The Mind Matters: The Scientific Case for Our Existence

Scientists don’t know quite what to make of the mind. They are torn between two extremes, the physical view that the brain is a machine and the mind is a process running in it, and the idealized view that the mind is non-physical or trans-physical, despite operating in a physical brain. The first camp, called eliminative materialism (or just eliminativism, materialism or reductionism), holds that only the physical exists, as supported by the hard scientific theories and evidence of physics, chemistry, and biology. At the far end of the latter camp are the solipsists, who hold that only one’s own mind exists, and a larger group of idealists, who hold that one or more minds exist but the physical world does not. Most members of the second camp, however, acknowledge physical existence but think something about our mental life can never be reduced to purely physical terms, though they can’t quite put their finger on what it is. Social science and our own intuition both assume that mental existence is something beyond physical existence. A materialist will hold either that (a) there are no mental states, just brain states (eliminativism), or that (b) there are mental states, but they can be viewed as (reduced to) brain states (reductionism). An idealist will hold that mental states can’t be reduced, which means they have some kind of existence beyond the physical, and that the physical might exist but if so it can never be conclusively proven. While the track record of the hard sciences to reduce phenomena to the physical appears impressive at first glance, their explanations are not physical, and neither are appeals to functionality (traits) in biology. Materialists think explanations are physical, being brain states, and would characterize biological traits as states in a (slower running) evolutionary system. While it is true that these functions operate in physical systems, the systems are not the functions; they only make them possible. The functions of brains and DNA are abstracted away from their underlying physical mechanisms. “Abstracted away” is called emergentism in philosophy, and refers to situations where even the full explanations of underlying mechanisms can’t provide the necessary explanatory power to account for certain behaviors that “emerge” from them. What emerges from brains and DNA is functionality, and the reason one can’t explain functionality by understanding the mechanism below is that functionality is a generalized conclusion and not a specific mechanism. Generalization is a process in which information is created by analyzing data for patterns. Information is a non-physical representation of patterns and their logic that is stored physically. I did not invent the idea that functional existence is independent of physical existence, but I do feel it is an overlooked and neglected existential state that is central to understanding the mind, so I will be developing and expanding on it for much of this book.

Before data came along the world was entirely physical; particles hit particles and followed scientific laws of the sort provided by physics and chemistry. Data is a second-order phenomenon which leverages small side-effects of events without changing the events themselves. It is theoretically possible that a logical analysis of such side-effects would reveal patterns in them that could be used to predict future events. The ability to predict the future would carry such profound benefits that any mechanism capable of doing it could do so to further its own ends, most notably to protect its own survival. At least two different mechanisms that predict the future have evolved, the first being entirely reactive and the second being proactive. Natural selection is a reactive process which measures the side effects of events blindly. It collects data about the value of a genetic trait to influence external events by statistically testing how well variations of the trait perform against each other. Because this statistical data is assessed using reproductive fitness, it takes years to millennia for adaptations to spread. Also, evolutionary “progress” is not directed but is limited to filling ecological niches with appropriate species. Animal brains, on the other hand, provide a proactive means to analyze patterns in data that can connect side effects to their underlying events to establish correlations between side effects and the effects themselves in a general way. Where a passive mechanism is general but blind to causation, this active connecting of side effects to effects actually defines causation, which put another way means to establish a predictive link between inputs that correlate to a predictive model and the outputs that one can expect. These data analysis, modeling, correlation, and matching operations will succeed based primarily on the qualities of the theoretical analysis and only secondarily on the qualities of the computing platform. The theoretical analysis, being an exercise in logic over abstract operands, is strictly nonphysical, while the computing platform is strictly physical. The operands are abstract because they are not tied strictly to any physical referent but are generalized entities (let’s call them concepts) that may be attached to a variety of possible referents depending on how we correlate and match. Put another way, indirection causes emergence. It is not at all the case that something has arisen from nothing; it is only the case that something has arisen that doesn’t directly connect back to what it came from.

Because information is decoupled this way from physical systems that employ it, we need to grant it, or more generally the function that it makes possible, a separate status of existence than physical things. I would call them physical and mental, but mental is a special case of how brains employ information, and this realm of existence extends to all abstracted entities used by information management systems to achieve function, so I will call them physical and functional. So we can conclude that scientific explanations themselves may secondarily consist of brain states, but primarily they are an abstract network of concepts which constitute an informational representation of possible sets of affairs. That is, their functional existence is paramount, and how they manifest physically in minds, computers or on paper are incidental to this function, whose existence can be discussed quite independently of any physical expression.

Having established a basis for the existence of functional entities, I will now turn my attention to physical entities. Do physical entities also have a valid claim to existence? Our continued existence as living creatures certainly seems to depend on recognizing sensory feedback as evidence of a fixed external world, and extensive analysis both personal and scientific has established the high reliability of viewing this external world as being completely fixed independent of our senses. So we can and should grant physical things a special existential status, but should not forget that it is necessarily provisional as all our knowledge of it, no matter how well corroborated, is ultimately indirect and mediated by our minds through thoughts and concepts, which exist functionally.

Ironically, considering it has a secondary claim to existence, physical science has made much more definitive, precise, and arguably useful claims about the world than functional sciences (in which I include biology and social sciences). And even worse for the cause of the functional sciences is the problem that the existence of function has inadvertently been discredited. Once an idea, like phlogiston or flat earth, has been cast out of the realm of scientific credibility, it is very hard to bring it back. So it is the case that dualism, the idea of a separate existence of mind and matter, has acquired the taint of the mystical or even the supernatural. But dualism is correct, scientific, and not at all mystical when formulated correctly. The eliminativist idea that everything that exists is physical, aka physical monism, is not exactly wrong because all physical things that exist are physical (by definition); all objects in the universe are wholly physical. But we can imagine something that doesn’t exist, and although our imagination physically lives in our brain, what we imagine still has a hypothetical existence regardless of whether we think about it or not, and that hypothetical existence is the essence of functional existence. I can think of an apple, and you can think of an apple, and completely different brain processes (in the sense that one is here and the other is there) happen. But our distinct concepts of apple will share many functional similarities, and the value of the concept APPLE (I will use the customary convention of capitalizing concepts) ultimately derives from its role in affecting our function and what we will do and not in the physical mechanisms we use to conceive of apples in our brains, which are, in this larger sense, irrelevant. Relevance itself is a fundamental property of function that is meaningless physically.

The laws of physical science can provide very reliable explanations for all physical phenomena. We are finding it very challenging to explain all the mechanisms that power biological systems, and our brains in particular, because they employ very complex electrochemical reactions, and, in the case of brains, complex networks of neural connections as well. It’s just very hard to unravel. But we are fairly sure that the mind arises as a consequence of brain activity, which is to say it is a process in the brain. The success of physical science coupled with the physical nature of the brain has led many to leap to the conclusion that the mind is physical, but if we take the mind to represent the functional aspect of the brain, then my arguments above show that the mind is not physical. Pursuing an eliminativists stance, the neurophilosopher Paul Churchland says the activities of the mind are just the “dynamical features of a massively recurrent neural network”1. From a physical perspective, this is entirely true, provided one takes the phrase “massively recurrent neural network” to be a simplification of the brain’s overall architecture. The problem lies in the word “features,” which is an inherently non-physical concept. Features are ideas, packets, or groupings of abstract relationships about other ideas, which, as I have been saying, are the very essence of non-physical, mental existence. These features are not part of the mechanism of the neural network; they are signals or information that travel through it. The same feature can be thought about in different ways at different times by different people but will still fundamentally refer to the same feature, i.e. the same functions of the feature. This “traveling” is a consequence of complex feedback loops in the brain that capture patterns as information to guide future behavior. This information is the basic unit of function, which is decoupled from and so independently existing from the physical. Physical and functional existence together form a complete ontology (philosophy of existence) which I call form and function dualism. While philosophers sometimes list a wider variety of categories of being (e.g. properties, events), I believe these additional categories can all be reduced to either form or function and no further.

Because physical systems that use information (which I generically call information management systems) have an entirely physical mechanism, one can easily overlook that something more is happening. Let’s take a closer look at the features of functional existence. Functional existence is not a separate substance in the brain as Descartes proposed. It is not even anywhere in the brain because only physical things have a location. Any given thought I might have simultaneously exists in two ways, physically and functionally. The physical form of a thought is the set of neurons thinking the thought, including the neurochemical processes they employ. While we can pinpoint certain neurons or brain areas that are more active when thinking certain thoughts, we also know that the whole brain and arguably the whole body participate in forming the thought. The thought also has a function, being the role or purpose it serves. This property, clearly not physical and seemingly intangible, is tangible (touchable) through the impact it can have in aiding prediction, so it is a kind of existence we can talk about and which affects us. A thought’s purpose is to assist the brain in its overall function, which is to control the body in ways that stand to further the survival and reproduction of the gene line. “Control” broadly refers to predictive strategies that can be used to guide actions, hopefully to achieve desired outcomes more often than chance alone would. And even seemingly purposeless thoughts participate in maintaining the overall performance of the mind and can be thought of as practice thoughts if nothing else. For a predictive strategy to work better than chance, it needs to gather information, i.e. to correlate past situations to the current situation based on similarities. The relationship between information and the circumstances in which it can be employed is inherently indirect, and abstractly so (abstract: “not based on a particular instance; theoretical”, meaning the information is general and not specific). We store this general information in one of two ways, genetically or neurally (i.e. via a neurochemical memory system). As noted above, genetic information is collected and processed reactively through heritability, and neural information is processed proactively using neural data analysis. Genetic information directs all the systems of the body, but also specifically directs the control systems of the brain to give us instincts, which go beyond simple urges to include things like giving us a visual and sensory understanding of our surroundings. We also create neurally-stored memories of our personal experience to record our interactions with the world. Not everything our instincts and memory enables us to do will be useful, but they do enable us to do some useful things, and this capacity is the measure of their existence as function.

The brain’s hardware is only very indirectly connected to its function, the same way software’s function is only indirectly connected to the computer hardware on which it runs. A nearly infinite string of genetic feedback stretching back millions of years designed our brains to achieve functional objectives by progressively selecting and adapting physical mechanisms based only on their ability to help. This kind of results-oriented design can produce complex mechanisms that are hard to unravel, but functions with specific mechanical and algorithmic needs can be pinpointed. The nerves that conduct sensory information from the eyes, ears, nose, etc. to specific brain areas dedicated to those senses have been identified. We have a sense of the kinds of processing the brain does to see color, 3D, detect edges, recognize objects, fill in our blind spots, etc., so the localized study of the mechanisms of vision and the other senses may well come to provide a comprehensive explanation of their function as well. All of the brain’s genetically-based functionality will eventually be explained pretty well by this kind of physical study alone, but we are still quite a ways off from that point and what our minds do is not driven solely by genetics. Everything we’ve learned stretching back to our conception is stored in some neurochemical way which is itself entirely genetic and explainable, but the contents of what we have learned are, like software, not knowable by studying our gene-based hardware. Until recently our sole method to access the brain’s informational content was through use, by observing behavior, listening to verbal accounts of knowledge, or contemplating the thoughts in our own heads. But now several techniques have been developed to see which brain areas and somethings which individual neurons activate as people do or think specific things. While this is certainly a good first step to reading a person’s mind, we are a long ways off from reading thoughts objectively as well as we can subjectively. I can certainly imagine that a machine learning algorithm given such information could become quite good at predicting what we are thinking about and even what we will do next, but this too will fall far short of understanding why.

Function in the mind doesn’t arise through a single mechanism as it does in a computer’s CPU, one instruction after the next. It is a highly distributed process in which genetically supported functions are activated chemically, electrochemically or even mechanically, while neural functions are activated neurochemically. Even so, it is useful to draw an analogy between the brain’s mechanisms and computer hardware and between learned knowledge and software. We can’t completely distinguish the hardware and software aspects of any brain function, but we can do so approximately. Knowledge extends the innate power of the brain the same way an advanced civilization outperforms an aboriginal one. Extending the analogy, software is more than just a series of instructions. Each instruction embeds functionality designed into the hardware, and instructions are then bundled into operating system subroutines and library functions that make high-level functions possible. Our brains perform the low-level functions subconsciously, which is to say outside direct conscious awareness or control. From habit, we can learn to perform even high-level functions with little or no direct conscious awareness or control. We can talk about high-level functions of the mind or software independent of the hardware or low-level routes, but to understand it well we need to know what the underlying mechanisms are doing. Physical architectures that make function possible both limit and enable what is possible, but function still retains a fundamentally non-physical character. What function makes possible could always be accomplished using different hardware. Also, form doesn’t drive function, function drives function, meaning that what needs to be done to achieve better and better results is the force that impels evolution and human design, not the toolkits available to do them. Consequently, I will be focusing on the mind more than the brain, i.e. the function more than the form, as I develop explanations.

Before I set out to develop theories to explain the high-level functions of the mind, I’d like to bring up some caveats. First, theories themselves are high-level functions of the mind, so I will be explaining the mind by using the mind. This is a bit of a catch-22, but we don’t hesitate to use theories to explain physics and chemistry, so what is good for the goose must be good for the gander. Second, I have admitted that I am going to propose theories to explain function without a detailed theory to explain the physical brain mechanism behind it. Again, I claim that all of science proceeds this way, starting with sketchy explanations of mechanisms and refining them as new evidence comes in. So long as my theories are plausible given our physical knowledge, they are sufficient. And third, there are many ways to explain anything, each creating a perspective that covers different aspects (or the same aspects in different ways). Science aims to merge mutually consistent perspectives into overarching theories based on a consistent set of underlying evidence, but it is not always practical or possible to merge them all at a given point in time. So we should expect and encourage a proliferation of theories on the frontier. In the case of the mind, there are still many ways we could describe the master control program and high-level functions of the brain, and I will present one scheme here, as unified as I can make it.

A stream has no purpose; water just flows downhill. But a blood vessel is built specifically to deliver resources to tissues and remove waste. Lifeforms have many specialized mechanisms to perform specific functions, but a single animal body can’t perform all its support functions simultaneously, except the most degenerate, sessile animals like sponges. A mobile animal needs to prioritize where it will go and what it will do. This functional requirement led to the evolution of animal brains, which, as noted above, collect information about their environment through senses and analyze it to predict likely outcomes from different interactions. This information can be used to plan actions. This kind of dynamic, proactive data analysis generalizes functional aspects of the environment into concepts which can be manipulated functionally as distinct entities. So an animal can thus interpret sensory data as food, friend or foe nearly instantaneously. In doing so, it has moved out of the realm of physical interactions into the realm of functional interactions, which are cut off from the physical by a barrier of indirection. Any functional interaction requires sensory processing that leads to recognition using stored models, evaluation of possible reactions, and prioritization against other reactions. These are not just “knee-jerk” reactions; they require complex information processing. Most of the information processing needed to do them is innate, and in some simple animals it may be entirely innate, but the advantages that learning and memory provide to tune responses to specific situations are so great that nearly all animals have some capacity to learn. Animals are better equipped to learn some things than others. For example, humans more readily store sensory experiences than abstract information and have distinct innate mechanisms for storing procedural vs. declarative memories.

What interests us most about the mind is not the huge fraction we generally take for granted that is pretty similar across all mammalian brains, it is the part that makes humans special. The richness of our lives and even our motivation for living depend on our senses, motor skills, emotions, drives, and awarenesses (of time, place, etc.), so I don’t want to mitigate their significance to our lives or their contribution to the mind overall. All mammals (and to some degree all sentient creatures) have those things, but they are not enough to form our sense of self, a sense which we are pretty sure is stronger in people than in animals. While arguably sheer arrogance, I am going to claim that this sense in humans derives from our greater capacity for abstract thought. Beyond just controlling themselves and our lives, people can also freely associate seemingly any ideas in their heads with any others ad infinitum, giving us the potential to see any situation from any angle. Whether and how this association is really free is a matter I will explore more later, but our flexibility with generalizations lets us leverage knowledge in vastly more ways than if we only applied our knowledge narrowly, which we can infer happens with animals based on their behavior. Some consider language to be at the root of human intelligence, and some look to our facility with procedural or spatial thinking, but the lever that drove these was the evolutionary expansion of abstract thought.

Though our mechanisms to learn and recall are innate, we have created an entirely artificial world of cultural knowledge that supports our civilization. We each adopt our share of cultural knowledge, but we also develop a store of personal knowledge specific to our own interactions with the world. What is all this knowledge and what makes it abstract? While we will in time come to understand the neurochemical mechanisms that support abstract thinking and memory, just as we will for vision processing, knowing how they work will say much less about what ideas we have created with them than vision processing says about what we see. This is because abstraction opens the door to the imagination, letting us create fictional worlds where anything might happen, while vision aims for as much realism as possible. Consider an analogy. If vision and abstract thinking are both higher-order functions like software libraries, then the vision library performs a specific function while the abstract thinking library provides a general toolkit one can use to perform a wide, even unlimited, range of functions. This library performs a higher-order process we call “thinking” that employs abstract data we call “concepts” to model general-purpose problems. General-purpose computer toolkits like this include programming languages (e.g. C++, Java) and software engines, of which there are now hundreds for developing games. The actual instructions computers use are in machine language, but we program them in programming languages or engines which package up commonly used functionality into a convenient and seamless environment. In the same way, our reasoning capacity provides us with a convenient and seamless way to use many of our brain’s underlying capabilities without worrying about the details. Although convenient, this also means we can’t “think our way out of the box”; we can only think using the small, tailored subset of functions available to our conscious minds.

While our linguistic, procedural, and spatial thinking leverage our enhanced facility for abstraction, they are still innate capacities. The crown jewel of our abstraction toolkit is the ability to reason. Reasoning is what extends our power to know things to knowing that we know them. Reasoning pushes the level of indirection inherent in all knowledge back one or more levels, to become self-referential or arbitrarily referential. Reasoning doesn’t just let us create new abstract concepts; it lets us connect them logically through deduction (entailment; cause and effect), induction (weight of evidence), or abduction (recognition or intuition). The ability to reason abstractly is the critical skill we would expect to find in a robot seeking to pass the Turing test, which entails convincing a human that it is human via written communication. If a robot could reason abstractly as well as we can, we would have to grant it intelligence. It would need to understand and have programming for senses, emotions, drives, etc., as well, but I posit that reasoning is the most significant and overarching skill.

In granting us the ability to reason abstractly, evolution opened a door for us to a much larger space than the physical world: the unbounded world of imagination. We know animals are sentient, feel pain and experience the world in much the same way we do, but we rather doubt their imaginations are really comparable to our own. When religions refer to the divine, it is this attribute in ourselves to which they refer. This nonphysical capacity, abstract reasoning, gives us arbitrary power over nature and our own fortunes, both in an imaginary sense and, through our actions, in a physical sense. It is hard to keep in mind, and counterintuitive, but the imaginary sense matters more than the physical sense. That is why we turn to religion, to remind us that the greatest part of us is not our physical bodies and deeds, but our internal world and the ways it connects us to each other and all living things. I am not saying this for the sake of spirituality or religion, but only to acknowledge a fact about our existence. Cogito ergo sum is not just a cool idea without much practical application, we are primarily beings of thought and only secondarily bodies in an environment. Minds made capable of self-awareness through a sufficiently powerful facility for abstraction and reason have to come to terms with this fact. We are not just physical organisms preserving a gene line; we are now also mental organisms floating in seas of possibility, detached from time and space through the force of conjecture. Because ideas are timeless, we are timeless.

Each religion captures the essence of this idea in its own doctrinal way, but they have in common a reverence for this divine capacity, because the important point here is there is more to us that we can ever see. Our imaginations that the ties that bind us to the world run deeper through the planes of possibility than we can perceive through a single stream of consciousness, but they are the biggest part of us, not our bodies. We sense it, we are humbled by it, and we take guidance from it because our actions are not just in the service of evolutionary (physical) goals but also in the service of the network of this abstract power cultivated by nature and nurture and handed down to us through countless generations. Is it perhaps going too far to suggest that minds and possibly free will elevate us above evolution’s thrall? As we become post-evolutionary, about to design the cybernetic organisms that will replace us, discussions about the conventional mechanisms of evolution become moot. Ultimately, we transcend evolution for the same reason we have free will: because Pandora’s box has been opened and anything might happen. But that is a deeper discussion I will develop more later.

Let me back up a bit. Our mental states are special cases of function that are highly tuned to meet our needs, but to understand them we need to dig a lot deeper into at the existential nature of function. I initially said that information is the basic unit of function, and I defined information in terms of its ability to correlate similar past situations to the current situation to make an informed prediction. This strategy hinges on the likelihood that similar kinds of things will happen repeatedly. At a subatomic level the universe presumably never exactly repeats itself, but we have discovered consistent laws of nature that are highly repeatable even for the macroscopic objects with which we typically interact. Lifeforms, as DNA-based information management systems, bank on the repeatable value of genetic traits when they use positive feedback to reward adaptive traits over nonadaptive ones. Hence DNA uses information to predict the future. Further adaptation will always be necessary as circumstances change, but some of what has been learned before (via genetic encoding) remains useful indefinitely, allowing lifeforms to expand their genetic repertoire over billions of years. Some of this information is made available to the mind through instincts. For example, the value of wanting to eat or to have kids has been demonstrated through countless generations, and having a mental awareness of these goals and their relative priority is central to the mind’s function. In the short-term, however, that is, over a single organism’s lifetime, instincts alone don’t and can’t capture enough detailed information to provide the level of decision support animals need. For example, food sources and threats vary too much to ingrain them entirely as instincts. To meet this challenge, animals learn, and to learn an animal must be able to store experiential information as memory that it can access as needed over its lifetime. In principle, minds continually learn throughout life, always assessing effects and associating them to their causes. In practice, the minds of all animals undergo rapid learning phases during youth followed by the confident application of those lessons during adulthood. Adults continue to learn, but reacting quickly is generally more valuable to adults than learning new tricks, so stubbornness overshadows flexibility. I have defined function and information as aids to prediction. This capacity of function to help us underlies its meaning and distinguishes it from form, even though we need mechanisms (form) to perform functions and store information. But form and function are distinct: the ability to predict has no physical aspect, and particles, molecules, and objects have no predictive aspect.

With that established, we can get to the most interesting aspect of the mind, which is this: brains acquire and manage information through reasoning and intuition, but minds only exist because of reasoning (I will get to why in a moment). Reasoning is the attribution of causes to effects, while intuition covers all information acquired without reasoning, e.g. by discerning patterns and associations. So reasoning is how the mind employs a causative approach while intuition is how it uses a pattern analysis approach. All information is helpful either because it connects causes to effects or because it finds patterns that can be exploited. The first approach is the domain of logic, while the second is the domain of data analysis. Reasoning is conducted using atomic units of information called concepts. A concept is a container or indirect reference that the mind uses to represent (stand for) something else. Intuition doesn’t use atomic information but rather stores and extracts information based on pattern analysis and recognition. For example, recognition is the subconscious matching of input data against our memory that causes a single concept to pop out, e.g. a known object or word. We often use words to label concepts, though most are unnamed. Every word actually refers to a host of concepts, including its dictionary definition(s) (approximately), its connotations, and also a constantly evolving set of variations specific to our experience. Concepts are generalizations that apply to similar situations. BASEBALL, THREE, and RED can refer to any number of physical phenomena but are not physical themselves. Even when they are applied to specific physical things, they are still only references and not the things themselves. Churchland would say these mental terms are a part of folk psychology that makes sense to us subjectively but have no place in the real world, which depends on the calculations that flow through the brain but does not care about our high-level “molar” interpretation of them as “ideas.” Really, though, the mysterious, folksy, molar property he can’t quite put his finger on is function, and it can’t be ignored or reduced. Brains manage information to achieve purposes, and only by focusing on those purposes (i.e. by regarding them as entities) can we understand what it is “really” doing. Concepts, intuition, and reasoning are basic tools the brain uses to achieve its primary function of controlling the body. But what is it about concepts and reasoning that creates the mind? Why can’t we just go about our business as unaware zombies?

The mind and the brain are not the same. Just distinguishing form from function is not enough to separate the mind from the brain. Holistically, all brain function can be called mind. That is, the actions our body takes as directed by the brain accomplish the functions that define the mind. In some cases, those directions may simply result from chemical triggers, e.g. hormone regulation, and not neural processing, but mental control is ultimately a combination of genetic and neural information processing, so holistically we need to lump all this feedback-mediated control under the label “mind.” But this holistic perspective is not the one of most interest to us. When we think of the mind, we are mostly thinking of our first-person capacity for awareness, attention, perception, feeling, thinking, will and reason. Collectively, we call these properties or mental states agency, and we call our human awareness of our own agency self. We currently have no scientific explanation for why we seem to be agents experiencing mental states, yet computer programs (and zombies) have no such states but can still do things. Some scientists have concluded that this means our subjective lives are illusions that are immaterial to our behavior, and evidence that our conscious minds only become aware of our actions slightly after the neural signals to perform them have happened seems to bolster this view. This conclusion is wrong; our subjective states are the product of very real processing happening in our brains, and this processing does control what we do and is not at all incidental to it. A subset of our brain’s information processing is dedicated to creating the mental states we experience, because agency is an effective way for animals to use reasoning, arguably the most effective way.

In more detail, concepts are generalizations and generalizations are fiction, literally figments of our imagination. These figments form imaginary worlds comprised of concepts defined in terms of other concepts via deductive, inductive and abductive links. The links are also imaginary; causation and patterns carry generalized information about things and events and are not the things or events themselves. This all happens with or without consciousness because simulations of the world via imaginary worlds are (arguably) the best route to predicting what will happen in the real world. One of these imaginary worlds — let’s call it the mind’s real world — has a special status: it is our best guess from current sensory information about what is happening in the real world. It is still imaginary; we are really inside our heads and not out in the world. Since the mind’s real world only exists in the present (and also the past to the extent we remember it), we can only predict the future using imaginary worlds, but we will try to choose imaginary worlds that seem plausible when making decisions about the real world. It is ultimately the mind’s whole job to make decisions based on such predictions, and one of the tools it uses is reason. But reason only works in a single stream, drawing conclusions one at a time from premises. Physical reality, on the other hand, unfolds in parallel because the waves and particles that comprise it act independently. Brains are physically located in bodies that also only work in a single stream, although they can potentially perform several activities concurrently if they can operate their extremities independently. The octopus, in particular, has a fairly independent brain for each arm, which makes sense because it is often helpful and practical for the arms to pursue independent objectives. But for most animals a single stream of reasoned decisions governing a coordinated stream of bodily actions works best. This is not to say that all decisions need reason or need the active oversight of reason. Many, even most, of the control functions of the brain either happen independently of reason or are habituated and are delegated to subconscious control outside or at the periphery of conscious awareness. This is also not to say that our reasoned decisions control or supersede all subconscious functions; quite the contrary, we are generally quite satisfied to let subconscious controls do what they do best. But the point I am driving toward is that there are good reasons for this top-level single stream of reasoned decisions to be processed as the subjective mental states I am calling agency. First, most bodies act as single agents most of the time, i.e. achieving one functional goal at a time. So it makes sense to generalize about an animal that it is currently eating and defending its food; these characterizations create a story that can help both the animal and others observing it make reasoned decisions. The mental capacity to recognize agency in others is called “theory of mind” (or TOM; not to be confused with the theories I am presenting about the mind). Second, the top-level reasoning process that either makes decisions or confirms decisions made subconsciously is itself an agent (the conscious mind itself) that must reconcile its book of reasoning with the single stream of actions the body performs. The way we feel free will leads us to believe that we willfully selected the actions we undertook, as opposed to simply observing actions our subconscious minds made for us. This is not because we actively make every decision on the spot using rational processes, it is because we can. The conscious mind is a supervisory process; it ultimately delegates all the nonsupervisory work to underlings, i.e. to subconscious processes. Most of what we do during the day is fairly habitual; we have previously established many techniques and preferences for things we like from a supervisory perspective, and we only have to nudge our minds in the right direction and subconscious processing takes care of the details. Even the explicit decisions we make throughout the day are enacted because our conscious minds “ask” our bodies to do our bidding, and subconscious processes which have been habituated to stimulate our nerves in just the right ways actually move our extremities. Experiments that demonstrate we the order to hit a button when we see a specific event comes from the subconscious and a split second later is “made” by our conscious minds should not surprise us at all. What has really happened is that we consciously habituated (programmed) the subconscious to act that way; it has been preapproved to behave a certain way, and the subconscious can react quicker if habituated this way than we can consciously. If we were to test our mental response in any situation where new rational thought is applied we will find that the subconscious waits for the decision to come down before acting. And third, I have to note that just because we can consciously control our actions with rational thought, this doesn’t mean we always will. Pressures from the subconscious to act in ways contrary to our conscious, rational stream of thought are immense and we will often succumb to them. Such pressures never provide us with a reason why, but we develop a rationalized view of our instinctive drives, emotional needs and moral responsibilities so we can incorporate them into our rational decision making processes. Evolutionary theory has started to provide scientific rationales for the adaptive value of these subconscious pressures which we can potentially incorporate into our rational thinking to make better decisions.

Summarizing, the rational stream of thought at the center of our conscious experience is only a small part of what the brain is doing, but it is the part coordinating overall activities. It appears to us as subjective theater principally because the job of the process is inherently first-person: it collects sensory and internal inputs, weighs them against priorities, and produces desirable actions as outputs. Although it is therefore first-person by construction, this is not enough for it to “know” it is first-person, which arises because it interprets the inputs through intuition and reason. Intuition just leverages data in useful ways without any attempt at explanation. Reason, however, condenses the inputs into generalized elements that create our cartoon-like imaginary worlds and the mind’s real world. Elements that sit on deep stores of intuitive knowledge feel much more real that purely abstract concepts (so APPLE feels more real than SPHERE). Our first-person perspective arises at the moment our supervisory process starts to act as if the mind’s real world is itself the real world. As humans, we know that they are different, but as subjects, we are convinced that our senses are themselves real, and, by extension, that all the objects we distinguish as discrete in the world are in fact discrete and bearers of the qualities we ascribe to them. In lower animals, useful behaviors can be controlled genetically with little or no imagination or reasoning. It is probably fair to say that the vast majority of behavior (and the mental processing behind it) in all animals, including us, is controlled genetically, but a fraction is controlled through neural reasoning processes in a proportion that roughly correlates to brain size and complexity. It is this fraction that creates minds that feel joy, pain, or anything, for that matter, because the brain has dedicated processing power to create a subjective theater with mental states as a way to make the mind’s real world pull strings effectively in the actual real world. Note that the “quality” of our mental states is a computational fiction (i.e. subjective), but the objective meaning of these states comes from how they help us function. Things feel like how they influence our behavior. We shy away from frightening/cold/hot things; we jump into happy/warm/cool things.

But why does our supervisory process start to act as if the mind’s real world is itself the real world? In a word, confidence. Confidence is the ability to make effective decisions quickly, taking all the available information into account, prioritizing what matters most, and choosing the action that will probably work best. Unlike chemical regulatory mechanisms or sensory information processioning, which can run continuously, reasoning is discrete. Reasoning simplifies a complex situation into a small number of relevant actors and the relationships between them from which logical implications can be deduced. In any situation, people (and animals) are keenly aware of the most relevant issues they face, especially those with immediate urgency. They draw on experience and project likely outcomes combining both overt rational logic and intuition, which is largely a memory-based lookup of successful strategies. Reason and intuition together create the confidence that a decision will succeed to an expected degree. This doesn’t eliminate doubt; every decision is also accompanied by doubt, but once we reach a level of confidence that says acting now is the best way to manage all our priorities, we will act. We are most confident performing repetitive actions that have succeeded so many times we have little reason to doubt them. We are naturally least confident performing actions which we suspect might fail or which we have never done before. We can’t avoid acting in these situations forever if our long-term goals depend on appropriate action. What we can and invariably will do in such situations is to analyze them further. We start thinking them over from many perspectives, drawing on both rational models and intuitive sensibilities. The amount of analysis depends on the situation. Every day we face a number of situations for which a few moments analysis brings us to a sufficient level of confidence that we can act. For more challenging goals we will procrastinate, giving ourselves more time to develop a good strategy rather than leaping before we have adequately looked. We run the risk of analysis paralysis, which is procrastination to the point where we are causing harm by not acting. To avoid reaching this point, we need to weigh the benefit of action against the possible harm, which itself can lead to analysis paralysis because any one goal must compete for our attention with all other goals. Things typically work out because we maintain a short list of personal priorities we must accomplish and we see to it that they are addressed. But as we look in a more open-ended way to the future our list of possible goals becomes infinite as there are an unlimited number of projects we might do to benefit ourselves and/or others. Analysis paralysis shuts down most of these avenues before we can even contemplate them because we only have so much time, so we invariably focus on the manageable subset that is most personally significant to us. We will have vague opinions on all possible goals, at least to the point that we are satisfied that they don’t require our attention. This brings us back to the goal of subjectivity. Subjectivity is an immersive, hands-on approach for establishing what goals we should pay attention to and what means to attain them will best manage all our priorities. By providing a theater in which we can reason, subjectivity can be viewed as a regulatory mechanism that keeps objectivity (reason) on track.

Non-sentient algorithms, such as a self-driving car might have, would lack confidence in this sense. Their algorithm would dictate the best strategy, and so long as their stored experience is sufficient to handle the driving challenges they encounter, they will do fine. But outside that range they can do nothing. For example, if the car in front of you stalled on the tracks with an oncoming train 30 seconds away, you could use your car to push the other car off the tracks, but if your car were self-driving with no experience in pushing cars, it would be helpless. In a novel situation, a human can figure out what to do but a program operating within a range of experience can’t. While we could, in principle, program machines to figure things out, the bigger gap is that they need to recognize what problems need to be solved, which depends on a sense of what is important. A human can immediately tell there is a risk of death to people in the cars and on the train, and, of somewhat less concern, a risk of damage to the cars and the train. A lifetime of subjective prioritization using feedback from instinctive preferences has honed this model to perceive and weigh the risks accordingly. Simply programming an algorithm to protect people first and property next doesn’t begin to address the complex subtleties of our prioritization model. Yes, we instinctively prioritize the safety of ourselves, our family, our tribe, strangers, and property in that order, but beyond such crude directives, we refine our priorities over a lifetime of first-person interactions in which the way each situation unfolds relative to our first-person priorities impacts them going forward. Without this implied “me,” as in “What does this do for me?”, we essentially have no skin in the game and thus no basis for preferring one action to another. It is the difference between Soviet five-year plans and a market economy. One is preprogrammed, but has no adaptability to circumstances, while the other evolves to changing circumstances. It is hard to imagine an algorithm for managing a mind and its body that could be more effective than one that principally interprets all top-level interactions with the world in terms of the holistic impact to itself via a subjective stance.

But why, exactly, should the agent approach be better than the (robotic) alternatives? This is a logical consequence of there being one brain to control one body. At any point in time, the parts of the body can each only undertake one action, so one overall, coordinated strategy must govern the whole body. At the top level, then, the brain needs to come to an unending series of decisions about what to do next. Each of these decisions should draw on all the relevant information at the animal’s disposal, including all sensory inputs and any information recorded instinctively in DNA or experientially in memory. With the agency approach, this is accomplished by having a conscious mind in which an attention subprocess focuses information from perception, feeling and memory into an awareness state on which thinking, feeling and reason can act to continually make the next decision. An enormous pool of information is filtered down to create consciousness, and it is specifically done to provide the agent process of the brain (i.e. consciousness) as logically simplified a train of thought as possible so that it can focus its efforts on what is relevant to making decisions while ignoring everything else. This logical simplification can be thought of as creating a cartoon-like representation of reality that captures the aspects most relevant to animal behavior in packets of information — concepts — for which generalized rules of causation can be applied. Intuition, which includes a broad set of algorithms to recognize patterns, can’t by itself process concepts using logical rules such as cause and effect. Reasoning does this, and the network of concepts it uses to do it creates the agent-oriented perspective with which we are so familiar. The addition of abstraction elevates this agency to the level of human intelligence. So, as I said above, we would recognize a robot that could demonstrate abstraction as being intelligent, but abstraction is a development of reasoning, not intuition, so the robot would need to be reasoning with a relevant set of concepts just as we do. Does this imply it would possess agency? If it were controlling a body in the world, then yes, I think this follows, because its relevant set of concepts would be akin to our own. It might subdivide the world into entirely different concepts, but it would still be using a concept-based simplification derived from sensory inputs that probably depends principally on cause and effect for predictive power. The distinct qualia (sensations) that make up our conscious experience are physically just information in the form of electrochemical signals. But each quale feels distinctive so the agent can tell them apart. We also have innate and learned associations for each quale, e.g. red seems dangerous, but the distinctiveness is the main thing as it lets a single-stream train of thought monitor many sensory input channels simultaneously without getting confused. Provided our putative robot had distinct streams of sensory inputs feeding a simplified concept-based central reasoning process then that distinctiveness could be said to be perceived as qualia just like our own. Note that intuition happens outside of our conscious control or awareness and so does not need qualia (i.e. it doesn’t feel), though it can make use of the information. We only have direct conscious awareness of a small amount of the processing done in our brains, and the rest of the processing is subconscious. I will use intuition and subconscious synonymously, though with different connotations. Reasoning and conscious are not synonyms because the conscious mind can access much intuitive knowledge and so uses both reasoning and intuition to reach decisions. Our mind seems to us to be a single entity, but it is really a partnership between the subconscious and the conscious. The conscious mind can override the subconscious on any deliberated decision, but to promote efficiency the simplest tasks are all delegated to the subconscious via instinct and learning. Though we feel like subconscious decisions are “ours”, we may find on conscious review that we don’t agree with them and will attempt to retrain ourselves to act differently next time, essentially adjusting the instructions that guide the subconscious.

Before I move on I’d like to explain the power of reason over intuition in one other way. If most of our mental processing is subconscious and does not use reason, and we can let our subconscious minds make so many of our daily decisions on autopilot, why do we need a conscious reasoning layer at the top to create a cartoon-like world? Note that our more complex subconscious behaviors got there in the first place from conscious programming (learning) using concepts, so although we can carry out such behaviors without further reasoning we used reasoning to establish them. The real question, though, is whether subconscious algorithms that glean patterns from information could theoretically solve problems as well, eliminating the need for consciousness. While people aren’t likely to change their ways, an intelligent computer program that didn’t need code for consciousness would be easier to develop. Let’s grant this computer program access to a library of learned behavior to cover a wide variety of situations, which is analogous to the DNA-based information the brain provides through instinct. Let’s further say the program can use concepts as containers to distinguish objects, events, and behaviors. Such a program could know from experience and data analysis how bullets can move. They can stay still, fall, be thrown, or be fired from a gun at great speed. Still things generally touch other things below them, falling things don’t touching other things below them, and thrown and fired things follow throwing and firing actions. What is missing from this picture is an explanatory theory of cause and effect, and more broadly the application of logic through reason. The analysis of patterns alone does not reveal why things happen because it doesn’t use a logical model with rules of behavior. The theory of gravity says that earth pulls all things toward it, and more generally that all things exert an attractive force to each other inversely proportional to the square of their distance apart. The weakness of physical intuition compared to theory is made clear by the common but mistaken intuition that the speed that objects fall is proportional to their weight. Given more experience observing falling objects one will eventually develop an intuitive sense that aligns well with the laws of physics, but trying to do science by studying data instead of theorizing about cause and effect relationships would be very slow and inconclusive. The intuitions we gather from large data sets are indispensable to our overall understanding but are only weakly suggestive compared to the near certainty we get from positing laws of nature. The subconscious is theory-free; it just circulates information looking for patterns, including information packaged up into concepts. When it encounters multiple factors in combinations it has not seen before, it has no way of predicting combined effects. In the real world, every situation is unique and so has a novel combination of factors. Reasoning with cause and effect can draw out the implications of those factors where pattern analysis could only see likelihoods relative to possibly irrelevant past experience.

A self-driving car must be able to evaluate current circumstances and select appropriate responses. While we have long had the technology to build sensors and mechanical or computer-based controllers, we haven’t been able to interpret sensor data well enough to replace human drivers. Machine learning has solved that problem, and we can now train algorithms using thousands of examples to recognize things. This recognition mirrors our subconscious approach by using data and positive feedback. Self-driving car algorithms plug recognized objects into a reason-based driving model that follows the well-defined rules of the road. To ensure good recognition and response in nearly any circumstance, these programs use data from millions of hours of “practice”. What they do is akin to us performing a learned behavior: we collect a little feedback from the environment to make sure our behavior is appropriate, and then we just execute it. To tie our shoes we need feedback to locate the laces and ensure the tension is appropriate through the process, but mostly we don’t think and it just happens. We need to be able to reason to drive well because we have to be prepared to act well when we encounter new situations, but a self-driving car, with all of its experience, is likely to have seen just about every kind of situation it will ever encounter and already has a response ready. That overwhelming edge in experience won’t help when it encounters a new situation that reason could have easily solved, but even so self-driving cars are already 20 times safer than humans and will soon be over 100 times safer, mostly because humans make more mistakes. Although computer algorithms still can’t do general purpose reasoning, our reasoning processes have lots of subconscious support, so applying machine learning to reasoning will continue to increase the cleverness of computers and may even bring them all the way to abstract intelligence. My goal is to unveil the algorithm of reason, to the extent that this can be done using reason. That will certainly include crediting subconscious support where it is due, but more significantly it will expose the role and structure of consciousness.

All animal minds bundle information into concepts through abstraction for convenient processing by their conscious minds. Abstract thought employs conceptual models, which are sets of rules and concepts that work together to characterize some topic to be explained (the subject of prediction). We often perceive conceptual models as images or representations of the outside world “playing” inside our heads. While we can’t exactly describe the structure of conceptual models, we can represent them outside the mind using language or a formal system. Formal systems, which often employ formal languages, can achieve much greater logical precision than natural language. But what both formal and natural languages have in common is that they treat concepts atomically. We ultimately need intuition, i.e. subconscious skills, to resolve concepts to meanings. Yes, we can reason out logical explanations of concepts in terms of other concepts, but these explanations can only cover certain aspects and invariably miss much detail that we grasp from consideration of our immense body of experience for any given concept, for which we depend on subconscious associations. Again, the bicameral mind (a partnership between the subconscious and the conscious, not the speaking/listening division proposed by Julian Jaynes2) feels to us quite unified even though it actually blends intuitive understandings based on subconscious processes with rational understandings orchestrated by rational, conscious processes. From this, we can conclude that formal systems simplify out a critical part of the model. Natural language also simplifies, but words carry subtleties through context and connotation. Mental models combine all of our intuitive, subconscious-based knowledge with the reasoned concept-based knowledge we use in conceptual models. Put another way, conceptual models manage the cartoon-like, simplified view at the center of reasoning, while mental models combine these logical views with all the sensory and experiential data that backs them up.

The logical positivists in the 1930’s and 1940’s claimed that all scientific explanations could fit into a formal system (called the Deductive Nomological Model), which basically said that scientific explanations follow solely from laws of nature, their causes, and their effects. The first flaw with this theory was that it committed the category mistake of equating function with form. Scientific explanations, and all understanding, exist to serve a function, which is to say they have predictive power and consequently are carriers of information. That which is to be explained, the explanandum or form, is explained by an explanans or function. It is not that the form doesn’t exist in its own right, it is that our only interest in it relates to what might happen to it, its function. The second flaw with the DN Model was that it presumes that explanations only require a deductive or logical approach, but as I explained above, patterns are fundamental to comprehension as they set the foundation that connects the observer to the observed. Logic may be perfect but can only be imperfectly married to nature, a connection established by detecting and correlating patterns. While postpositivists have tried to salvage some of the certainty of positivism by acknowledging that human understanding introduces uncertainty, but they can’t because the real problem is that function doesn’t reduce to form. No matter how appealing, scientific realism (the view that the universe actually exists) is irrelevant to science. Science is indifferent to the noumenal (what actually exists); it is concerned only with the phenomenal (that which is observed) and what we can learn from observation. Form and function dualism gives postpositivism solid ground to stand on and is the missing link to support the long-sought unity of science. I contend that functional explanations are always partial by their nature, providing the ability to predict the future better than chance but guaranteeing nothing. It is consequently unfair to call such explanations “partial” because there is no such thing as a “full” explanation.

  1. Paul Churchland, Neurophilosophy at Work, Cambridge University Press, 2007, p2
  2. Julian Jayne’s, Bicameralism (psychology), The Origin of Consciousness in the Breakdown of the Bicameral Mind, Houghton Mifflin Harcourt, 1976

1 thought on “The Mind Matters: The Scientific Case for Our Existence”

Leave a Reply