“Meta-phorin”

This is a poem I wrote about how the brain structures its own neural connectivity in order to produce metaphors, poetry, analogies, allegory, and the like, including through its use of semaphorin guidance molecules and such. So one can think of it as a type of meta-poetry I suppose.

“Meta-phorin”

Branches born from distant gardens
Fed by the fruits of senses streamed
Spreading out, a vibrant pattern
Crawling along those ancient trees
Toward the scents, hypnotic dance
Winding paths until they meet

Their tips begin to touch at last
Caressing as they’re intertwined
Hebbian journey, webs of gnosis
Embodied frames are now sublime

Synaptic waters flowing faster
Emotions growing, bearing passion
Creative means no longer foreign
By the meta-semaphorin

Advertisement

“Black Mirror” Reflections: U.S.S. Callister (S4, E1)

This Black Mirror reflection will explore season 4, episode 1, which is titled “U.S.S. Callister”.  You can click here to read my last Black Mirror reflection (season 3, episode 2: “Playtest”).  In U.S.S. Callister, we’re pulled into the life of Robert Daly (Jesse Plemons), the Chief Technical Officer at Callister Inc., a game development company that has produced a multiplayer simulated reality game called Infinity.  Within this game, users control a starship (an obvious homage to Star Trek), although Daly, the brilliant programmer behind this revolutionary game, has his own offline version of the game which has been modded to look like his favorite TV show Space Fleet, where Daly is the Captain of the ship.

We quickly learn that most of the employees at Callister Inc. don’t treat Daly very kindly, including his company’s co-founder James Walton (Jimmy Simpson).  Daly appears to be an overly passive, shy, introvert.  During one of Daly’s offline gaming sessions at home, we come to find out that the Space Fleet characters aboard the starship look just like his fellow employees, and as Captain of his simulated crew, he indulges in berating them all.  Due to the fact that Daly is Captain and effectively controls the game, he is rendered nearly omnipotent, and able to force his crew to perpetually bend to his will, lest they suffer immensely.

It turns out that these characters in the game are actually conscious, created from Daly having surreptitiously acquired his co-workers’ DNA and somehow replicated their consciousness and memories and uploaded them into the game (and any biologists or neurologists out there, let’s just forget for the moment that DNA isn’t complex enough to store this kind of neurological information).  At some point, a wrench is thrown into Daly’s deviant exploits when a new co-worker, programmer Nanette Cole (Cristin Milioti), is added to his game and manages to turn the crew against him.  Once the crew finds a backdoor means of communicating with the world outside the game, Daly’s world is turned upside down with a relatively satisfying ending chock-full of poetic justice, as his digitized, enslaved crew members manage to escape while he becomes trapped inside his own game as it’s being shutdown and destroyed.

This episode is rife with a number of moral issues that build on one another, all deeply coupled with the ability to engineer a simulated reality (perceptual augmentation).  Virtual worlds carry a level of freedom that just isn’t possible in the real world, where one can behave in countless ways with little or no consequence, whether acting with beneficence or utter malice.  One can violate physical laws as well as prescriptive laws, opening up a new world of possibilities that are free to evolve without the feedback of social norms, legal statutes, and law enforcement.

People have long known about various ways of escaping social and legal norms through fiction and game playing, where one can imagine they are somebody else, living in a different time and place, and behave in ways they’d never even think of doing in the real world.

But what happens when they’re finished with the game and go back to the real world with all its consequences, social norms and expectations?  Doesn’t it seem likely that at least some of the behaviors cultivated in the virtual world will begin to rear their ugly heads in the real world?  One can plausibly argue that violent game playing is simply a form of psychological sublimation, where we release many of our irrational and violent impulses in a way that’s more or less socially acceptable.  But there’s also bound to be a difference between playing a very abstract game involving violence or murder, such as the classic board-game Clue, and playing a virtual reality game where your perceptions are as realistic as can be and you choose to murder some other character in cold blood.

Clearly in this episode, Daly was using the simulated reality as a means of releasing his anger and frustration, by taking it out on reproductions of his own co-workers.  And while a simulated experiential alternative could be healthy in some cases, in terms of its therapeutic benefit and better control over the consequences of the simulated interaction, we can see that Daly took advantage of his creative freedom, and wielded it to effectively fashion a parallel universe where he was free to become a psychopath.

It would already be troubling enough if Daly behaved as he did to virtual characters that were not actually conscious, because it would still show Daly pretending that they are conscious; a man who wants them to be conscious.  But the fact that Daly knows they are conscious makes him that much more sadistic.  He is effectively in the position of a god, given his powers over the simulated world and every conscious being trapped within it, and he has used these powers to generate a living hell (thereby also illustrating the technology’s potential to, perhaps one day, generate a living heaven).  But unlike the hell we hear about in myths and religious fables, this is an actual hell, where a person can suffer the worst fates imaginable (it is in fact only limited by the programmer’s imagination) such as experiencing the feeling of suffocation, yet unable to die and thus with no end in sight.  And since time is relative, in this case based on the ratio of real time to simulated time (or the ratio between one simulated time and another), a character consciously suffering in the game could feel as if they’ve been suffering for months, when the god-like player has only felt several seconds pass.  We’ve never had to morally evaluate these kinds of situations before, and we’re getting to a point where it’ll be imperative for us to do so.

Someday, it’s very likely that we’ll be able to create an artificial form of intelligence that is conscious, and it’s up to us to initiate and maintain a public conversation that addresses how our ethical and moral systems will need to accommodate new forms of conscious moral agents.

Jude Law, Haley Joel Osment, Brendan Gleeson, and Brian Turk in Artificial Intelligence: AI (2001)

We’ll also need to figure out how to incorporate a potentially superhuman level of consciousness into our moral frameworks, since these frameworks often have an internal hierarchy that is largely based on the degree or level of consciousness that we ascribe to other individuals and to other animals.  If we give moral preference to a dog over a worm, and preference to a human over a dog (for example), then where would a being with superhuman consciousness fit within that framework?  Most people certainly wouldn’t want these beings to be treated as gods, but we shouldn’t want to treat them like slaves either.  If nothing else, they’ll need to be treated like people.

Technologies will almost always have that dual potential, where they can be used for achieving truly admirable goals and to enhance human well being, or used to dominate others and to exacerbate human suffering.  So we need to ask ourselves, given a future world where we have the capacity to make any simulated reality we desire, what kind of world do we want?  What kind of world should we want?  And what kind of person do you want to be in that world?  Answering these questions should say a lot about our moral qualities, serving as a kind of window into the soul of each and every one of us.

Predictive Processing: Unlocking the Mysteries of Mind & Body (Part VI)

This is the last post I’m going to write for this particular post-series on Predictive Processing (PP).  Here’s the links to parts 1, 2, 3, 4, and 5.  I’ve already explored a bit on how a PP framework can account for folk psychological concepts like beliefs, desires, and emotions, how it accounts for action, language and ontology, knowledge, and also perception, imagination, and reasoning.  In this final post for this series, I’m going to explore consciousness itself and how some theories of consciousness fit very nicely within a PP framework.

Consciousness as Prediction (Predicting The Self)

Earlier in this post-series I explored how PP treats perception (whether online or an offline form like imagination) as simply predictions pertaining to incoming visual information with varying degrees of precision weighting assigned to the resulting prediction error.  In a sense then, consciousness just is prediction.  At the very least, it is a subset of the predictions, likely those that are higher up in the predictive hierarchy.  This is all going to depend on which aspect or level of consciousness we are trying to explain, and as philosophers and cognitive scientists well know, consciousness is difficult to pin down and define in any way.

By consciousness, we could mean any kind of awareness at all, or we could limit this term to only apply to a system that is aware of itself.  Either way we have to be careful here if we’re looking to distinguish between consciousness generally speaking (consciousness in any form, which may be unrecognizable to us) and the unique kind of consciousness that we as human beings experience.  If an ant is conscious, it doesn’t likely have any of the richness that we have in our experience nor is it likely to have self-awareness like we do (even though a dolphin, which has a large neocortex and prefrontal cortex, is far more likely to).  So we have to keep these different levels of consciousness in mind in order to properly assess their being explained by any cognitive framework.

Looking through a PP lens, we can see that what we come to know about the world and our own bodily states is a matter of predictive models pertaining to various inferred causal relations.  These inferred causal relations ultimately stem from bottom-up sensory input.  But when this information is abstracted at higher and higher levels, eventually one can (in principle) get to a point where those higher level models begin to predict the existence of a unified prediction engine.  In other words, a subset of the highest-level predictive models may eventually predict itself as a self.  We might describe this process as the emergence of some level of self-awareness, even if higher levels of self-awareness aren’t possible unless particular kinds of higher level models have been generated.

What kinds of predictions might be involved with this kind of emergence?  Well, we might expect that predictions pertaining to our own autobiographical history, which is largely composed of episodic memories of our past experience, would contribute to this process (e.g. “I remember when I went to that amusement park with my friend Mary, and we both vomited!”).  If we begin to infer what is common or continuous between those memories of past experiences (even if only 1 second in the past), we may discover that there is a form of psychological continuity or identity present.  And if this psychological continuity (this type of causal relation) is coincident with an incredibly stable set of predictions pertaining to (especially internal) bodily states, then an embodied subject or self can plausibly emerge from it.

This emergence of an embodied self is also likely fueled by predictions pertaining to other objects that we infer existing as subjects.  For instance, in order to develop a theory of mind about other people, that is, in order to predict how other people will behave, we can’t simply model their external behavior as we can for something like a rock falling down a hill.  This can work up to a point, but eventually it’s just not good enough as behaviors become more complex.  Animal behavior, most especially that of humans, is far more complex than that of inanimate objects and as such it is going to be far more effective to infer some internal hidden causes for that behavior.  Our predictions would work well if they involved some kind of internal intentionality and goal-directedness operating within any animate object, thereby transforming that object into a subject.  This should be no less true for how we model our own behavior.

If this object that we’ve now inferred to be a subject seems to behave in ways that we see ourselves as behaving (especially if it’s another human being), then we can begin to infer some kind of equivalence despite being separate subjects.  We can begin to infer that they too have beliefs, desires, and emotions, and thus that they have an internal perspective that we can’t directly access just as they aren’t able to access ours.  And we can also see ourselves from a different perspective based on how we see those other subjects from an external perspective.  Since I can’t easily see myself from an external perspective, when I look at others and infer that we are similar kinds of beings, then I can begin to see myself as having both an internal and external side.  I can begin to infer that others see me similar to the way that I see them, thus further adding to my concept of self and the boundaries that define that self.

Multiple meta-cognitive predictions can be inferred based on all of these interactions with others, and from the introspective interactions with our brain’s own models.  Once this happens, a cognitive agent like ourselves may begin to think about thinking and think about being a thinking being, and so on and so forth.  All of these cognitive moves would seem to provide varying degrees of self-hood or self-awareness.  And these can all be thought of as the brain’s best guesses that account for its own behavior.  Either way, it seems that the level of consciousness that is intrinsic to an agent’s experience is going to be dependent on what kinds of higher level models and meta-models are operating within the agent.

Consciousness as Integrated Information

One prominent theory of consciousness is the Integrated Information Theory of Consciousness, otherwise known as IIT.  This theory, initially formulated by Giulio Tononi back in 2004 and which has undergone development ever since, posits that consciousness is ultimately dependent on the degree of information integration that is inherent in the causal properties of some system.  Another way of saying this is that the causal system specified is unified such that every part of the system must be able to affect and be affected by the rest of the system.  If you were to physically isolate one part of a system from the rest of it (and if this part was the least significant to the rest of the system), then the resulting change in the cause-effect structure of the system would quantify the degree of integration.  A large change in the cause-effect structure based on this part’s isolation from the system (that is, by having introduced what is called a minimum partition to the system) would imply a high degree of information integration and vice versa.  And again, a high degree of integration implies a high degree of consciousness.

Notice how this information integration axiom in IIT posits that a cognitive system that is entirely feed-forward will not be conscious.  So if our brain processed incoming sensory information from the bottom up and there was no top-down generative model feeding downward through the system, then IIT would predict that our brain wouldn’t be able to produce consciousness.  PP on the other hand, posits a feedback system (as opposed to feed-forward) where the bottom-up sensory information that flows upward is met with a downward flow of top-down predictions trying to explain away that sensory information.  The brain’s predictions cause a change in the resulting prediction error, and this prediction error serves as feedback to modify the brain’s predictions.  Thus, a cognitive architecture like that suggested by PP is predicted to produce consciousness according to the most fundamental axiom of IIT.

Additionally, PP posits cross-modal sensory features and functionality where the brain integrates (especially lower level) predictions spanning various spatio-temporal scales from different sensory modalities, into a unified whole.  For example, if I am looking at and petting a black cat lying on my lap and hearing it purr, PP posits that my perceptual experience and contextual understanding of that experience are based on having integrated the visual, tactile, and auditory expectations that I’ve associated to constitute such an experience of a “black cat”.  It is going to be contingent on a conjunction of predictions that are occurring simultaneously in order to produce a unified experience rather than a barrage of millions or billions of separate causal relations (let alone those which stem from different sensory modalities) or having millions or billions of separate conscious experiences (which would seem to necessitate separate consciousnesses if they are happening at the same time).

Evolution of Consciousness

Since IIT identifies consciousness with integrated information, it can plausibly account for why it evolved in the first place.  The basic idea here is that a brain that is capable of integrating information is more likely to exploit and understand an environment that has a complex causal structure on multiple time scales than a brain that has informationally isolated modules.  This idea has been tested and confirmed to some degree by artificial life simulations (animats) where adaptation and integration are both simulated.  The organism in these simulations was a Braitenberg-like vehicle that had to move through a maze.  After 60,000 generations of simulated brains evolving through natural selection, it was found that there was a monotonic relationship between their ability to get through the maze and the amount of simulated information integration in their brains.

This increase in adaptation was the result of an effective increase in the number of concepts that the organism could make use of given the limited number of elements and connections possible in its cognitive architecture.  In other words, given a limited number of connections in a causal system (such as a group of neurons), you can pack more functions per element if the level of integration with respect to those connections is high, thus giving an evolutionary advantage to those with higher integration.  Therefore, when all else is equal in terms of neural economy and resources, higher integration gives an organism the ability to take advantage of more regularities in their environment.

From a PP perspective, this makes perfect sense because the complex causal structure of the environment is described as being modeled at many different levels of abstraction and at many different spatio-temporal scales.  All of these modeled causal relations are also described as having a hierarchical structure with models contained within models, and with many associations existing between various models.  These associations between models can be accounted for by a cognitive architecture that re-uses certain sets of neurons in multiple models, so the association is effectively instantiated by some literal degree of neuronal overlap.  And of course, these associations between multiply-leveled predictions allows the brain to exploit (and create!) as many regularities in the environment as possible.  In short, both PP and IIT make a lot of practical sense from an evolutionary perspective.

That’s All Folks!

And this concludes my post-series on the Predictive Processing (PP) framework and how I see it as being applicable to a far more broad account of mentality and brain function, than it is generally assumed to be.  If there’s any takeaways from this post-series, I hope you can at least appreciate the parsimony and explanatory scope of predictive processing and viewing the brain as a creative and highly capable prediction engine.

Predictive Processing: Unlocking the Mysteries of Mind & Body (Part V)

In the previous post, part 4 in this series on Predictive Processing (PP), I explored some aspects of reasoning and how different forms of reasoning can be built from a foundational bedrock of Bayesian inference (click here for parts 1, 2, or 3).  This has a lot to do with language, but I also claimed that it depends on how the brain is likely generating new models, which I think is likely to involve some kind of natural selection operating on neural networks.  The hierarchical structure of the generative models for these predictions as described within a PP framework, also seems to fit well with the hierarchical structure that we find in the brain’s neural networks.  In this post, I’m going to talk about the relation between memory, imagination, and unconscious and conscious forms of reasoning.

Memory, Imagination, and Reasoning

Memory is of course crucial to the PP framework whether for constructing real-time predictions of incoming sensory information (for perception) or for long-term predictions involving high-level, increasingly abstract generative models that allow us to accomplish complex future goals (like planning to go grocery shopping, or planning for retirement).  Either case requires the brain to have stored some kind of information pertaining to predicted causal relations.  Rather than memories being some kind of exact copy of past experiences (where they’d be stored like data on a computer), research has shown that memory functions more like a reconstruction of those past experiences which are modified by current knowledge and context, and produced by some of the same faculties used in imagination.

This accounts for any false or erroneous aspects of our memories, where the recalled memory can differ substantially from how the original event was experienced.  It also accounts for why our memories become increasingly altered as more time passes.  Over time, we learn new things, continuing to change many of our predictive models about the world, and thus have a more involved reconstructive process the older the memories are.  And the context we find ourselves in when trying to recall certain memories, further affect this reconstruction process, adapting our memories in some sense to better match what we find most salient and relevant in the present moment.

Conscious vs. Unconscious Processing & Intuitive Reasoning (Intuition)

Another attribute of memory is that it is primarily unconscious, where we seem to have this pool of information that is kept out of consciousness until parts of it are needed (during memory recall or or other conscious thought processes).  In fact, within the PP framework we can think of most of our generative models (predictions), especially those operating in the lower levels of the hierarchy, as being out of our conscious awareness as well.  However, since our memories are composed of (or reconstructed with) many higher level predictions, and since only a limited number of them can enter our conscious awareness at any moment, this implies that most of the higher-level predictions are also being maintained or processed unconsciously as well.

It’s worth noting however that when we were first forming these memories, a lot of the information was in our consciousness (the higher-level, more abstract predictions in particular).  Within PP, consciousness plays a special role since our attention modifies what is called the precision weight (or synaptic gain) on any prediction error that flows upward through the predictive hierarchy.  This means that the prediction errors produced from the incoming sensory information or at even higher levels of processing are able to have a greater impact on modifying and updating the predictive models.  This makes sense from an evolutionary perspective, where we can ration our cognitive resources in a more adaptable way, by allowing things that catch our attention (which may be more important to our survival prospects) to have the greatest effect on how we understand the world around us and how we need to act at any given moment.

After repeatedly encountering certain predicted causal relations in a conscious fashion, the more likely those predictions can become automated or unconsciously processed.  And if this has happened with certain rules of inference that govern how we manipulate and process many of our predictive models, it seems reasonable to suspect that this would contribute to what we call our intuitive reasoning (or intuition).  After all, intuition seems to give people the sense of knowing something without knowing how it was acquired and without any present conscious process of reasoning.

This is similar to muscle memory or procedural memory (like learning how to ride a bike) which is consciously processed at first (thus involving many parts of the cerebral cortex), but after enough repetition it becomes a faster and more automated process that is accomplished more economically and efficiently by the basal ganglia and cerebellum, parts of the brain that are believed to handle a great deal of unconscious processing like that needed for procedural memory.  This would mean that the predictions associated with these kinds of causal relations begin to function out of our consciousness, even if the same predictive strategy is still in place.

As mentioned above, one difference between this unconscious intuition and other forms of reasoning that operate within the purview of consciousness is that our intuitions are less likely to be updated or changed based on new experiential evidence since our conscious attention isn’t involved in the updating process. This means that the precision weight of upward flowing prediction errors that encounter downward flowing predictions that are operating unconsciously will have little impact in updating those predictions.  Furthermore, the fact that the most automated predictions are often those that we’ve been using for most of our lives, means that they are also likely to have extremely high Bayesian priors, further isolating them from modification.

Some of these priors may become what are called hyperpriors or priors over priors (many of these believed to be established early in life) where there may be nothing that can overcome them, because they describe an extremely abstract feature of the world.  An example of a possible hyperprior could be one that demands that the brain settle on one generative model even when it’s comparable to several others under consideration.  One could call this a “tie breaker” hyperprior, where if the brain didn’t have this kind of predictive mechanism in place, it may never be able to settle on a model, causing it to see the world (or some aspect of it) as a superposition of equiprobable states rather than simply one determinate state.  We could see the potential problem in an organism’s survival prospects if it didn’t have this kind of hyperprior in place.  Whether or not a hyperprior like this is a form of innate specificity, or acquired in early learning is debatable.

An obvious trade-off with intuition (or any kind of innate biases) is that it provides us with fast, automated predictions that are robust and likely to be reliable much of the time, but at the expense of not being able to adequately handle more novel or complex situations, thereby leading to fallacious inferences.  Our cognitive biases are also likely related to this kind of unconscious reasoning whereby evolution has naturally selected cognitive strategies that work well for the kind of environment we evolved in (African savanna, jungle, etc.) even at the expense of our not being able to adapt as well culturally or in very artificial situations.

Imagination vs. Perception

One large benefit of storing so much perceptual information in our memories (predictive models with different spatio-temporal scales) is our ability to re-create it offline (so to speak).  This is where imagination comes in, where we are able to effectively simulate perceptions without requiring a stream of incoming sensory data that matches it.  Notice however that this is still a form of perception, because we can still see, hear, feel, taste and smell predicted causal relations that have been inferred from past sensory experiences.

The crucial difference, within a PP framework, is the role of precision weighting on the prediction error, just as we saw above in terms of trying to update intuitions.  If precision weighting is set or adjusted to be relatively low with respect to a particular set of predictive models, then prediction error will have little if any impact on the model.  During imagination, we effectively decouple the bottom-up prediction error from the top-down predictions associated with our sensory cortex (by reducing the precision weighting of the prediction error), thus allowing us to intentionally perceive things that aren’t actually in the external world.  We need not decouple the error from the predictions entirely, as we may want our imagination to somehow correlate with what we’re actually perceiving in the external world.  For example, maybe I want to watch a car driving down the street and simply imagine that it is a different color, while still seeing the rest of the scene as I normally would.  In general though, it is this decoupling “knob” that we can turn (precision weighting) that underlies our ability to produce and discriminate between normal perception and our imagination.

So what happens when we lose the ability to control our perception in a normal way (whether consciously or not)?  Well, this usually results in our having some kind of hallucination.  Since perception is often referred to as a form of controlled hallucination (within PP), we could better describe a pathological hallucination (such as that arising from certain psychedelic drugs or a condition like Schizophrenia) as a form of uncontrolled hallucination.  In some cases, even with a perfectly normal/healthy brain, when the prediction error simply can’t be minimized enough, or the brain is continuously switching between models, based on what we’re looking at, we experience perceptual illusions.

Whether it’s illusions, hallucinations, or any other kind of perceptual pathology (like not being able to recognize faces), PP offers a good explanation for why these kinds of experiences can happen to us.  It’s either because the models are poor (their causal structure or priors) or something isn’t being controlled properly, like the delicate balance between precision weighting and prediction error, any of which that could result from an imbalance in neurotransmitters or some kind of brain damage.

Imagination & Conscious Reasoning

While most people would tend to define imagination as that which pertains to visual imagery, I prefer to classify all conscious experiences that are not directly resulting from online perception as imagination.  In other words, any part of our conscious experience that isn’t stemming from an immediate inference of incoming sensory information is what I consider to be imagination.  This is because any kind of conscious thinking is going to involve an experience that could in theory be re-created by an artificial stream of incoming sensory information (along with our top-down generative models that put that information into a particular context of understanding).  As long as the incoming sensory information was a particular way (any way that we can imagine!), even if it could never be that way in the actual external world we live in, it seems to me that it should be able to reproduce any conscious process given the right top-down predictive model.  Another way of saying this is that imagination is simply another word to describe any kind of offline conscious mental simulation.

This also means that I’d classify any and all kinds of conscious reasoning processes as yet another form of imagination.  Just as is the case with more standard conceptions of imagination (within PP at least), we are simply taking particular predictive models, manipulating them in certain ways in order to simulate some result with this process decoupled (at least in part) from actual incoming sensory information.  We may for example, apply a rule of inference that we’ve picked up on and manipulate several predictive models of causal relations using that rule.  As mentioned in the previous post and in the post from part 2 of this series, language is also likely to play a special role here where we’ll likely be using it to help guide this conceptual manipulation process by organizing and further representing the causal relations in a linguistic form, and then determining the resulting inference (which will more than likely be in a linguistic form as well).  In doing so, we are able to take highly abstract properties of causal relations and apply rules to them to extract new information.

If I imagine a purple elephant trumpeting and flying in the air over my house, even though I’ve never experienced such a thing, it seems clear that I’m manipulating several different types of predicted causal relations at varying levels of abstraction and experiencing the result of that manipulation.  This involves inferred causal relations like those pertaining to visual aspects of elephants, the color purple, flying objects, motion in general, houses, the air, and inferred causal relations pertaining to auditory aspects like trumpeting sounds and so forth.

Specific instances of these kinds of experienced causal relations have led to my inferring them as an abstract probabilistically-defined property (e.g. elephantness, purpleness, flyingness, etc.) that can be reused and modified to some degree to produce an infinite number of possible recreated perceptual scenes.  These may not be physically possible perceptual scenes (since elephants don’t have wings to fly, for example) but regardless I’m able to add or subtract, mix and match, and ultimately manipulate properties in countless ways, only limited really by what is logically possible (so I can’t possibly imagine what a square circle would look like).

What if I’m performing a mathematical calculation, like “adding 9 + 9”, or some other similar problem?  This appears (upon first glance at least) to be very qualitatively different than simply imagining things that we tend to perceive in the world like elephants, books, music, and other things, even if they are imagined in some phantasmagorical way.  As crazy as those imagined things may be, they still contain things like shapes, colors, sounds, etc., and a mathematical calculation seems to lack this.  I think the key thing to realize here is the fundamental process of imagination as being able to add or subtract and manipulate abstract properties in any way that is logically possible (given our current set of predictive models).  This means that we can imagine properties or abstractions that lack all the richness of a typical visual/auditory perceptual scene.

In the case of a mathematical calculation, I would be manipulating previously acquired predicted causal relations that pertain to quantity and changes in quantity.  Once I was old enough to infer that separate objects existed in the world, then I could infer an abstraction of how many objects there were in some space at some particular time.  Eventually, I could abstract the property of how many objects without applying it to any particular object at all.  Using language to associate a linguistic symbol for each and every specific quantity would lay the groundwork for a system of “numbers” (where numbers are just quantities pertaining to no particular object at all).  Once this was done, then my brain could use the abstraction of quantity and manipulate it by following certain inferred rules of how quantities can change by adding to or subtracting from them.  After some practice and experience I would now be in a reasonable position to consciously think about “adding 9 + 9”, and either do it by following a manual iterative rule of addition that I’ve learned to do with real or imagined visual objects (like adding up some number of apples or dots/points in a row or grid), or I can simply use a memorized addition table and search/recall the sum I’m interested in (9 + 9 = 18).

Whether we consider imagining a purple elephant, mentally adding up numbers, thinking about what I’m going to say to my wife when I see her next, or trying to explicitly apply logical rules to some set of concepts, all of these forms of conscious thought or reasoning are all simply different sets of predictive models that I’m simply manipulating in mental simulations until I arrive at a perception that’s understood in the desired context and that has minimal prediction error.

Putting it all together

In summary, I think we can gain a lot of insight by looking at all the different aspects of brain function through a PP framework.  Imagination, perception, memory, intuition, and conscious reasoning fit together very well when viewed as different aspects of hierarchical predictive models that are manipulated and altered in ways that give us a much more firm grip on the world we live in and its inferred causal structure.  Not only that, but this kind of cognitive architecture also provides us with an enormous potential for creativity and intelligence.  In the next post in this series, I’m going to talk about consciousness, specifically theories of consciousness and how they may be viewed through a PP framework.

Predictive Processing: Unlocking the Mysteries of Mind & Body (Part IV)

In the previous post which was part 3 in this series (click here for parts 1 and 2) on Predictive Processing (PP), I discussed how the PP framework can be used to adequately account for traditional and scientific notions of knowledge, by treating knowledge as a subset of all the predicted causal relations currently at our brain’s disposal.  This subset of predictions that we tend to call knowledge has the special quality of especially high confidence levels (high Bayesian priors).  Within a scientific context, knowledge tends to have an even stricter definition (and even higher confidence levels) and so we end up with a smaller subset of predictions which have been further verified through comparing them with the inferred predictions of others and by testing them with external means of instrumentation and some agreed upon conventions for analysis.

However, no amount of testing or verification is going to give us direct access to any knowledge per se.  Rather, the creation or discovery of knowledge has to involve the application of some kind of reasoning to explain the causal inputs, and only after this reasoning process can the resulting predicted causal relations be validated to varying degrees by testing it (through raw sensory data, external instruments, etc.).  So getting an adequate account of reasoning within any theory or framework of overall brain function is going to be absolutely crucial and I think that the PP framework is well-suited for the job.  As has already been mentioned throughout this post-series, this framework fundamentally relies on a form of Bayesian inference (or some approximation) which is a type of reasoning.  It is this inferential strategy then, combined with a hierarchical neurological structure for it to work upon, that would allow our knowledge to be created in the first place.

Rules of Inference & Reasoning Based on Hierarchical-Bayesian Prediction Structure, Neuronal Selection, Associations, and Abstraction

While PP tends to focus on perception and action in particular, I’ve mentioned that I see the same general framework as being able to account for not only the folk psychological concepts of beliefs, desires, and emotions, but also that the hierarchical predictive structure it entails should plausibly be able to account for language and ontology and help explain the relationship between the two.  It seems reasonable to me that the associations between all of these hierarchically structured beliefs or predicted causal relations at varying levels of abstraction, can provide a foundation for our reasoning as well, whether intuitive or logical forms of reasoning.

To illustrate some of the importance of associations between beliefs, consider an example like the belief in object permanence (i.e. that objects persist or continue to exist even when I can no longer see them).  This belief of ours has an extremely high prior because our entire life experience has only served to support this prediction in a large number of ways.  This means that it’s become embedded or implicit in a number of other beliefs.  If I didn’t predict that object permanence was a feature of my reality, then an enormous number of everyday tasks would become difficult if not impossible to do because objects would be treated as if they are blinking into and out of existence.

We have a large number of beliefs that require object permanence (and which are thus associated with object permanence), and so it is a more fundamental lower-level prediction (though not as low level as sensory information entering the visual cortex) and we use this lower-level prediction to build upon into any number of higher-level predictions in the overall conceptual/predictive hierarchy.  When I put money in a bank, I expect to be able to spend it even if I can’t see it anymore (such as with a check or debit card).  This is only possible if my money continues to exist even when out of view (regardless of if the money is in a paper/coin or electronic form).  This is just one of many countless everyday tasks that depend on this belief.  So it’s no surprise that this belief (this set of predictions) would have an incredibly high Bayesian prior, and therefore I would treat it as a non-negotiable fact about reality.

On the other hand, when I was a newborn infant, I didn’t have this belief of object permanence (or at best, it was a very weak belief).  Most psychologists estimate that our belief in object permanence isn’t acquired until after several months of brain development and experience.  This would translate to our having a relatively low Bayesian prior for this belief early on in our lives, and only once a person begins to form predictions based on these kinds of recognized causal relations can we begin to increase that prior and perhaps eventually reach a point that results in a subjective experience of a high degree in certainty for this particular belief.  From that point on, we are likely to simply take that belief for granted, no longer questioning it.  The most important thing to note here is that the more associations made between beliefs, the higher their effective weighting (their priors), and thus the higher our confidence in those beliefs becomes.

Neural Implementation, Spontaneous or Random Neural Activity & Generative Model Selection

This all seems pretty reasonable if a neuronal implementation worked to strengthen Bayesian priors as a function of the neuronal/synaptic connectivity (among other factors), where neurons that fire together are more likely to wire together.  And connectivity strength will increase the more often this happens.  On the flip-side, the less often this happens or if it isn’t happening at all then the connectivity is likely to be weakened or non-existent.  So if a concept (or a belief composed of many conceptual relations) is represented by some cluster of interconnected neurons and their activity, then it’s applicability to other concepts increases its chances of not only firing but also increasing the strength of wiring with those other clusters of neurons, thus plausibly increasing the Bayesian priors for the overlapping concept or belief.

Another likely important factor in the Bayesian inferential process, in terms of the brain forming new generative models or predictive hypotheses to test, is the role of spontaneous or random neural activity and neural cluster generation.  This random neural activity could plausibly provide a means for some randomly generated predictions or random changes in the pool of predictive models that our brain is able to select from.  Similar to the role of random mutation in gene pools which allows for differential reproductive rates and relative fitness of offspring, some amount of randomness in neural activity and the generative models that result would allow for improved models to be naturally selected based on those which best minimize prediction error.  The ability to minimize prediction error could be seen as a direct measure of the fitness of the generative model, within this evolutionary landscape.

This idea is related to the late Gerald Edelman’s Theory of Neuronal Group Selection (NGS), also known as Neural Darwinism, which I briefly explored in a post I wrote long ago.  I’ve long believed that this kind of natural selection process is applicable to a number of different domains (aside from genetics), and I think any viable version of PP is going to depend on it to at least some degree.  This random neural activity (and the naturally selected products derived from them) could be thought of as contributing to a steady supply of new generative models to choose from and thus contributing to our overall human creativity as well whether for reasoning and problem solving strategies or simply for artistic expression.

Increasing Abstraction, Language, & New Rules of Inference

This kind of use it or lose it property of brain plasticity combined with dynamic associations between concepts or beliefs and their underlying predictive structure, would allow for the brain to accommodate learning by extracting statistical inferences (at increasing levels of abstraction) as they occur and modifying or eliminating those inferences by changing their hierarchical associative structure as prediction error is encountered.  While some form of Bayesian inference (or an approximation to it) underlies this process, once lower-level inferences about certain causal relations have been made, I believe that new rules of inference can be derived from this basic Bayesian foundation.

To see how this might work, consider how we acquire a skill like learning how to speak and write in some particular language.  The rules of grammar, the syntactic structure and so forth which underlie any particular language are learned through use.  We begin to associate words with certain conceptual structures (see part 2 of this post-series for more details on language and ontology) and then we build up the length and complexity of our linguistic expressions by adding concepts built on higher levels of abstraction.  To maximize the productivity and specificity of our expressions, we also learn more complex rules pertaining to the order in which we speak or write various combinations of words (which varies from language to language).

These grammatical rules can be thought of as just another higher-level abstraction, another higher-level causal relation that we predict will convey more specific information to whomever we are speaking to.  If it doesn’t seem to do so, then we either modify what we have mistakenly inferred to be those grammatical rules, or depending on the context, we may simply assume that the person we’re talking to hasn’t conformed to the language or grammar that my community seems to be using.

Just like with grammar (which provides a kind of logical structure to our language), we can begin to learn new rules of inference built on the same probabilistic predictive bedrock of Bayesian inference.  We can learn some of these rules explicitly by studying logic, induction, deduction, etc., and consciously applying those rules to infer some new piece of knowledge, or we can learn these kinds of rules implicitly based on successful predictions (pertaining to behaviors of varying complexity) that happen to result from stumbling upon this method of processing causal relations within various contexts.  As mentioned earlier, this would be accomplished in part by the natural selection of randomly-generated neural network changes that best reduce the incoming prediction error.

However, language and grammar are interesting examples of an acquired set of rules because they also happen to be the primary tool that we use to learn other rules (along with anything we learn through verbal or written instruction), including (as far as I can tell) various rules of inference.  The logical structure of language (though it need not have an exclusively logical structure), its ability to be used for a number of cognitive short-cuts, and it’s influence on our thought complexity and structure, means that we are likely dependent on it during our reasoning processes as well.

When we perform any kind of conscious reasoning process, we are effectively running various mental simulations where we can intentionally manipulate our various generative models to test new predictions (new models) at varying levels of abstraction, and thus we also manipulate the linguistic structure associated with those generative models as well.  Since I have a lot more to say on reasoning as it relates to PP, including more on intuitive reasoning in particular, I’m going to expand on this further in my next post in this series, part 5.  I’ll also be exploring imagination and memory including how they relate to the processes of reasoning.

Predictive Processing: Unlocking the Mysteries of Mind & Body (Part III)

In the first post in this series I re-introduced the basic concepts underling the Predictive Processing (PP) theory of perception, in particular how the brain uses a form of active Bayesian inference to form predictive models that effectively account for perception as well as action.  I also mentioned how the folk psychological concepts of beliefs, desires and emotions can fit within the PP framework.  In the second post in this series I expanded the domain of PP a bit to show how it relates to language and ontology, and how we perceive the world to be structured as discrete objects, objects with “fuzzy boundaries”, and other more complex concepts all stemming from particular predicted causal relations.

It’s important to note that this PP framework differs from classical computational frameworks for brain function in a very big way, because the processing and learning steps are no longer considered separate stages within the PP framework, but rather work at the same time (learning is effectively occurring all the time).  Furthermore, classical computational frameworks for brain function treat the brain more or less like a computer which I think is very misguided, and the PP framework offers a much better alternative that is far more creative, economical, efficient, parsimonious and pragmatic.  The PP framework holds the brain to be more of a probabilistic system rather than a deterministic computational system as held in classical computationalist views.  Furthermore, the PP framework puts a far greater emphasis on the brain using feed-back loops whereas traditional computational approaches tend to suggest that the brain is primarily a feed-forward information processing system.

Rather than treating the brain like a generic information processing system that passively waits for new incoming data from the outside world, it stays one step ahead of the game, by having formed predictions about the incoming sensory data through a very active and creative learning process built-up by past experiences through a form of active Bayesian inference.  Rather than utilizing some kind of serial processing scheme, this involves primarily parallel neuronal processing schemes resulting in predictions that have a deeply embedded hierarchical structure and relationship with one another.  In this post, I’d like to explore how traditional and scientific notions of knowledge can fit within a PP framework as well.

Knowledge as a Subset of Predicted Causal Relations

It has already been mentioned that, within a PP lens, our ontology can be seen as basically composed of different sets of highly differentiated, probabilistic causal relations.  This allows us to discriminate one object or concept from another, as they are each “composed of” or understood by the brain as causes that can be “explained away” by different sets of predictions.  Beliefs are just another word for predictions, with higher-level beliefs (including those that pertain to very specific contexts) consisting of a conjunction of a number of different lower level predictions.  When we come to “know” something then, it really is just a matter of inferring some causal relation or set of causal relations about a particular aspect of our experience.

Knowledge is often described by philosophers using various forms of the Platonic definition: Knowledge = Justified True Belief.  I’ve talked about this to some degree in previous posts (here, here) and I came up with what I think is a much better working definition for knowledge which could be defined as such:

Knowledge consists of recognized patterns of causality that are stored into memory for later recall and use, that positively and consistently correlate with reality, and for which that correlation has been validated by empirical evidence (i.e. successful predictions made and/or goals accomplished through the use of said recalled patterns).

We should see right off the bat how this particular view of knowledge fits right in to the PP framework, where knowledge consists of predictions of causal relations, and the brain is in the very business of making predictions of causal relations.  Notice my little caveat however, where these causal relations should positively and consistently correlate with “reality” and be supported by empirical evidence.  This is because I wanted to distinguish all predicted causal relations (including those that stem from hallucinations or unreliable inferences) from the subset of predicted causal relations that we have a relatively high degree of certainty in.

In other words, if we are in the game of trying to establish a more reliable epistemology, we want to distinguish between all beliefs and the subset of beliefs that have a very high likelihood of being true.  This distinction however is only useful for organizing our thoughts and claims based on our level of confidence in their truth status.  And for all beliefs, regardless of the level of certainty, the “empirical evidence” requirement in my definition given above is still going to be met in some sense because the incoming sensory data is the empirical evidence (the causes) that support the brain’s predictions (of those causes).

Objective or Scientific Knowledge

Within a domain like science however, where we want to increase the reliability or objectivity of our predictions pertaining to any number of inferred causal relations in the world, we need to take this same internalized strategy of modifying the confidence levels of our predictions by comparing them to the predictions of others (third-party verification) and by testing these predictions with externally accessible instrumentation using some set of conventions or standards (including the scientific method).

Knowledge then, is largely dependent on or related to our confidence levels in our various beliefs.  And our confidence level in the truth status of a belief is just another way of saying how probable such a belief is to explain away a certain set of causal relations, which is equivalent to our brain’s Bayesian prior probabilities (our “priors”) that characterize any particular set of predictions.  This means that I would need a lot of strong sensory evidence to overcome a belief with high Bayesian priors, not least because it is likely to be associated with a large number of other beliefs.  This association between different predictions seems to me to be a crucial component not only for knowledge generally (or ontology as mentioned in the last post), but also for our reasoning processes.  In the next post of this series, I’m going to expand a bit on reasoning and how I view it through a PP framework.

Predictive Processing: Unlocking the Mysteries of Mind & Body (Part II)

In the first post of this series I introduced some of the basic concepts involved in the Predictive Processing (PP) theory of perception and action.  I briefly tied together the notions of belief, desire, emotion, and action from within a PP lens.  In this post, I’d like to discuss the relationship between language and ontology through the same framework.  I’ll also start talking about PP in an evolutionary context as well, though I’ll have more to say about that in future posts in this series.

Active (Bayesian) Inference as a Source for Ontology

One of the main themes within PP is the idea of active (Bayesian) inference whereby we physically interact with the world, sampling it and modifying it in order to reduce our level of uncertainty in our predictions about the causes of the brain’s inputs.  Within an evolutionary context, we can see why this form of embodied cognition is an ideal schema for an information processing system to employ in order to maximize chances of survival in our highly interactive world.

In order to reduce the amount of sensory information that has to be processed at any given time, it is far more economical for the brain to only worry about the prediction error that flows upward through the neural system, rather than processing all incoming sensory data from scratch.  If the brain is employing a set of predictions that can “explain away” most of the incoming sensory data, then the downward flow of predictions can encounter an upward flow of sensory information (effectively cancelling each other out) and the only thing that remains to propagate upward through the system and do any “cognitive work” (i.e. the only thing that needs to be processed) on the predictive models flowing downward is the remaining prediction error (prediction error = predictions of sensory input minus the actual sensory input).  This is similar to data compression strategies for video files (for example) that only worry about the information that changes over time (pixels that change brightness/color) and then simply compress the information that remains constant (pixels that do not change from frame-to-frame).

The ultimate goal for this strategy within an evolutionary context is to allow the organism to understand its environment in the most salient ways for the pragmatic purposes of accomplishing goals relating to survival.  But once humans began to develop culture and evolve culturally, the predictive strategy gained a new kind of evolutionary breathing space, being able to predict increasingly complex causal relations and developing technology along the way.  All of these inferred causal relations appear to me to be the very source of our ontology, as each hierarchically structured prediction and its ability to become associated with others provides an ideal platform for differentiating between any number of spatio-temporal conceptions and their categorical or logical organization.

An active Bayesian inference system is also ideal to explain our intellectual thirst, human curiosity, and interest in novel experiences (to some degree), because we learn more about the world (and ourselves) by interacting with it in new ways.  In doing so, we are provided with a constant means of fueling and altering our ontology.

Language & Ontology

Language is an important component as well and it fits well within a PP framework as it serves to further link perception and action together in a very important way, allowing us to make new kinds of predictions about the world that wouldn’t have been possible without it.   A tool like language makes a lot of sense from an evolutionary perspective as well since better predictions about the world result in a higher chance of survival.

When we use language by speaking or writing it, we are performing an action which is instantiated by the desire to do so (see previous post about “desire” within a PP framework).  When we interpret language by listening to it or by reading, we are performing a perceptual task which is again simply another set of predictions (in this case, pertaining to the specific causes leading to our sensory inputs).  If we were simply sending and receiving non-lingual nonsense, then the same basic predictive principles underlying perception and action would still apply, but something new emerges when we send and receive actual language (which contains information).  With language, we begin to associate certain sounds and visual information with some kind of meaning or meaningful information.  Once we can do this, we can effectively share our thoughts with one another, or at least many aspects of our thoughts with one another.  This provides for an enormous evolutionary advantage as now we can communicate almost anything we want to one another, store it in external forms of memory (books, computers, etc.), and further analyze or manipulate the information for various purposes (accounting, inventory, science, mathematics, etc.).

By being able to predict certain causal outcomes through the use of language, we are effectively using the lower level predictions associated with perceiving and emitting language to satisfy higher level predictions related to more complex goals including those that extend far into the future.  Since the information that is sent and received amounts to testing or modifying our predictions of the world, we are effectively using language to share and modulate one brain’s set of predictions with that of another brain.  One important aspect of this process is that this information is inherently probabilistic which is why language often trips people up with ambiguities, nuances, multiple meanings behind words and other attributes of language that often lead to misunderstanding.  Wittgenstein is one of the more prominent philosophers who caught onto this property of language and its consequence on philosophical problems and how we see the world structured.  I think a lot of the problems Wittgenstein elaborated on with respect to language can be better accounted for by looking at language as dealing with probabilistic ontological/causal relations that serve some pragmatic purpose, with the meaning of any word or phrase as being best described by its use rather than some clear-cut definition.

This probabilistic attribute of language in terms of the meanings of words having fuzzy boundaries also tracks very well with the ontology that a brain currently has access to.  Our ontology, or what kinds of things we think exist in the world, are often categorized in various ways with some of the more concrete entities given names such as: “animals”, “plants”, “rocks”, “cats”, “cups”, “cars”, “cities”, etc.  But if I morph a wooden chair (say, by chipping away at parts of it with a chisel), eventually it will no longer be recognizable as a chair, and it may begin to look more like a table than a chair or like nothing other than an oddly shaped chunk of wood.  During this process, it may be difficult to point to the exact moment that it stopped being a chair and instead became a table or something else, and this would make sense if what we know to be a chair or table or what-have-you is nothing more than a probabilistic high-level prediction about certain causal relations.  If my brain perceives an object that produces too high of a prediction error based on the predictive model of what a “chair” is, then it will try another model (such as the predictive model pertaining to a “table”), potentially leading to models that are less and less specific until it is satisfied with recognizing the object as merely a “chunk of wood”.

From a PP lens, we can consider lower level predictions pertaining to more basic causes of sensory input (bright/dark regions, lines, colors, curves, edges, etc.) to form some basic ontological building blocks and when they are assembled into higher level predictions, the amount of integrated information increases.  This information integration process leads to condensed probabilities about increasingly complex causal relations, and this ends up reducing the dimensionality of the cause-effect space of the predicted phenomenon (where a set of separate cause-effect repertoires are combined into a smaller number of them).

You can see the advantage here by considering what the brain might do if it’s looking at a black cat sitting on a brown chair.  What if the brain were to look at this scene as merely a set of pixels on the retina that change over time, where there’s no expectations of any subset of pixels to change in ways that differ from any other subset?  This wouldn’t be very useful in predicting how the visual scene will change over time.  What if instead, the brain differentiates one subset of pixels (that correspond to what we call a cat) from all the rest of the pixels, and it does this in part by predicting proximity relations between neighboring pixels in the subset (so if some black pixels move from the right to the left visual field, then some number of neighboring black pixels are predicted to move with it)?

This latter method treats the subset of black-colored pixels as a separate object (as opposed to treating the entire visual scene as a single object), and doing this kind of differentiation in more and more complex ways leads to a well-defined object or concept, or a large number of them.  Associating sounds like “meow” with this subset of black-colored pixels, is just one example of yet another set of properties or predictions that further defines this perceived object as distinct from the rest of the perceptual scene.  Associating this object with a visual or auditory label such as “cat” finally links this ontological object with language.  As long as we agree on what is generally meant by the word “cat” (which we determine through its use), then we can share and modify the predictive models associated with such an object or concept, as we can do with any other successful instance of linguistic communication.

Language, Context, and Linguistic Relativism

However, it should be noted that as we get to more complex causal relations (more complex concepts/objects), we can no longer give these concepts a simple one word label and expect to communicate information about them nearly as easily as we could for the concept of a “cat”.  Think about concepts like “love” or “patriotism” or “transcendence” and realize how there’s many different ways that we use those terms and how they can mean all sorts of different things and so our meaning behind those words will be heavily conveyed to others by the context that they are used in.  And context in a PP framework could be described as simply the (expected) conjunction of multiple predictive models (multiple sets of causal relations) such as the conjunction of the predictive models pertaining to the concepts of food, pizza, and a desirable taste, and the word “love” which would imply a particular use of the word “love” as used in the phrase “I love pizza”.  This use of the word “love” is different than one which involves the conjunction of predictive models pertaining to the concepts of intimacy, sex, infatuation, and care, implied in a phrase like “I love my wife”.  In any case, conjunctions of predictive models can get complicated and this carries over to our stretching our language to its very limits.

Since we are immersed in language and since it is integral in our day-to-day lives, we also end up being conditioned to think linguistically in a number of ways.  For example, we often think with an interior monologue (e.g. “I am hungry and I want pizza for lunch”) even when we don’t plan on communicating this information to anyone else, so it’s not as if we’re simply rehearsing what we need to say before we say it.  I tend to think however that this linguistic thinking (thinking in our native language) is more or less a result of the fact that the causal relations that we think about have become so strongly associated with certain linguistic labels and propositions, that we sort of automatically think of the causal relations alongside the labels that we “hear” in our head.  This seems to be true even if the causal relations could be thought of without any linguistic labels, in principle at least.  We’ve simply learned to associate them so strongly to one another that in most cases separating the two is just not possible.

On the flip side, this tendency of language to associate itself with our thoughts, also puts certain barriers or restrictions on our thoughts.  If we are always preparing to share our thoughts through language, then we’re going to become somewhat entrained to think in ways that can be most easily expressed in a linguistic form.  So although language may simply be along for the ride with many non-linguistic aspects of thought, our tendency to use it may also structure our thinking and reasoning in large ways.  This would account for why people raised in different cultures with different languages see the world in different ways based on the structure of their language.  While linguistic determinism seems to have been ruled out (the strong version of the Sapir-Whorf hypothesis), there is still strong evidence to support linguistic relativism (the weak version of the Sapir-Whorf hypothesis), whereby one’s language effects their ontology and view of how the world is structured.

If language is so heavily used day-to-day then this phenomenon makes sense as viewed through a PP lens since we’re going to end up putting a high weight on the predictions that link ontology with language since these predictions have been demonstrated to us to be useful most of the time.  Minimal prediction error means that our Bayesian evidence is further supported and the higher the weight carried by these predictions, the more these predictions will restrict our overall thinking, including how our ontology is structured.

Moving on…

I think that these are but a few of the interesting relationships between language and ontology and how a PP framework helps to put it all together nicely, and I just haven’t seen this kind of explanatory power and parsimony in any other kind of conceptual framework about how the brain functions.  This bodes well for the framework and it’s becoming less and less surprising to see it being further supported over time with studies in neuroscience, cognition, psychology, and also those pertaining to pathologies of the brain, perceptual illusions, etc.  In the next post in this series, I’m going to talk about knowledge and how it can be seen through the lens of PP.

Predictive Processing: Unlocking the Mysteries of Mind & Body (Part I)

I’ve been away from writing for a while because I’ve had some health problems relating to my neck.  A few weeks ago I had double-cervical-disc replacement surgery and so I’ve been unable to write and respond to comments and so forth for a little while.  I’m in the second week following my surgery now and have finally been able to get back to writing, which feels very good given that I’m unable to lift or resume martial arts for the time being.  Anyway, I want to resume my course of writing beginning with a post-series that pertains to Predictive Processing (PP) and the Bayesian brain.  I’ve written one post on this topic a little over a year ago (which can be found here) as I’ve become extremely interested in this topic for the last several years now.

The Predictive Processing (PP) theory of perception shows a lot of promise in terms of finding an overarching schema that can account for everything that the brain seems to do.  While its technical application is to account for the acts of perception and active inference in particular, I think it can be used more broadly to account for other descriptions of our mental life such as beliefs (and knowledge), desires, emotions, language, reasoning, cognitive biases, and even consciousness itself.  I want to explore some of these relationships as viewed through a PP lens more because I think it is the key framework needed to reconcile all of these aspects into one coherent picture, especially within the evolutionary context of an organism driven to survive.  Let’s begin this post-series by first looking at how PP relates to perception (including imagination), beliefs, emotions, and desires (and by extension, the actions resulting from particular desires).

Within a PP framework, beliefs can be best described as simply the set of particular predictions that the brain employs which encompass perception, desires, action, emotion, etc., and which are ultimately mediated and updated in order to reduce prediction errors based on incoming sensory evidence (and which approximates a Bayesian form of inference).  Perception then, which is constituted by a subset of all our beliefs (with many of them being implicit or unconscious beliefs), is more or less a form of controlled hallucination in the sense that what we consciously perceive is not the actual sensory evidence itself (not even after processing it), but rather our brain’s “best guess” of what the causes for the incoming sensory evidence are.

Desires can be best described as another subset of one’s beliefs, and a set of beliefs which has the special characteristic of being able to drive action or physical behavior in some way (whether driving internal bodily states, or external ones that move the body in various ways).  Finally, emotions can be thought of as predictions pertaining to the causes of internal bodily states and which may be driven or changed by changes in other beliefs (including changes in desires or perceptions).

When we believe something to be true or false, we are basically just modeling some kind of causal relationship (or its negation) which is able to manifest itself into a number of highly-weighted predicted perceptions and actions.  When we believe something to be likely true or likely false, the same principle applies but with a lower weight or precision on the predictions that directly corresponds to the degree of belief or disbelief (and so new sensory evidence will more easily sway such a belief).  And just like our perceptions, which are mediated by a number of low and high-level predictions pertaining to incoming sensory data, any prediction error that the brain encounters results in either updating the perceptual predictions to new ones that better reduce the prediction error and/or performing some physical action that reduces the prediction error (e.g. rotating your head, moving your eyes, reaching for an object, excreting hormones in your body, etc.).

In all these cases, we can describe the brain as having some set of Bayesian prior probabilities pertaining to the causes of incoming sensory data, and these priors changing over time in response to prediction errors arising from new incoming sensory evidence that fails to be “explained away” by the predictive models currently employed.  Strong beliefs are associated with high prior probabilities (highly-weighted predictions) and therefore need much more counterfactual sensory evidence to be overcome or modified than for weak beliefs which have relatively low priors (low-weighted predictions).

To illustrate some of these concepts, let’s consider a belief like “apples are a tasty food”.  This belief can be broken down into a number of lower level, highly-weighted predictions such as the prediction that eating a piece of what we call an “apple” will most likely result in qualia that accompany the perception of a particular satisfying taste, the lower level prediction that doing so will also cause my perception of hunger to change, and the higher level prediction that it will “give me energy” (with these latter two predictions stemming from the more basic category of “food” contained in the belief).  Another prediction or set of predictions is that these expectations will apply to not just one apple but a number of apples (different instances of one type of apple, or different types of apples altogether), and a host of other predictions.

These predictions may even result (in combination with other perceptions or beliefs) in an actual desire to eat an apple which, under a PP lens could be described as the highly weighted prediction of what it would feel like to find an apple, to reach for an apple, to grab it, to bite off a piece of it, to chew it, and to swallow it.  If I merely imagine doing such things, then the resulting predictions will necessarily carry such a small weight that they won’t be able to influence any actual motor actions (even if these imagined perceptions are able to influence other predictions that may eventually lead to some plan of action).  Imagined perceptions will also not carry enough weight (when my brain is functioning normally at least) to trick me into thinking that they are actual perceptions (by “actual”, I simply mean perceptions that correspond to incoming sensory data).  This low-weighting attribute of imagined perceptual predictions thus provides a viable way for us to have an imagination and to distinguish it from perceptions corresponding to incoming sensory data, and to distinguish it from predictions that directly cause bodily action.  On the other hand, predictions that are weighted highly enough (among other factors) will be uniquely capable of affecting our perception of the real world and/or instantiating action.

This latter case of desire and action shows how the PP model takes the organism to be an embodied prediction machine that is directly influencing and being influenced by the world that its body interacts with, with the ultimate goal of reducing any prediction error encountered (which can be thought of as maximizing Bayesian evidence).  In this particular example, the highly-weighted prediction of eating an apple is simply another way of describing a desire to eat an apple, which produces some degree of prediction error until the proper actions have taken place in order to reduce said error.  The only two ways of reducing this prediction error are to change the desire (or eliminate it) to one that no longer involves eating an apple, and/or to perform bodily actions that result in actually eating an apple.

Perhaps if I realize that I don’t have any apples in my house, but I realize that I do have bananas, then my desire will change to one that predicts my eating a banana instead.  Another way of saying this is that my higher-weighted prediction of satisfying hunger supersedes my prediction of eating an apple specifically, thus one desire is able to supersede another.  However, if the prediction weight associated with my desire to eat an apple is high enough, it may mean that my predictions will motivate me enough to avoid eating the banana, and instead to predict what it is like to walk out of my house, go to the store, and actually get an apple (and therefore, to actually do so).  Furthermore, it may motivate me to predict actions that lead me to earn the money such that I can purchase the apple (if I don’t already have the money to do so).  To do this, I would be employing a number of predictions having to do with performing actions that lead to me obtaining money, using money to purchase goods, etc.

This is but a taste of what PP has to offer, and how we can look at basic concepts within folk psychology, cognitive science, and theories of mind in a new light.  Associated with all of these beliefs, desires, emotions, and actions (which again, are simply different kinds of predictions under this framework), is a number of elements pertaining to ontology (i.e. what kinds of things we think exist in the world) and pertaining to language as well, and I’d like to explore this relationship in my next post.  This link can be found here.

The Brain as a Prediction Machine

Over the last year, I’ve been reading a lot about Karl Friston and Andy Clark’s work on the concept of perception and action being mediated by a neurological schema centered on “predictive coding”, what Friston calls “active inference”, the “free energy principle”, and Bayesian inference in general as it applies to neuro-scientific models of perception, attention, and action.  Here’s a few links (Friston here; Clark here, here, and here) to some of their work as it is worth reading for those interested in neural modeling, information theory, and learning more about meta-theories pertaining to how the brain integrates and processes information.

I find it fascinating how this newer research and these concepts relate to and help to bring together some content from several of my previous blog posts, in particular, those that mention the concept of hierarchical neurological hardware and those that mention my own definition of knowledge “as recognized causal patterns that allow us to make successful predictions.”  For those that may be interested, here’s a list of posts I’ve made over the last few years that I think contain some relevant content (in chronological order).

• Knowledge and the “Brain in a Vat” scenario
• Knowledge: An Expansion of the Platonic Definition
• Neuroscience Arms Race & Our Changing World View
• Neurological Configuration & the Prospects of an Innate Ontology
• Darwin’s Big Idea May Be The Biggest Yet

The ideas formulated by Friston and expanded on by Clark center around the brain being (in large part) a prediction generating machine.  This fits in line with my own conclusions about what the brain seems to be doing when it’s acquiring knowledge over time (however limited my reading is on the subject).  Here’s an image of the basic predictive processing schema:

The basic Predictive Processing schema (adapted from Lupyan and Clark (2014))

One key element in Friston and Clark’s work (among the work of some others) is the amalgamation of perception and action.  In this framework, perception itself is simply the result of the brain’s highest level predictions of incoming sensory data.  But also important in this framework is that prediction error minimization is accomplished through embodiment itself.  That is to say, their models posit that the brain not only tries to reduce prediction errors by updating its prediction models based on the actual incoming sensory information (with only the error feeding forward to update the models, similar to data compression schema), but the concept of active inference involves the minimization of prediction error through the use of motor outputs.  This could be taken to mean that motor outputs themselves are, in a sense, caused by the brain trying to reduce prediction errors pertaining to predicted sensory input — specifically sensory input that we would say stems from our desires and goals (e.g. desire to fulfill hunger, commuting to work, opening the car door, etc.).

To give a simple example of this model in action, let’s consider an apple resting on a table in front of me.  If I see the apple in front of me and I have a desire to grab it, my brain would not only predict what that apple looks like and how it is perceived over time (and how my arm looks while reaching for it), but it would also predict what it should feel like to reach for the apple.  So if I reach for it based on the somato-sensory prediction and there is some error in that prediction, corroborated by my visual cortex observing my arm moving in some wrong direction, the brain would respond by updating its models that predict what it should feel so that my arm starts moving in the proper direction.  This prediction error minimization is then fine-tuned as I get closer to the apple and can finally grab it.

This embodiment ingrained in the predictive processing models of Friston and Clark can also be well exemplified by the so-called “Outfielder’s Problem”.  In this problem, an outfielder is trying to catch a fly ball.  Now we know that outfielders are highly skilled at doing this rather effectively.  But if we ask the outfielder to merely stand still and watch a batted ball and predict where it will land, their accuracy is generally pretty bad.  So when we think about what strategy the brain takes to accomplish this when moving the body quickly, we begin to see the relevance of active inference and embodiment in the brain’s prediction schema.  The outfielder’s brain employs a brilliant strategy called “optical acceleration cancellation” (OAC).  Here, the well-trained outfielder sees the fly ball, and moves his or her body (while watching the ball) in order to cancel out any optical acceleration observed during the ball’s flight.  If they do this, then they will end up exactly where the ball was going to land, and then they’re able to catch it successfully.

We can imagine fine-grained examples of this active inference during everyday tasks, where I may simply be looking at a picture on my living room wall, and when my brain is predicting how it will look over the span of a few seconds, my head may slightly change its tilt, or direction, or my eyes may slowly move a fraction of a degree this way or that way, however imperceptible to me.  My brain in this case is predicting what the picture on the wall will look like over time and this prediction (according to my understanding of Clark) is identical to what we actually perceive.  One key thing to note here is that the prediction models are not simply updated based on the prediction error that is fed forward through the brain’s neurological hierarchies, but it is also getting some “help” from various motor movements to correct for the errors through action, rather than simply freezing all my muscles and updating the model itself (which may in fact be far less economical for the brain to do).

Another area of research that pertains to this framework, including ways of testing its validity, is that of evolutionary psychology and biology, where one would surmise (if these models are correct) that evolution likely provided our brains with certain hard-wired predictive models and our learning processes over time use these as starting points to produce innate reflexes (such as infant suckling to give a simple example) that allow us to survive long enough to update our models with actual new acquired information.  There are many different facets to this framework and I look forward to reading more about Friston and Clark’s work over the next few years.  I have a feeling that they have hit on something big, something that will help to answer a lot of questions about embodied cognition, perception, and even consciousness itself.

I encourage you to check out the links I provided pertaining to Friston and Clark’s work, to get a taste of the brilliant ideas they’ve been working on.

Conscious Realism & The Interface Theory of Perception

A few months ago I was reading an interesting article in The Atlantic about Donald Hoffman’s Interface Theory of Perception.  As a person highly interested in consciousness studies, cognitive science, and the mind-body problem, I found the basic concepts of his theory quite fascinating.  What was most interesting to me was the counter-intuitive connection between evolution and perception that Hoffman has proposed.  Now it is certainly reasonable and intuitive to assume that evolutionary natural selection would favor perceptions that are closer to “the truth” or closer to the objective reality that exists independent of our minds, simply because of the idea that perceptions that are more accurate will be more likely to lead to survival than perceptions that are not accurate.  As an example, if I were to perceive lions as inert objects like trees, I would be more likely to be naturally selected against and eaten by a lion when compared to one who perceives lions as a mobile predator that could kill them.

While this is intuitive and reasonable to some degree, what Hoffman actually shows, using evolutionary game theory, is that with respect to organisms with comparable complexity, those with perceptions that are closer to reality are never going to be selected for nearly as much as those with perceptions that are tuned to fitness instead.  More so, truth in this case will be driven to extinction when it is up against perceptual models that are tuned to fitness.  That is to say, evolution will select for organisms that perceive the world in a way that is less accurate (in terms of the underlying reality) as long as the perception is tuned for survival benefits.  The bottom line is that given some specific level of complexity, it is more costly to process more information (costing more time and resources), and so if a “heuristic” method for perception can evolve instead, one that “hides” all the complex information underlying reality and instead provides us with a species-specific guide to adaptive behavior, that will always be the preferred choice.

To see this point more clearly, let’s consider an example.  Let’s imagine there’s an animal that regularly eats some kind of insect, such as a beetle, but it needs to eat a particular sized beetle or else it has a relatively high probability of eating the wrong kind of beetle (and we can assume that the “wrong” kind of beetle would be deadly to eat).  Now let’s imagine two possible types of evolved perception: it could have really accurate perceptions about the various sizes of beetles that it encounters so it can distinguish many different sizes from one another (and then choose the proper size range to eat), or it could evolve less accurate perceptions such that all beetles that are either too small or too large appear as indistinguishable from one another (maybe all the wrong-sized beetles whether too large or too small look like indistinguishable red-colored blobs) and perhaps all the beetles that are in the ideal size range for eating appear as green-colored blobs (that are again, indistinguishable from one another).  So the only discrimination in this latter case of perception is between red and green colored blobs.

Both types of perception would solve the problem of which beetles to eat or not eat, but the latter type (even if much less accurate) would bestow a fitness advantage over the former type, by allowing the animal to process much less information about the environment by not focusing on relatively useless information (like specific beetle size).  In this case, with beetle size as the only variable under consideration for survival, evolution would select for the organism that knows less total information about beetle size, as long as it knows what is most important about distinguishing the edible beetles from the poisonous beetles.  Now we can imagine that in some cases, the fitness function could align with the true structure of reality, but this is not what we ever expect to see generically in the world.  At best we may see some kind of overlap between the two but if there doesn’t have to be any then truth will go extinct.

Perception is Analogous to a Desktop Computer Interface

Hoffman analogizes this concept of a “perception interface” with the desktop interface of a personal computer.  When we see icons of folders on the desktop and drag one of those icons to the trash bin, we shouldn’t take that interface literally, because there isn’t literally a folder being moved to a literal trash bin but rather it is simply an interface that hides most if not all of what is really going on in the background — all those various diodes, resistors and transistors that are manipulated in order to modify stored information that is represented in binary code.

The desktop interface ultimately provides us with an easy and intuitive way of accomplishing these various information processing tasks because trying to do so in the most “truthful” way — by literally manually manipulating every diode, resistor, and transistor to accomplish the same task — would be far more cumbersome and less effective than using the interface.  Therefore the interface, by hiding this truth from us, allows us to “navigate” through that computational world with more fitness.  In this case, having more fitness simply means being able to accomplish information processing goals more easily, with less resources, etc.

Hoffman goes on to say that even though we shouldn’t take the desktop interface literally, obviously we should still take it seriously, because moving that folder to the trash bin can have direct implications on our lives, by potentially destroying months worth of valuable work on a manuscript that is contained in that folder.  Likewise we should take our perceptions seriously, even if we don’t take them literally.  We know that stepping in front of a moving train will likely end our conscious experience even if it is for causal reasons that we have no epistemic access to via our perception, given the species-specific “desktop interface” that evolution has endowed us with.

Relevance to the Mind-body Problem

The crucial point with this analogy is the fact that if our knowledge was confined to the desktop interface of the computer, we’d never be able to ascertain the underlying reality of the “computer”, because all that information that we don’t need to know about that underlying reality is hidden from us.  The same would apply to our perception, where it would be epistemically isolated from the underlying objective reality that exists.  I want to add to this point that even though it appears that we have found the underlying guts of our consciousness, i.e., the findings in neuroscience, it would be mistaken to think that this approach will conclusively answer the mind-body problem because the interface that we’ve used to discover our brains’ underlying neurobiology is still the “desktop” interface.

So while we may think we’ve found the underlying guts of “the computer”, this is far from certain, given the possibility of and support for this theory.  This may end up being the reason why many philosophers claim there is a “hard problem” of consciousness and one that can’t be solved.  It could be that we simply are stuck in the desktop interface and there’s no way to find out about the underlying reality that gives rise to that interface.  All we can do is maximize our knowledge of the interface itself and that would be our epistemic boundary.

Predictions of the Theory

Now if this was just a fancy idea put forward by Hoffman, that would be interesting in its own right, but the fact that it is supported by evolutionary game theory and genetic algorithm simulations shows that the theory is more than plausible.  Even better, the theory is actually a scientific theory (and not just a hypothesis), because it has made falsifiable predictions as well.  It predicts that “each species has its own interface (with some similarities between phylogenetically related species), almost surely no interface performs reconstructions (read the second link for more details on this), each interface is tailored to guide adaptive behavior in the relevant niche, much of the competition between and within species exploits strengths and limitations of interfaces, and such competition can lead to arms races between interfaces that critically influence their adaptive evolution.”  The theory predicts that interfaces are essential to understanding evolution and the competition between organisms, whereas the reconstruction theory makes such understanding impossible.  Thus, evidence of interfaces should be widespread throughout nature.

In his paper, he mentions the Jewel beetle as a case in point.  This beetle has a perceptual category, desirable females, which works well in its niche, and it uses it to choose larger females because they are the best mates.  According to the reconstructionist thesis, the male’s perception of desirable females should incorporate a statistical estimate of the true sizes of the most fertile females, but it doesn’t do this.  Instead, it has a category based on “bigger is better” and although this bestows a high fitness behavior for the male beetle in its evolutionary niche, if it comes into contact with a “stubbie” beer bottle, it falls into an infinite loop by being drawn to this supernormal stimuli since it is smooth, brown, and extremely large.  We can see that the “bigger is better” perceptual category relies on less information about the true nature of reality and instead chooses an “informational shortcut”.  The evidence of supernormal stimuli which have been found with many species further supports the theory and is evidence against the reconstructionist claim that perceptual categories estimate the statistical structure of the world.

More on Conscious Realism (Consciousness is all there is?)

This last link provided here shows the mathematical formalism of Hoffman’s conscious realist theory as proved by Chetan Prakash.  It contains a thorough explanation of the conscious realist theory (which goes above and beyond the interface theory of perception) and it also provides answers to common objections put forward by other scientists and philosophers on this theory.