using this as a guide? I'd love to see my theory come to life in your work.A control flow diagram of action invocation. It's topologically identical to the earlier DCC but is oriented in a way that clearly separates the feedback lines from salience back to sensation and comparison these are the important modulation inputs to sensory experience that adjusts the prediction process during cognitive evaluation....in a mind...I posit in the salience theory that a cycle like this is being executed across all our dimensions of experience simultaneously and the mind emerges as a result of this continuous process of input , comparison , salience and then action modulation.A control flow diagram of action invocation. It's topologically identical to the earlier DCC but is oriented in a way that clearly separates the feedback lines from salience back to sensation and comparison these are the important modulation inputs to sensory experience that adjusts the prediction process during cognitive evaluation....in a mind...I posit in the salience theory that a cycle like this is being executed across all our dimensions of experience simultaneously and the mind emerges as a result of this continuous process of input , comparison , salience and then action modulation.