Free Agents

How Evolution Gave Us Free Will

by Kevin Mitchell

Rating: 10/10

Summary

In this book Kevin Mitchell explains how evolution produced living things that can make choices for their own reasons and why this means that we have free will, even if our cognitive machinery is entirely based on the physical instantiation of our brains.

This is possible because physics is not enough to determine everything about a system for two reasons:

quandum indeterminancy: which means that the lowest levels of physics behaves fundamentally randomly
non-linear dynamics: which means that systems as a whole can be unpredictable, even if governed by known equations, because they are hyper sensitive to their input conditions

These two acting together create a world that is not fully specified everywhere all at once since the Big Bang... they rule out determinism, our universe is not deterministic and the possibility of choice is a result of this. Low level quantum fluctuations can be amplified by chaos effects to create wildly different outcomes. In other words: there is wiggle room for higher order systems (like living things), to create their own causal power because the world itself is not 100% pre-determined. They can fix their own systems into specific modes of action, because of the informational states they represent within. In a way, living organisms can stabilize the chaotic nature of physical reality into attractor states that have meaning and represent useful information about the world. And they can decide to act on these representations. This is not circular, because the system arises in time through natural means, but then evolves according to it's own rules, once it is there, through active work. This flux is the core process of life.

Ever since organisms started moving around, evolution favoured those with more complex systems to navigate their environments, because they could do this more effectively, which lead to a decisive advantage in surviving and thriving. At first, organisms coupled information about the outside and inside to their actions very tightly, automatically and directly. But slowly, adaptation led these systems to become more complex and enable more advanced decision making. Because again, that was beneficial to the survival of those organisms. Those who could learn about the world and decide better, outcompeted the simpler ones. Evolution settled on neural systems, because they are really good at aggregating and computing information. Neural networks can create new abstract representations. They do so by computation: summing up, averaging, and summarizing inputs from the lower levels into higher order patterns with meanings. These meanings are always relative to the survival of the organism in some way or another, because that's the only meaning that is given by evolution initially.

The idea of the self is an abstraction that arises once these abstract representations can be summed up over time. The self is the selective aggregation of the history of experiences of an organism. The sense of self binds us though: it constrains and limits our possibilities for action, and therefore we are not truly free. However we wouldn't want to be truly free, because that would mean we couldn't act for reasons. We would just do random things without any meaning. Every reason is a constraint. These constraints of a self, enable us to have free will. Because without selfhood, there wouldn't be a "us" that we could talk about in the first place.

In humans, there is something on top of all of that cognition: populations of neurons that can monitor higher level abstract representations and make us aware of them, linking the higher order representations into consciousness and therefore giving us the ability to directly manipulate our abstractions directly. This means we can think, reason, plan, imagine and so on because our internal world is recursive and this gives us metacognition. It also means that we can share our ideas with other people, we can talk about our own reasons and our mental representations. This is one of the Beginnings of Infinity that David Deutsch talks about.

To sum up the main idea of this book:

Once life does exist, everything changes. The universe doesn't have purpose, but life does. Natural selection ensures it.

The rest of it is really the story of how this unfolded. For more see the detailed notes below. They are quite in depth. The book itself is 100% worth reading for it's own sake. I first picked it up to compare it against another book I have recently read by Robert Sapolsky called Determined, which argues the opposite case, namely that we don't have free will at all, because well... neuroscience.

Detailed Notes

Preface

Scientists now have data to help answer the question of free will.

We understand the machinery of the brain which doesn't leave any room for "mental causation".

The implication of reductionism is that biology is not really a science unto itself but is just complicated chemistry, and chemistry is just complicated physics (and psychology is just complicated neuroscience).

Life is driven by information, not by energy.

Agency is a property of living systems.

The story of agency is the story of life itself.

Chapter 1 - Player One

When looked at closely, the machinery in our brain is something outside of our control, and in this sense we are programmed, much like a very complicated robot or NPC in a game. There is no choice for what we want because it's a product of our history, in which we had no choice.

Essay: Free Will - Sam Harris

Essay: Essay on the Freedom of the Will - Arthur Schopenhauer

Research by Libet. This comes up again and again. Same as in Determined by Robert Sapolsky.

Research by Wilder Penfield

Optogenetics - genetically engineered neurons that respond to different wavelengths of light by opening up ion channels in the membrane - thereby inducing/inhibiting action signals in neurons with light controlled by the researchers.

Using these tools we can control what the rats do, their behavior, but also their memories. Their thinking.

Research by Tomás Ryan

We may have set out, as neuroscientists, to explain how the workings of the brain generate or realize psychological phenomena, but we are in danger of explaining those phenomena away.

You cannot push the stocks in your brain around with a thought.

Book Recommendation: The Grand Design - Stephen Hawking

Philosophers argue about the meaning of every word in the question of "Do we have free will?"

What are "we", where is the boundary for "you" and "I"? What does "have" mean? And what do we mean by "free will" exactly?

Philosophers, by defining these terms, choose the free will they want to have, and then argue backward from it.

The book is more concerned with the question of: what kind of free will do we actually have!

The answer to this question comes from looking at evolutionary processes. How do very simple living things do stuff. Do they already have agency?

All living things have some degree of agency. That is their defining characteristic, what sets them apart from the mostly lifeless, passive universe.

From the chemistry of rocks and hydrothermal vents—the chemistry of the evolving planet itself—life emerged as systems of interacting molecules, interlocked in dynamic patterns that became self-sustaining.

Living things literally incorporate their history into their own physical structure to inform future action.

There is no self in a given moment: the self is defined by persistence over time.

The nervous system runs on meaning.

Chapter 2 - Life Goes On

Life is not a state it's a process. You are not just alive, you are living.

Book Recommendation: What is Life - Erwin Schrödinger

Life is a constant metabolic activity, the persistence of a pattern, that naturally would dissolve.

It originated near hydrothermal vents, using the H+ proton gradients in order to harvest energy, from there it got membranes, and later on a template for restoring it's chemical state (RNA and DNA).

DNA is an aperiodic crystal, it is stable, yet doesn't have a pattern, and can therefore store information. Because of its double strand nature it can also be easily copied.

When it's copied there can be mistakes, natural variations occur, they compete with one another, evolution is born.

DNA is a record of history. It records the experience of a cellular lineage. What worked and didn't work in the environment.

Genomes are models for the cell, but by extension also a model of the environment, because the cell is adapted to the environment.

Thought: This doesn't take into account epigenetics and the existence of xenobots and is therefore most likely an incomplete picture. Cell internal environment itself also stores information about outside environment and gets passed along in division.

Life exists because it can.

Once life does exist, everything changes. The universe doesn't have purpose, but life does. Natural selection ensures it.

Purpose is fittedness.

The existence of a goal imbues things with properties that previously never existed relative to that goal: function, meaning and value.

Life is literally a meaning making biochemical machine. It produces goal states and then, if it's still alive, strives to fulfill them. Persistence is the ultimate goal.

Chapter 3 - Action!

Organisms can act in the world, changing the environment that they are in.

Organisms have to be prepared for change because they cause change.

Life is building it's own environments. introducing higher level goal states, "simply" by playing upon itself. Whole biospheres like the Amazon rainforest or coral reefs are an emergent property of that.

Proteins can carry out logical functions by binding or not binding to certain places along translation machinery, in the presence of certain chemicals. They can sense molecules by binding to them, and then alter the state of metabolism by changing their shape and then interfering with or enabling certain processes because of their changed shape.

If A is there, do X else do Y

This enables cellular decision making. Attach a protein like this to a motor control, make the motor control function of the protein dependent on the concentration of a desired molecule, and you can have a cell, which moves into the direction of the chemical gradient of the desired molecule. It does so by 2 modes of movement, tumbling and swimming and resetting the ratio of the two based on how often the sensing molecules are getting triggered.

Usually the proteins doing the sensing of outside molecules sit inside the cell membrane, one end sticking out of and another into the cell. The shape of the protein changes when something binds on the outside, all the way to the inside part, leading to a signaling cascade within the cell.

Thought: The Clockwork video on Rhodopsin and G-Proteins comes into mind.

Sensing stuff on the outside is costly, hence evolutionary pressure limits the amounts of things organisms can sense to the slice of reality they need to know about in order to survive.

We inhabit a selected slice of reality.

The organism needs to integrate multiple measurements and compare them effectively to make a model of the outside world. Even at this simplest level of perception it is not just passively responding but is actively involved in making sense of that information to infer what is out in its environment—to build a kind of spatial map if how things are distributed. The organism doesn't really "know" this in any conscious sense, of course, because consciousness has not evolved yet.

Thought: This is slightly misguided. It's not of course that the organism isn't conscious yet at this point. I would say, according to Michael Levin, it has proto consciousness once it does this sort of integration of information and acting on that information. The difference between this and "real" consciousness is only one in degree of complexity, not in kind.

There is indeterminacy built into these molecular systems. They operate based on concentrations and random tumbling and spontaneous reactions that happen and sometimes don't. The idea of "same stimulus" makes no sense, because each time the cell will be in a wildly different state. There's chaos at the lowest levels, but statistical order imposed on top because of concentration differences. Overall this means that even at this single celled level life is not a "simple" stimulus- response machine.

Thought: The corollary of this is interesting: sufficiently advanced NPCs in a computer games, with statistical fluctuation and integration of data from their internal and external environment built into their "decision making" would be agentic in the same way as single celled organisms.

The causation in single celled organisms actions is not physical in origin it's informational. It's not about the physical stimulus that happens, but what it means for the cell.

The paradox of Shannon Information: Randomness carries the most information. But relative information is what is meaningful to us. We only care about certain patterns.

Before there were living organisms, there were no good things and bad things. Things are not good or bad in themselves: they only have meaning and value with respect to some goal or purpose.

Chapter 4 - Life Gets Complicated

Harvesting more energy allows for more complexity to arise. Endosymbiosis of mitochondria is comparable to the agricultural revolution.

This extra energy can be put to use for maintaining bigger genomes, because it releases the energy constraints on the maximum complexity that could be attained. This added potential for complexity can be exploited by evolution to produce cells with more bells and whistles. Even cells that could cooperate and achieve goals by working together. However when working together, dynamics of cheating force collectives of cells to be similar to one another genetically. The idea of multicellular life originates here. This allows for much bigger bodies, with specialized cell types and parts. But now these parts need to be coordinated over long distances. For this, signalling needs to be there. Biochemical electricity ion gradients can help in this signaling and that's exactly what neurons do.

Neurons are highly specialized to traverse these long distances with dendritic and axonal projections. They can be single cells with meter long appendages. Therefore they connect cells that are fat apart from one another and can integrate the information coming from those cells, including those of other neurons. This allows networks of neurons to do all kinds of information processing and therefore enhance the possibilities for really complex decision making greatly. Neurons and the nervous systems they form supercharge the idea of agency that proto electrochemical cellular networks have. They enable you reading this and me writing this.

More complex networks of neurons allow the coupling between sensing and action to loosen. There is more and more time for the decision, more and more wiggle room and freedom within the decision the bigger the neural networks get because the meanings and representations that they can create, increases with the added complexity.

Thought: This reminds me a lot of the ideas that David Deutsch expounds in his book the Beginning of Infinity.

Biological neural networks can learn facts about the environment that the animal is in, even in simple model organisms like the worm C.Elegans. The information is stored in the different strengths of connections between different neurons (as far as we know at least). Worms have associative, cellular memory.

This is a big idea, because learning allows for the integration of signals over time into decision making. Which again means, agency goes up. Learning gives the ability for organisms to make up their own reasons for deciding one thing over another. Their own reasons being informed by situations they encountered in the past. In a way, evolution does the same, but on a different time scale (multiple generations), whereas neural networks allow for updates within a single lifetime.

Chapter 5 - The Perceiving Self

Responding to stimuli is far from "real" cognition. It lacks representation. And the horizon of the representations matters, the more complex, the more the organism has to "think about" before making a decision.

Photoreceptors are called opsins, they have a small molecule called retinal inside them that is photosensitive and changes it's shape when hit by a photon. This shape change releases the molecule from the protein, which lead to the protein changing it's shape, triggering some other signal within the cell.

Photoreceptors are thus really chemoreceptors, but for a molecule that is itself sensitive to light.

Evolution has burdened living things with a glorious purpose: to persist, either in their own right or through reproduction. In the service of this overarching imperative, animals evolved sub goals: find food, avoid threats, and mate.

The visual cortex is made up of different visual areas, that feed signals into one another, realizing higher and higher level features, correlating signals across more and more neurons of the retina.

In doing so the visual cortex is using knowledge about the world, to see what is important for the organism. It interprets the data coming from the retina, imposing and guessing the meaning of it. Because the mapping of potential reasons for a certain activation patterns of retinal nerve cells is not 1-1 there are many options for the same pattern to arise. The visual cortex has to chose the most useful one. Visual illusions happen when that choice is wrong or particularly hard to make.

What we see is a belief that we map onto the sensory data, not the processed data itself.

Most of human sensory perception works this way, and brains can even correlate data across different senses. What we hear or smell can alter what we see.

We compare the data incoming from the retina and our top down expectation from past experience against one another, updating our expectations dynamically if they clash with the data. In this sense we try to predict what the environment is going to hold next, based on our past experience. This way we can make our internal world representations more accurate.

Perceptual systems are not just processing information—they are extracting meaning.

The possibility of movement creates a problem for the visual system. How to differentiate if new visual states happen because we moved or because the environment moved?

Chapter 6 - Choosing

When an organism has multiple attractive things in the world it would want to pursue, summing up or averaging all the actions, in other words going towards all of them, leads nowhere. It's like the donkey that died of thirst because it couldn't make up its mind where to drink. In reality, brains do winner takes all decisions around motor actions by inhibiting other areas in the brain that could also decide.

The hippocampus is encoding memories by linking particular random activation patterns to certain inputs. Essentially, reinforcing those random patterns to re-occur when the same stimulus is around again. It does this strengthening/weakening via Synaptic Plasticity and good old Hebbian Learning. What fires together, wires together. But we don't have a full understanding yet.

The meaning of the neural activity patterns at any time is grounded by the interpretation through the patterns of stored synaptic weights: incoming data acquire meaning by reference with stored knowledge.

The cellular processes of synaptic plasticity do not happen by default: they are gated by processes of attention, arousal, salience, reward, or surprise and signaled by neuromodulators like dopamine, serotonin, and acetylcholine.

We learn to perceive the world in a state of possibilities. The questions of what can and what should I do here drive our perception.

To chose which action to take an animal can pump it's own world model, put itself inside the situations where the goals are achieved and then use the valence of these predicted states to make a decision. They thus anticipate and simulate the future.

Animals develop, through experience, habits of thought.

Not everything possible that we could do at any moment "comes to mind", we only ever consider a vastly smaller subset of possible actions, compared to what we really could do at any point in time.

We inhibit actions by default, then different plans for action get simulated and compete for motor control and only the "best" ones gets released, and we act those and only those out.

Every time we chose an action, we learn if we get the desired outcome (or even something) better and then make that action more likely the next time. This is reinforcement learning.

Habits are shortcuts that help navigate familiar situations on autopilot without involving the cortical-striatum decision loop.

Animals are not waiting for external stimuli. They are cycling through possible actions, chosing, doing, re-evaluating, learning.

It is a recursive loop of mutual interaction.

Chapter 7 - The Future Is Not Written

Book Recommendation: A philosophical essay on probabilities - Simon Laplace

Bool Recommendation: Just Deserts - Daniel Dennett

A completely determined universe is a meaningless one. Because there wouldn't be choices that could matter.

Causal potential is accumulated across evolution through feedback processes that enable organisms to build information about their environment into the configuration if their own physical structures.

Book Recommendation: The Big Picture - Sean Carroll

Book Recommendation: Consciousness and the Laws of Physics - Sean Carroll

In a deterministic universe, things just happen, then other things happen, then other things happen, and so on. Nothing does anything—certainly not for a reason.

Book Recommendation: Fundamentals: Ten Keys to Reality - Frank Wilczek

Book Recommendation: Helgoland - Carlo Rovelli

Quantum physics might have true randomness at its core. Hidden variables are excluded by Bells Theorem and the multiverse theory has the problem of determining in which universe we end up in, which is again, random.

Quantum decoherence removes random fluctuations in realms where Newtonian mechanics acts out. Therefore the world that we think about and interact with can be treated as if it were deterministic, even if the quantum building blocks making it up aren't.

Feedback loops can lead to nonlinear dynamics and therefore unpredictable chaos as well as semi stable order. Life fits that bill.

The universe has finite precision for storing information and some of the precision of the information is only unfolding in time. Like computing PI, where PI itself is deterministic, yet you don't know the next state.

It would be better to follow the myth about the gods than to be a slave to the destiny of the physicists. – Epicurus

Emergent structures can causally control their lower level constituents.

Many scientists are suspicious of "why" questions. They smack of cosmic teleology.

Scientists instead pursue how questions. How does this mechanism work?

There are abstract principles of mathematics and engineering that just "exist" or hold true, independent of any physical substrate.

Natural selection and evolution are examples of such principles. Same would be true for Nash Equilibria or something like the digits of Pi.

A cause is a "difference that makes a difference."

Chapter 8 - Harnessing Indeterminacy

The world is undetermined because of Quantum Effects and Non-linear Dynamics+ Chaotic Systems.

Protein binding and conformation changes are probabilistic. A problem for neurons and their neurotransmitter vesicles and signal strength/reliability.

The noisiness in neurons could be by design. Nature varies and so the same input could lead to slightly different outputs each time, meaning that the behavior of the organism is more varied and hence maybe more adaptive. It doesn't try the same strategy over and over again. It also enables learning, neurons re-adjust their baseline all the time. Stronger signals over time get tuned down by regulating receptors etc.

Organisms can sometimes choose to do something random.

Cockroaches can run away in random directions and we don't really know how they do that precisely.

Neural circuits can accumulate activations over time because their activity is dynamic. They are on and off, but the action is not spontaneous, there is a time component to it. This way fluctuations can be harnessed. Suppressed as well as amplified and used to tip the dynamic state one way or another.

Libets experiments are looking at decision data in an averaged out way which leads to a potentially wrong conclusion. There is more research, that disagrees with Libets argument by Aaron Schurger. The data can be explained by random noise being used to generate a random decision. Moving the finger "on a whim". The effects disappear when people care about the choice and it has consequences. There's more research for this by Uri Maoz and Liad Mudrik.

Overall then, Libet's experiments have very little relevance for the question of free will. They do not relate to deliberative decisions at all, where the readiness potential is not observed. Instead, they confirm first, that neural activity in the brain is not completely deterministic, and second, that organisms can choose to harness the inherent randomness to make arbitrary decisions in a timely fashion.

Book Recommendation: The Origin of Species - Charles Darwin

Songbirds learn songs by producing a motor pattern, then randomly adding noise to it, then comparing it to the result and keeping variations that are closer, over time their song starts matching the original.

Chapter 9 - Meaning

Book Recommendation: A Metaphysics for Freedom - Helen Steward

Book Recommendation: The Astonishing Hypothesis - Francis Crick

The neural patterns only have causal power in the system by virtue of what they mean.

Neurons course grain information signals in time, they don't transmit every pulse rather they aggregate pulses and sum them up to reach the firing threshold. This way they can generate new information by relating the firing rates of independent other neurons. Higher order information about the coinciding (or not coinciding) of firing of other neurons.

In the case of color or visual spatial input this is very clear. Neurons create new information by the computation that they run on their input neurons.

Color doesn't even exist in the world: it is the result of the organism creating what are essentially arbitrary categories to help distinguish objects from each other.

Information is present in earlier layers of neurons, but it is entangled with other information so as to become essentially useless. Only through more layers of neurons and their filtering is the information processed and made visible.

The Hallmark of understanding: the ability to generalize to new situations by reference to an organized, hierarchical, categorical framework of regular causal relations in the world.

Objects and concepts are represented by neuron clusters in the brain as maps. We are napping the world within our brains.

There appears to be an anatomical progression from areas representing percepts to ones representing concepts.

First we perceive, then we map, then we progressively conceptualize.

These maps and representations are surprisingly well conserved, between different humans and some even between monkeys and humans.

Representations are not mental. They are happening at a structural level of neuronal organization. They could be or correspond mental representations too but right now we don't know.

Low level details of how meaning or representations are instantiated are not important. Different arrangements of neurons could represent the same thing. It doesn't even have to be neurons.

The brain doesn't run on spikes, but on patterns.

Although it is necessarily instantiated in a set of physical mechanisms, the causation in the system is informational. The criteria for how each neuron or each population responds to incoming information are set over multiple time-frames: over millennia by evolution; over a lifetime by individual experience; over years, months days or hours by the adoption of different goals; and over minutes seconds, or even tens of milliseconds by processes of attention, arousal, and moment-to-moment decision making.

Abstract entities like thoughts and beliefs and desires can have causal inluence in a physical system.

The meaning of the various (often arbitrary) neural patterns arrises through the grounded interaction of the organism with its environment over time.

Organisms are separate from the environment because they push back on it. Transforming a pocket of environment into themselves and then keeping it that way, even against outside influences trying to undo the particular organization.

Book Recommendation: The Critique of Judgement - Immanuel Kant

Book Recommendation: Opus Postumum - Immanuel Kant

Organisms are made out of processes, that relate to one another, both ways.

Once you break apart the relations, there is no system left to understand.

Agency—the capacity of organisms to act with causal power in the world, for their own reasons—is the defining feature of life itself.

Chapter 10 - Becoming Ourselves

Book Recommendation: On the Freedom of the Will - Arthur Schopenhauer

If we are a product of our past environment, how can we make free decisions? Aren't we constrained by our own histories in the things that we wish for and want?

Genetic heritability is a complicated process, because it doesn't imply determinism, especially with behavior. Genetics influence psychology which influences behavior. But the environment does the same, and so even though some behavioral traits are heritable, that is more a testament to the fact that similar genetics builds similar brains. Similar, but not equal.

Human DNA programs human nature.

However we can alter and build on that nature by our own experience.

We have our unique version of the pattern because our genetics and upbringing was different from everybody else in the world.

Genetics means we are not blank slates.

Genes do shape how we become ourselves. Even if they don't determine it outright.

What do the Big Five traits represent biologically? What does it mean on a neuron / pattern level for somebody to be more neurotic or more open?

Researchers haven't found neurological correlates yet...

Extraversion is not a singular "thing in the brain." Rather, it is a construct reflecting multiple neural parameters.

We can make decisions that are constricting future decisions. Long term goals and plans, that shape the space of potential actions that we would consider as viable.

Character is about virtues and vices, it's a moral judgement, pro social vs. anti social aspects of being. Personality isn't normative like that.

Humans survived and thrived and colonized practically every niche on the planet by working together.

A sense of fairness develops from evolution to protect against abuse by freeloaders.

Book Recommendation: Social Cognitive Theory - Albert Bandura

We signal to other people if their actions are virtuous or not, in an attempt to change them, making them conform to the pro-social norms. Character is shaped by the environment.

The external signals from other people's reactions get internalized to form part of a self-regulatorx apparatus—a moral compass—that we use to guide our own behavior.

The central question of Buddhism, once the self is lost, why do anything, the dilemma of the awakened person, is arising within the book.

Even if your conscious self were somehow able to choose what to want to do, free from any constraints arising from prior caused or subconscious influences, then on what basis would you decide?

Conscious or not, reasons (or preferences) are constraints—that's how they guide action.

There is no self in a moment. The self is defined by continuity through time.

The organism is not a pattern of stuff; it is a pattern of interacting processes, and the self is that pattern persisting.

I would add that it is also that pattern changing and evolving. Adapting to new experiences, learning.

Completely free of any constraints, you "could do whatever, but not whatever you want". Having a you, that wants things is a constraints, a pretty limiting one in fact.

Chapter 11 - Thinking About Thinking

We might just tell ourselves stories of why we decided something, after the fact. Rationalizing and narrating our actions, reasoning about them after the fact they have been made automatically by the subconscious brain.

Book Recommendation: Thinking Fast and Slow - Daniel Kahneman

Many of the priming literature is fake or badly conducted experiments. Even Daniel Kahneman admitted this.

We talk about our goals, intentions and planned actions with other people, if they were not predictive or descriptive of actual behavior, this wouldn't make any sense. It would be selected against.

Different organisms have different cognitive depth. Humans have super deep cognitive depth because of meta cognition. We can recursively think about thinking. We are generally speaking because of this recursion a Beginning of Infinity.

We combine simple features into more distinct categories. Things that have feathers + things that lay eggs + things that can fly => well that's kinda a "bird like" thing. We have a category for it.

We can combine these nested hierarchies if concepts and maps of causal relations and system dynamics in new, creative ways within the abstract cognitive space and thereby engage in open-ended, model-based reasoning. We can imagine things.

We have neurons that are monitoring the confidence of these hierarchical concepts. And we have yet more neurons, that monitor those neurons. In other words, we are consciously aware of our confidence in our beliefs.

With a first order model, animals can realize they made the wrong choice and should do something else next time. With our higher order models, humans can realize that they made that choice for the wrong reasons and should think something else next time.

Book Recommendation: Synthese - Douglas Hofstaedter

Our prefrontal cortex can hold multiple concepts in working memory and do cognitive work on them, and then use the result to bias the actions in the rest of our cognitive decision making machinery.

Conscious awareness only presents select information in order to aid decision making. You don't want all the information all the time, just the right information at the right time.

We can learn what to pay attention to. The filtering of what makes it into consciousness can be altered.

As we evolved into the ultra-social, ultra-cooperative, obligately cultural species we are, other minds became, by far, the most important things in our environments.

It's not a question whether immaterial thoughts can push around physical stuff. Thoughts are not immaterial: they are physically instantiated in patterns of neural activity in various parts of the brain.

There's no need to posit a "ghost in the machine"—you're not haunting your own brain. The "ghost" is the machine at work.

Thoughts are not just patterns of neural activity. They have meaning. And this meaning gives them causal power. Their exact pattern doesn't matter, as long as the meaning stays the same. Thoughts are software.

Self control can vary between individuals and higher self control is seen as better character in most societies.

Chapter 12 - Free Will

The story of agency is really the story of life itself.

Living things strive, actively, to keep themselves organized. They take in free energy and perform work to stay out of thermodynamic equilibrium with their surroundings.

Through both biological innovations and cumulative cultural evolution, humans developed capacities for creative, open-ended, recursive thoughts and boundless imagination that truly set our minds free to combine and manipulate ideas in more and more abstract ways.

Selfhood this entails constraint. It is only constraint. The freedom to be you involves constraining the elements that make you up from becoming not you.

The low-level details of physical systems plus the equations governing the evolution of quantum fields do not completely determine the evolution of the whole system. They are not causally comprehensive.

Our decision making changes when we change. When we are tired, or sad, or angry, how we decide changes.

Our free will is thus not some nebulous, spooky, mystical property granted to us by the gods. It is an evolved biological function that depend on the proper functioning of a distributed set of neural resources.

Reducing people to their genetics and their brains, is not correct. And we shouldn't do it in the legal system.

Brains do not commit crimes: people do.

Genetic and neural variation play a role in decision making, but that role is probabilistic. Except in pathological cases, like tumors.

Higher level systems have causal power because physics is not enough to completely determine them. Therefore emergence can happen and the reasons embodied in the structure can fix the physics into certain states, especially in living things.

Living beings do not cause themselves in an instant, but they do cause themselves through time. That's what being alive entails—continuing to cause yourself.

This time aspect is important because it removes a circular argument.

Meaning is contextual. The meaning of the words on a page in the book is only existing between the relationships of minds, language, arrangements of paper and ink, information stored and decoded in a specific way, affecting a mind, which because of its context and history can extract the meaning from the visual stimulus.

Reductionism is effective in teasing apart the workings of a system. But the system is not only it's parts. Many systems have emergent, holistic effects that a reductive approach will miss.

Even Consciousness itself may have evolved in the service of better cognitive control and self regulation.

And we don't do this alone. The true power of human thought comes through collective interaction and cumulative culture. We share and accumulate knowledge and deeper understanding over generations, with young people easily grasping concepts that were literally unthinkable just decades earlier.

We have, as individuals and as a species, the power to transcend the immediacies of our own biology.

Epilogue - Artificial Agents

General intelligence can not be gained by passively consuming data. You need to perturbe the system to test hypotheses to test understanding and learn. AGI needs to be embodied somewhere, where it can act to learn, that somewhere can be a simulation.

Aritificial general intelligence may have to be earned through the exercise of agency.

Evolution has given us a roadmap of how to get to intelligence: by building systems that can learn but in a way that is grounded in real experience and causal autonomy from the get-go. You can't build a bunch of algorithms and expect an entity to pop into existence. Instead, you may have to build something with the architecture of an entity and let the algorithms emerge.