Evolution of the Brain

Note: to be reduced from rough notes

600m ya - Bilaterians and Steering

Valence

Bilaterians are the only animals that have brains
Nematodes (legless wormlike creatures about the size of a grain of rice) emerge in the Edicaran period from 635 to 539m ya.
Brain had 302 neurons (against 85 billion today
Initial steering is obtained through assessing the valence (goodness or badness) of a stimulus, and going towards the things that smelled good and away from the things that smelled bad.
There were negative and positive valence sensing neurons and move forward neurons and turning neurons.
The various sensory inputs acted as votes for going one way or another and the first brains evolved as a mega-integration place to take in all these votes and then decide who had won and thus where to steer.

Emotions

Affect is the name for the unifying foundation of emotions
In addition to valance (good or bad), there is arousal (high or low)
A primitive good mood encourages feeding, digesting, and sexual activity
A primitive bad mood inhibits these activities
An aroused good mood leads to exploiting nearby food sources or sexual partners
An aroused bad mood leads to escaping from bad feelings - hunger, fear
The brain generates affective states using neuromodulators like dopanmine and serotonin.
In the nematode, dopamine is released to create arousal and drive the search for food and serotonin is released to suppress arousal and drive the enjoyment of digesting it.
Dopamine is less about liking things and more about wanting them.
Other neuromodulators - norepinephrine, octopamine, and epinephrine drive escape behavior by suppressing the effectiveness of serotonin and stopping an animal from being able to rest and feel safe - acute stress response.
Opioids initiate recovery processes and inhibit negative valence neurons to stop and recover from stress episodes.
Chronic stress turns off arousal and motivation, activates serotonin and leads to numbness and depression (anhedonia). It can cause learned helplessness
Affect answers two questions:
- Do I want to expend energy by moving?
- Do I want to stay here or leave?

Associating, Predicting, Learning

The digestive organs are under the control of the nervous system
Conditional reactions are involuntary associations - associative learning happens automatically without conscious involvement.
At the same time as valence, the ability to use experience to change what is considered good and bad also emerges.
Learning in biological brains has always been continual.
Pavlov’s conditional reflexes are always strengthening (acquisition) or weakening (extinction) with each new experience. Extinction may be followed by spontaneous recovery (instantaneous) or reacquisition (faster than first time)
Spontaneous recovery is a primitive form of long-term memory.
The credit assignment problem - which cue really predicted something subsequently happening?:
- Eligibility traces - Immediately follows cue
- Overshadowing - Pick strongest cue
- Latent inhibition - Pick the cue you haven’t seen before.
- Blocking - Use existing cues and ignore others.
Learning occurs when synapses change strength or when new ones are formed or old ones are removed.
Association, prediction, and learning emerged to tweak the goodness and badness of things

500m ya - Vertebrates and Reinforcing

Reinforcement Learning

Cambrian period (Cambrian Explosion) is 540-485m ya.
The brains of all vertebrates, from fish to humans, develop in the same initial steps:
- Brains differentiate into three bulbs - a forebrain, midbrain, and hindbrain
- The forebrain unfolds into two subsystems:
  - The cortex and the basal ganglia
  - The thalamus and the hypothalamus
Animals learn by first performing random exploratory actions and then adjusting future actions based on valence outcomes.
Reinforcement learning is the ability to learn arbitrary sequences of actions through trial and error with reinforcing and punishing depending on the valence outcomes.

Temporal Distance Learning

Most drugs of abuse - alcohol, cocaine, nicotine - work by triggering the release of dopamine. All vertebrates, from fish to rats to monkeys to humans, are susceptible to becoming addicted to such dopamine-enhancing chemicals.
Discounting drives AI systems (or animals) to choose actions that lead to rewards sooner rather than later.
Dopamine is not a signal for reward but for reinforcement. Reinforcement and reward must be decoupled for reinforcement learning to work. To solve the temporal credit assignment problem, brains must reinforce behaviors based on changes in predicated future rewards, not actual rewards. This is why animals get addicted to dopamine-releasing behaviors despite it not being pleasurable, and this is why dopamine responses quickly shift their activations to the moments when animals predict upcoming reward and away from rewards themselves.
Dopamine was originally a signal for good things nearby - a primitive version of wanting. Evolution reshaped it into a temporal difference learning signal, from a fuzzy average of recently detected food to an ever fluctuating, precisely measured, and meticulously computed predicted-future-reward signal.
Disappointment and relief are emergent properties of a brain designed to learn by predicting future rewards.
The omission of an expected punishment is itself reinforcing; it is relieving. And the omission of an expected reward is itself punishing; it is disappointing.
Vertebrates are unique in the precision with which they can measure time.
Temporal distance learning, disappointment, relief, and the perception of time are all related.
The basal ganglia is in a perpetual state of gating and ungating specific actions, operating as a global puppeteer of an animal's behavior.
- It learns to repeat actions that maximize dopamine release.
- It is a system designed to repeat behaviors that lead to reinforcement and inhibit behaviors that lead to punishment.
The hypothalamus houses valence neurons inherited from the valence sensory apparatus of ancestral bilaterians.
- It is, in principle, a more sophisticated version of the steering brain of early bilaterians; it reduces external stimuli to good and bad and triggers reflexive responses to each.
- When the hypothalamus is happy, it floods the basal ganglia with dopamine, and when it is upset, it deprives the basal ganglia of dopamine.
- The basal ganglia is a student, always trying to satisfy its vague but stern hypothalamic judge.
- The hypothalamus is the decider of actual rewards.
How is dopamine transformed from a valence signal for actual rewards to a temporal difference signal for changes in predicted future reward? The basal ganglian student initially learns solely from the hypothalamic judge, but over time learns to judge itself, knowing when it makes a mistake before the hypothalamus gives any feedback.
This is why dopamine neurons initially respond when rewards are delivered, but over time shift their activation toward predictive cues.
This is also why receiving a reward that you knew you were going to receive doesn't trigger dopamine release; predictions from the basal ganglia cancel out the excitement from the hypothalamus.

Pattern Recognition

Sometime around 500m ya, our ancestor evolve pattern recognition to remember the smell of that dangerous arthropod, to remember the sight of its eyes peeking through the sand.
Early vertebrates could recognize things using brain structures that decoded patterns of neurons. Within a small mosaic of only fifty typos of olfactory neurons lived a universe of different patterns that could be recognized. Fifty cells can represent over one hundred trillion patterns.
Patterns can be similar but not the same.
Your iPhone needs to be able to tell the difference between your face and other people's faces, despite the fact that faces have overlapping features (discrimination). It must also be able to identify your face despite changes in shading, angle, facial hair, and more (generalization).
In the first cortex evolved a new morphology of neuron:
- Pyramidal neurons have hundreds of dendrites and receive inputs across thousands of synapses.
- These were the first neurons designed for the purpose of recognizing patterns.
- A small number of olfactory neurons connect to a much larger number of cortical neurons. They connect sparsely - a given olfactory cell will connect to only a subset of these cortical cells. This leads to pattern separation, decorrelation, or orthogonalization.
The problem of "catastrophic forgetting is why we don't let AI systems learn things sequentially; they learn things all at once and then stop learning. But even early bilaterians learned continually.
The retina contains over 100m neurons of five different types. The visual cortex decodes and memorizes the visual pattern the same way the olfactory cortex decodes and memorizes smell patterns.
But the same visual object can activate different patterns depending on its rotation distance, or location in your visual field. This creates the invariance problem - how to recognize a pattern as the same despite large variance in its inputs.
The same issue arises with words spoken by a child and an adult or in different accents. Your brain is somehow recognizing a common pattern despite huge variances in the sensory input.
Visual (and audio) processing in mammals is hierarchical:
- The lateral geniculate nucleus (LGN) is a small, ovoid, ventral projection of the thalamus where the thalamus connects with the optic nerve.
- The V1 area decomposes the complex pattens of visual input into simpler features like lines and edges
- V1 sends its output to V2, which then sends information to an area called V4, both of which are sensitive to more complex shapes and objects
- V4 sends its output to the inferior temporal gyrus or IT cortex, which is sensitive to complex whole objects like specific faces.
In the predatory arms race of the Cambrian, evolution shifted from arming animals with new sensory neurons for detecting specific things to general mechanisms for recognizing anything, and this caused new sensory organs and each incremental improvement in the brain's pattern recognition expanded the benefits to be gained by having more detailed sensory organs:
- Noses evolved to detect chemicals
- Inner ears evolved to detect frequencies of sound
- Eyes evolved to detect sights
In the brain, the result was the vertebrate cortex, which somehow recognizes patterns without supervision, accurately discriminates overlapping patterns and generalizes patterns to new experiences. It somehow continually learns patterns without catastrophic forgetting and despite larges variances in its inputs.
Pattern recognition and reinforcement learning evolved simultaneously in evolution. The greater the brain's ability to kearn arbitrary actions in response to things in the world, the greater the benefit to be gained from recognizing more things in the world. The more unique objects and places a brain can recognize, the more unique actions it can learn to take.
And so the cortex, basal ganglia, and sensory organs evolved together, all emerging from the same machinations of reinforcement learning.

Curiosity

It was early vertebrates that first became curious.
In vertebrates, surprise itself triggers the release of dopamine, even if there is no "real" reward.
To make animals curious, we evolved to find surprising and novel things reinforcing, which drives us to pursue and explore them. Even if the reward of an activity is negative, if it is novel, we might pursue it anyway.
Games of gambling are designed to exploit this with a 48% chance of winning it is high enough to be possible, but uncertain enough to make it surprising.
Social networks also hack into our 500m year preference for surprise, by showing us surprising things, but only sometimes.
Curiosity is a requirement for reinforcement learning to work. For the first time, learning became, in and of itself, an extremely valuable activity.

Modeling the World

Your brain has built a spatial map of your home so that you can make your way around (with a few stubbed toes) in the dark.
All vertebrates can learn spatial maps, but simple bilaterians cannot.
The vestibular sense feels the direction of head movement through "head-direction neurons".
The cortex of early vertebrates had three subareas:
- Lateral cortex - Recognizes smells and will evolve into the olfactory cortex in early mammals.
- Ventral cortex - Recognizes patterns of sights and sounds and will evolve into the amygdala.
- Medial cortex - Visual, vestibular, and head direction signals propagate here, where they are mixed together and converted into a spatial map. Later became the hippocampus. Contains place cells that activate when an animal is in a specific location
This was the first time that an organism could recognize where it was.
The first time a brain differentiated the self from the world.
The first tiem that a brain constructed an internal model - a representation of the external world.

200m ya - Mammals and Simulation

The Devonian and Permian Eras

420m to 375m ya is called the Devonian period - arthropods walked out of the oceans to populate the land, plants first evolved leaves for better absorption of sunlight and seeds for spreading, and trees first developed.
The Late Devonian Extinction caused the Carbon dioxide levels to plummet and the climate to cool, freezing the oceans.
Reptiles and therapsids evolved, with the therapsids (our ancestors) developing warm-bloodedness - the ability to use energy to generate their own internal heat. During the Permian era (300-250m ya) they became the most successful land animals
During the Permian-Triassic mass extinction event, 250m ya, over the course of 5-10m years, 96% of all marine life and 70 of land life died.
The reptiles became dominant while the bigger therapsids died out and only small therapsids, like the cynodont survived.
These burrowing or arboreal four-inch mammals, like birds or squirrels, had one advantage, they could make the first move.
The neocortex evolved to give these mammals the ability to simulate actions before they occurred.
Early vertebrates learned by doing, while these mammals could learn before doing, by imagining.
There were two requirements needed for simulation to evolve:
- Far-ranging vision - To see much of your surrounding and simulate various paths through them
- Warm-bloodedness - To let mammal brains operate much faster than fish or reptile brains.
The ventral cortex of the vertebrates became the associative amygdala in mammals - learning to recognize patterns that were predictive of valence outcomes

Generative Models, Perception, and the Neocortex

The neocortex is a sheet about two to four millimeters thick, folded around the outside of the brain. Unfolded, it is about three square feet
There are regions for vision, audition, touch, pain, and taste, for movement, language, and music.
In the middle of the 20th century, Vernon Mountcastle discovered that the neocortex is made up of repeating and duplicated microcircuits that he called neocortical columns:
- Neurons within a vertical column about five hundred microns in diameter of the neocortical sheet seem to all respond similarly to sensory stimuli, while neurons horizontally farther away do not. Eg an individual column in the visual cortex might contain neurons that all similarly responded to bars of light at specific orientations at a specific location in the visual field. However, neurons within nearby columns responded only to bars of light at different orientations or locations. And this separation is found across multiple other modalities.
- There are many connections vertically within a column and comparatively fewer connections between columns.
- The neocortex looks largely identical everywhere. All the different ares contain the same types of neurons organized in the same way in all specied of mammals.
Each neocortical column does exactly the same thing - the only difference is the type of input they receive and where they send their output. They seem to all implement some algorithm that is so general and universal that it can be applied to extremely diverse functions such as movement, language, and perception across every sensory modality
The neocortex contains six layers of neurons connected in a complicated buy consistent way:
- In layer six, there are neurons that always project to the thalamus
- In layer five, a specific type of neuron always projects to the basal ganglia, the thalamus, and the motor areas.
- In layer four, there are neurons that always get input directly from the thalamus
The properties of perception help us understand how the neocortex works:
- Filling in - the mind automatically and unconsciously fills in missing things.
- One at a time - the mind can see only one interpretation at a time
- Can't unsee - once it perceives an interpretation, the mind cannot unsee it. It likes to have an interpretation that explains sensory input
Hermann von Helmholtz proposed that a person doesn't perceive what is experience; instead, he perceives what the brain thinks is there, through a process called inference.
Brains must somehow recognize aspects of the world without being told the right answer. A suggestion for understanding this arises from Hinton's generative models, which have two modes:
- In recognition mode information flows up the network
- In generative mode, information flows backwards down the network.
- The network learns to recognize without supervision by generating its own data.
Hallucinations, dreams, and imagination suggest that the brain is like a generative model:
- Cutting off sensory input to the neocortex makes it unstable. It gets stuck in a drifting generative process in which visual scenes are simulated without being constrained to actual sensory input
- Some neuroscientists refer to perception, even when it is functioning properly, as a "constrained hallucination".
- Only mammals and birds show unequivocal evidence of dreaming.
- The neocortex is always in an unstable balance between recognition and generation, and during our waking life, humans spend an unbalanced amount of time recognizing and comparatively less time generating. Perhaps dreams are a counterbalance to this, a way to stabilize the generative model through a process of forced generation. If we are deprived of sleep, the imbalance becomes so severe that the generative model in the neocortex becomes unstable and we start to hallucinate.
- It is natural for humans to imagine things that they are not currently experiencing and when you are imagining, this could simply be the neocortex in generation mode
- When people are imagining things, their pupils dilate as their brains stop processing actual visual data
- The same neocortical neurons activate during recognition and when you simply imagine the same thing. People with neocortical damage that impairs certain sensory data become equally impaired at simply imagining features of that same sensory data. Perception and imagination seem not to be separate systems but two sides of the same coin.
The generative model in the neocortex seems to render a simulation of your environment so that it can predict things before they happen.
Early mammals learned to predict everything, to monitor everything, and to experience surprise when things did not go as expected.
If reflex circuits are reflex-prediction machines, and the critic in the basal ganglia is a reward prediction machine, then the neocortex is a world-prediction machine.
The neocortex may be prewired to assume that incoming sensor data, whether visual, auditory, or somatosensory, represent 3D objects that exist separately from ourselves and can move on their own. It does not have to learn about space, time, and the difference between the self and others. Instead, it tries to explain all incoming sensory information it receives by assuming it must have been derived from a 3D world that unfolds over time.
It is when the simulation in your neocortex becomes decoupled from the real external world around you - when it imagines things that are not there - that its power becomes most evident.

Imagination

The neocortex brought three new abilities:
- Vicarious trial and error - Rats can play out each option in a maze before trying it. They do this only when decision are hard, when the costs are close to the benefits, or when the rules change. When the rat stopped at the decision point and turned its head back and forth, its hippocampus ceased to encode the actual location of the rat and instead went back and forth rapidly playing out the sequence of place codes that made up both possible future paths from the choice point. Once a rat has a world model of their environment, they can rapidly mentally explore it until they find a way to get around obstacles to get what they want.
- Counterfactual learnings - An experiment presenting rats with a continuous set of irreversible choices saw that the rats that turned back and reactivated the representation of a forgone choice also ended up changing their future choices. Counterfacgual leaning was a major enhancement in how ancestral brains solved the credit assignment problem. Causation may live more in psychology than in physics. It is constructed by our brains to enable us to learn vicariously from alternative past choices.
- Episodic memory - We don't truly remember episodic events. The process of episodic remembering is one of simulating an approximate re-creation of the past. When imagining future events, you are simulating a future reality; When remembering past events, you are simulating a past reality. Both are simulations. People's episodic memories are filled in during the remembering process.
  - The distinction between a made-up imagined scene and an actual episodic memory is thin in the neocortex - repeatedly imagining a past even that did not occur falsely increases a person's confidence that the event did occur.
  - In mammal brains, episodic memory emerges from a partnership between the older hippocampus and the newer neocortex. The hippocampus can quickly learn patterns, but cannot render a simulation of the world; the neocortex can simulate detailed aspects of the world, but cannot learn new patterns quickly. Episodic memory must be stored quickly, and thus the hippocampus, designed for the rapid pattern recognition of places, was repurposed to also aid in the rapid encoding of episodic memories. Distributed neural activations of sensory neocortex (ie simulations) can be retrieved by reactivating the corresponding pattern in the hippocampus. By retrieving and replaying recent memories alongside old memories, the hippocampus aids the neocortex in incorporating new memories without disrupting old ones.

Model-Based Reinforcement Learning

Unlike TD-Gammon, AlphaZero was a model-based reinforcement learning algorithm. AlphaZero searched through possible future moves before deciding what to do next. It didn't simulate the trillions of possible futures; it prioritized and simulated only a thousand.
The brilliance of simulation in mammal brains is likely to be the flexibility with which they employ different strategies and the intelligence with which they decide between them:
- Sometimes we pause to simulate our option, but sometimes we just act instinctively.
- Sometimes we pause to consider possible futures, but other times we pause to simulate some past event or alternative past choices.
- Sometimes we imagine rich details in our plans, playing out each individual detailed subtask, and sometimes we render just the general idea of the plan.
A patient affected by a stroke in her prefrontal neocortex had lost all intention.
In the back half of the neocortex is the sensory neocortex, where a simulation of the external world is rendered, but the frontal neocortex contains three main subregions:
- Motor cortex
- Granular prefrontal cortex (gPFC)
- Agranular cortex (aPFC)
The frontal neocortex decides when and what to imagine.
The primary input to the aPFC comes from the hippocampus, hypothalamus, and amygdala. The aPFC learns to model the animal itself, inferring the intent of behavior it observes, and uses this intent to predict what the animal will do next. The aPFC creates a model of an animal's goals.
The columns of the aPFC might always be in one of three states:
- Silent (no intent)
- Recognize an intent and all predict the same next behavior
- Recognize and intent but predict different and inconsistent behaviors - the aPFC gets most excited when something goes wrong or something unexpected happens.
The process might go something like:
- The degree of disagreement of predictions is a measure of uncertainty. It might be this that triggers simulations. If everything is going as expected, you can let the basal ganglia drive decisions without a model, but when uncertainty emerges (something nex, some contingency is broken, or costs are close to benefits), then simulation is triggered.
- The aPFC specifically explores the paths that it is already predicting an animal will take. It could be triggering the sensory neocortex to render a specific simulation of the world, or it could be that the basal ganglia determines the actions taken during these simulations.
- The basal ganglia accumulates votes for competing choices, with different populations of neurons representing each competing action ramping up excitement until it passes a choice threshold, at which point an action is selected. The aPFC is vicariously training the basal ganglia, which doesn't know whether the sensory neocortex is simulating the current world or an imaginary world, all it knows is that it is getting reinforced when it makes its choice.
Habits are automated actions triggered by stimuli directly (they are model-free), and are controlled directly by the basal ganglia to save time and energy by avoiding unnecessary simulation and planning. Brains attempt to intelligently select when to model and when to rely on habit, but sometimes they make mistakes, and this is the origin of many of our irrational behaviors.
Just as you don't perceive what you see, so intent is not real - it is a computational trick for making predictions about what an animal will do next.
The basal ganglia has no intent or goal - it simply learns to repeat behaviors that have previously been reinforced.
The aPFC, however, does have explicit goals. By simulating a future that terminates at some end result, it has an end state (a goal) that it seeks to achieve.
The sensory cortex engages in passive inference - merely explaining and predicting sensory input. The aPFC engages in active inference - explaining one's own behavior and then using its predictions to actively change that behavior. It is repurposing the neocortical generative model for prediction to create volition.
The basal ganglia begins as the teacher of the aPFC, but as a mammal develops, these roles flip, and the aPFC becomes the teacher of the basal ganglia. Perhaps this is part of a developmental program for constructing a model of self, starting by matching one's internal model to its observations, and then transitioning to pushing behavior to match one's internal model.
Attention, working memory, executive control, and planning are all different applications of controlling the neocortical simulation.
- The aPFC's triggering of simulation is called imagination when it is unconstrained by current sensory input and attention when it is so constrained.
- Controlling ongoing behavior often also requires working memory - the maintenance of representations in the absence of any sensory cues while waiting. Working memory is just your aPFC trying to keep re-invoking an inner simulation until you no longer need it.
- The aPFC can inhibit the amygdala. This is the evolutionary beginning of behavioral inhibition, willpower, and self-control. In moments of willpower, you can inhibit your amygdala cravings, while in moments of weakness, the amygdala wins. But the aPFC is expensive to run and so when you are tired or stressed, it is less effective.
The aPFC controls behavior not by showing, but by telling.

The Hierarchy of Goals, and Learning and Automation

The motor cortex is a thin band of neocortex on the edge of the frontal cortex. It makes up a map of the entire body, dedicating lots of space to the parts of the body that animals have skilled motor control over (like the mouth and hands) and less to areas that they control less finely (like feet)
This map is mirrored in the adjacent somatosensory cortex, the region of the neocortex that processes information coming from touch sensors in the skin and proprioceptive signals from muscles.
The motor cortex is the primary system for controlling movement. It emerged tens of millions of years after the first mammals and only in the placental lineage.
Perhaps it doesn't generate motor commands, but rather predictions. Perhaps it is in a constant state of observing body movements that occur in the nearby somatosensory cortex and then tries to explain the behavior and use these explanations to predict what the animal will do next. The motor cortex is wired to make its predictions come true.
The aPFC learns to predict movements of navigational paths while the motor cortex learns to predict movements of specific body parts.
This is "embodiment" - parts of the neocortex have an entire model of an animal's body that can be simulated, manipulated, and adjusted as time unfolds.
The motor cortex was perhaps originally for motor planning. When learning a new movement the motor cortex simulations vicariously train the basal ganglia. Once a movement is well learned, the motor cortex is no longer needed.
There is plenty of evidence that the premotor and motor cortices are activated both by doing movements and by imagining movements.
Mental rehearsal of motor skills substantially increases performance across speaking, golf swings, and even surgical maneuvers.
- At the top of the hierarchy is the aPFC, where high-level goals are constructed based on amygdala and hypothalamus activation. It then propagates these goald to a nearby frontal region,
- The premotor cortex, which constructs subgoals and propagates these further until they reach
- The motor cortex, which then constructs subgoals
- The basal ganglia makes loops of connectivity with the frontal cortex, with the aPFC connecting to the front region of the basal ganglia (which then connects back to aPFC through the thalamus), and the motor cortex connecting to the back region of the basal ganglia (which then connects back to the motor cortex through a different region of the thalamus)
So any level of goal, whether high-level or low-level, has both a self model in the frontal neocortex and a model-free system in the basal ganglia. The neocortex has a slower but more flexible system for training, and the basal ganglia offers a faster but less flexible version for well-trained paths and movements:
- The front part of the basal ganglia associates stimuli with high-level goals. It generates cravings.
- The aPFC, however, is what makes you pause and consider if you actually want to pursue these cravings
- The back part of the basal ganglia associates stimuli with low-level goals. It generates automatic skilled movements
- The motor cortex, however, makes you pause and plan out your exact movements ahead of time
Learning a new behavior activates all levels of the motor hierarchy first, but as the behavior becomes automatic, it activates only lower levels of the hierarchy.
The frontal neocortex is the locus of simulation, while the basal ganglia is the locus of automation.

15m ya - Primates and Metathinking

Theory of Mind: Group Living, Political Savvy, and Deception

The ecology of Earth had found a beautiful equilibrium, with dinosaurs comfortably at the top of the food chain for well over 150m years, fish ruling the sea for even longer, and mammals and other animals finding their respective tiny but liveable niches.
The Permian-Triassic extinction - An asteroid a few miles wide hits the Earth, killing over 70% of land-living vertebrates.
Almost every dinosaur specied was extinct, except for the birds, and the following era is the Era of Mammals.
The first primates lived in groups, and as they grew, they became relatively free from predation and food competition, and their brains exploded to well over a hundred time their original size.
These early mammals uniquely gave birth to helpless children.
Mammals engage in play much more than other vertebrates
Group living led to unique social demands, and we find that the bigger the neocortex of a primate, the bigger its social group. This correlation does no hold for most other animals. Group living helps stave off predators.
Animals in group living situations evolved tools to resolve disputes, leading to the development of mechanisms to signal strength and submission without having to actually engage in a physical altercation.
Most lineages of mammals fell into one of four buckets of social systems: solitary, pair-bonded, harems, and multi-male groups.
Groups also minimize competition through hierarchal rigidity. The strongest, biggest, and toughest become dominant.
Something new was happening in sociality with these early mammals.
There are processes of ever-escalating deceptions and counter-deceptions. Each individual is able to understand the other's intent, and that it is possible to manipulate the other's beliefs.
Apes can tell the difference between "accidental" and "intentional", and between "unable" and "unwilling".
Understanding the minds of others requires understanding not only their intentions but also their knowledge
The act of inferring someone's intent and knowledge is called "theory of mind".
Monkeys keep track of and remember each individual in their group and are able to recognize them by appearance and voice.
They keep track not only of individuals, but also of the relationships between individuals, and are extremely sensitive to interactions that violate the social hierarchy
It is not only physical power that determines one's social ranking, but also political power. One's evolutionary fitness improves with one's rank.
Friendship and trust - monkeys most often rescue those whom they have previously formed grooming partnerships with. and they go out of their way to "make up" after aggressive interactions, especially those with nonfamily members.
Early primates were frugivores, and had little competition for food from other species, and this may have opened the way to large brains and comples social groups because they had an abundance of callories and of time.
Instead of building bigger muscles, they could build bigger brains to politic their way to the top. Monkey's social behavior shows an incredible degree of political forethought.
Today's primates spend up to 20% of their day socializing, much more than most other mammals.
They have many human social instincts, both good (friendships, reciprocity, reconciliation, trust, sharing) and bad (tribalism, nepotism, deception).

Modeling Our Own Minds and Other Minds

From 70m ya, at half a grm to 10m ya at 350 grams, brains grew almost 1,000 times.
New areas include two which are extremely interconnected with each other and have new input and output connectivities:
- The granular prefronal cortex (gPFC), which wraps around the much older agranular prefrontal cortex (aPFC)
- The primate sensory cortex (PSC)
The gPFC becomes uniquely active during tasks that require self-reference, such as evaluating your own personality traits, general self-related mind wandering, considering your own feelings, thinking about your own intentions, and thinking about yourself in general. It seems to allow you to project yourself - your intentions, feelings, thoughts, personality, and knowledge - into your rendered simulations. Some people with damage to it no longer recognize themselves in a mirror.
The older aPFC gets input directly from the amygdala and hippocampus, while the new primate gPFC receives most of its input directly from the aPFC. These new primate areas seem to be constructing a generative model of the older aPFC and sensory cortex itself. As the aPFC constructs explanations (intent) of amygdala and hippocampus activity, so perhaps the gPFC constructs explanations of the aPFC's model of intent - possibly inventing a "mind".
There are levels of abstraction to explain our behavior:
- Reflexes say - I turn to the smell coming from the left because it is good
- Vertebrate structures say - Going left maximizes predicted future reward
- Mammalian structures say - Left leads to food
- Primate structures say - I'm hungry - ie, the gPFC constructs explanations of the simulation itself, of what the animal wants and knows and thinks - this is metacognition - the ability to think about thinking:
  - Reflexes drive valence responses
  - The vertebrate basal ganglia an amygdala can then learn new behaviors
  - The mammalian aPFC can then learn a generative model of this model-free behavior and construct explanation - this is a first-order model
  - The primate gPFC can then learn a more abstract generative model (a second-order model) of this aPFC-driven behavior and construct explanations of intent itself - making choices based on mind states and knowledge
The gPFC also activates when inferring other people's intent and other people's knowledge and when recognizing false-beliefs
People with gPFC damage are worse at recognizing and empathizing with other people's emotions, distinguishing jokes, and identifying faux pas.
The bigger a primate's gPFC, the higher in the social hierarchy it tends to be. And in humans, the larger is his social network and his performance on theory of mind tasks.
Our understanding of ourselves often gets cross-wired with our understanding of others, suggesting a common system for each
Perhaps the gPFC builds a generative model of your own mind to use it to simulate the minds of others. Or perhaps the work of understanding the motivations of others helps us to construct the notion of a mind of our own.
Theory of mind helps you to manage your reputation and hide your transgressions. It isn't about managing hungry predators or inaccessible prey, but rather about the subtler and far more cutting dangers of politics.

Simulation in Imitation, Teaching, and Anticipating the Future

Primate tool use is more sophisticated than in other animals. Chimps exhibit over twenty different tool-using behaviors and actively manufacture their tools.
The premotor and motor areas in a monkey activate when they perform specific fine motor skills but also when they watch others do them.
Perhaps these "mirror neurons" are the mechanism by which primates engage in theory of mind - they imagine the action and then ask themselves "Why would I do this?" to deduce the other's intentions.
People with impairments in performing specific movements also cannot understand the intentions of others doing those same movements.
These mirror neurons help us learn new skills through observation.
Without transmission from others, most chimps never figure tool use out on their own. The ability to use tools is less about ingenuity and more about transmissibility
Acquiring an entirely novel motor skill by observation may have required, or at least hugely benefited from, entirely new machinery.
Theory of mind enabled our ancestors to "actively teach", students can identify the intent of a complex skill and differentiate between the intentional and unintentional movements of experts, filtering out extraneous movements and extracting the essence of a skill.
Theory of mind may have emerged for politicking and then been repurposed for imitation learning.
The social-brain hypothesis is rivaled by the ecological-brain hypothesis - that it was the frugivore diet of early primates that drove the rapid expansion of their brains.
Anticipating future needs may be another application of theory of mind - we can infer the intent of mind - our own or another's - in a different situation from our current one. People make similar types of mistakes in tasks of theory of mind and of anticipating future needs.

100k ya - Humans and Speech

Language for Sharing, Accumulating, and Complexifying Internal Simulations

The "Great Ape Dictionary" - chronicles almost 100 sounds and gestures.
No other animals use declarative labels or symbols - "That is a cow". Other animals may make a sound meaning "Predator", but this has an implied imperative "Run"
Humans are the only animals with grammar. And all human languages have (a fundamentally similar) grammar. We are the only ones to have a natural tendency to construct and use language.
Language lets us transfer our inner simulations to each other with an unprecedented degree of detail and accuracy. While concepts, ideas, thoughts, episodic memories and plans are not unique to humans, our ability to deliberately transfer them is only possible due to language
Language expands the scope of sources a brain can extract learnings from:
- Vertebrates (reinforcing) - Own actual actions (trial and error)
- Mammals (simulating) - Imagined actions (vicarious trial and error)
- Primates (mentalizing) - Others' actual actions (imitation learning)
- Humans (speaking) - Others' imagined actions
We share the outcomes of our own inner vicarious trial and errors so that the whole group can learn from our imaginations.
We can form common myths and have entirely made-up entities and stories persist merely because they hop between our brains. Common myths of things like countries, money, corporations, and governments allow us to cooperate with billions of strangers.
The power of language is not its products but the process of ideas being transferred, accumulated, and modified across generations.
Human children are "over-imitators", performing all steps they see, including irrelevant ones.
Without language, the inner simulations of chimpanzees and other animals do not accumulate, and thus inventions that are above a certain threshold of complexity - the best ones - are forever out of their reach.
The corpus of ideas accumulated reached a tipping point when the total sum no longer fits into the brain of a single human:
- First we got bigger brains
- Then we became more specialized in our groups
- Then population size expanded
- Then we invented writing - now we accumulate our shared simulations across generations. We are the hive brain apes.
Language transformed the human brain from an ephemeral organ to an eternal medium of accumulating inventions.
Our ascent over the last few thousand years had nothing to do with better genes and everything to do with the accumulation of better and more sophisticated ideas.

Language in the Brain

Broca's area supports the production of speech and writing.
Wernicke's area supports the understanding of speech and reading.
The human neocortex has unique control of the vocal cords, surely an adaptation for using verbal language.
Language emerges from specific regions in the brain and is contained in a subnetwork almost always found on the left side of the neocortex.
Language is a specific and independent skill that evolution wove into our brains.
In other primates, these language areas are present in the neocortex, but have nothing to do with communication.
Emotional expression emerge from a system entirely separate from language:
- Other primates have only a single emotional-expression system located in the amygdala and brainstem. Human laughs, cries, and scowls are evolutionary remnants of the system from which ape hoots and gestures emerge.
- We have a second separate system, supporting voluntary control of facial muscles that is controlled by the neocortex.
The newer language system needs to be taught - if a child goes long enough without being taught language, he or she will be unable to acquire it later in life. Unlike innate emotional expressions, features of language differ greatly across cultures. And indeed a human baby born without any neocortex will still express these emotions in the usual way but will never speak.
A hardwired tendency toward gestural and vocal turn-taking seems to be the platform on which language is built.
By nine months, babies show joint attention to objects. In contrast, chimps show no interest in ensuring someone else attends to the same object they do. The more joint attention expressed by an infant at the age of one year, the larger will be her vocabulary 12 months later.
Humans may have evolved a unique hardwired instinct to ask questions to inquire about the inner simulations of others. It isn't Broca's or Wernicke's areas that are new, but rather the underlying learning program that repurposed them for language. There is not a single region but rather a curriculum that forces a complex network of regions to work together.

The Perfect Storm

Around 2.5m ya something mysterious happened with the human brain rapidly becoming over three times larger.
There are various adaptations:
- There is a shift to eating meat (up to 85% meat diet!) and to making tools to cut it.
- They're walking upright.
- Their shoulders and torsos become adapted for throwing.
- Their legs and feet adapt for endurance running
- While other mammals pant to lower their body temperature, humans sweat
- Their mouths and guts shrank and they invent cooking with less digestion. Every human culture uses cooking
- They have controlled use of fire
- Big brains are hard to fit through birth canals and the human solution is premature birthing - babies are not born when they are ready to be born, but when their brains hit the max size to fit through the birth canal. It takes a human brain 12 years before it has reached its full adult size.
- As a result the parenting style changed, leading to longer-term pairings.
- "Grandmothering - only humans and orcas have females that are not reproductively capable until death. Maybe the menopause evolved to push grandmothers to shift their focus to supporting their children's children.
- The larynx and vocal cords of our ancestors were not adapted to vocal language until about 500k ya, and there is substantial evidence that language existed at least 100k ya. Our common ancestor from 100k ya almost definitely spoke a language of a complexity equal to ours.
Most altruistic behaviors are the result of kin selection, and the essential feature for reciprocal altruism to successfully propagate throughout a group is the detection and punishment of defectors.
Humans are, relative to other animals, both by far the most altruistic to unrelated strangers and the most cruel - no other animal commits genocide.
Homo Erectus probably had the ability to assign declarative labels and perhaps even use some simplified grammars*
Language my have emerged as a trick between parents and children, since the most prominent learning program for language is the interplay of joint attention and proto-conversations between parents and children.
As much as 70% of human conversation is gossip:
- Gossip enables a stable system of reciprocal altruism by sharing information about betrayals and about heroic acts.
- The more severe the costs of cheating, the more altruistic it was optimal to behave.
- For every incremental increase in gossip and punishment of violators, the more altruistic it was optimal to be.
As social groups got bigger and more ideas accumulated across generations, humans became better hunters and cooks and brains got bigger and births got earlier.
By 100k ya, there were at least four species of humans spread out across the planet:
- Homo floresiensis was in Indonesia
- Homo erectus was in Asia
- Homo neanderthalensis was in colder Europe
- Homo sapiens had remained in Africa
By 40k ya, only Homo sapiens remained.

ChatGPT

Both LLMs and the neocortical areas for language seem to be engaging in prediction, generalizing past experiences and applying them to new sentences, and guessing what comes next, though more is happening in your brain than merely the automatic prediction of words.
The foundation of language learning is not sequence learning but the tethering of symbols to components of a child's already present inner simulation.
The neocortex evolved long before words, already wired to render a simulated world that captures an incredibly vast and accurate set of physical rules and attributes of the actual world.
The human brain contains both a language prediction system and an inner simulation.
The intertwining of mentalizing and language is ubiquitous. Wernicke's area is right in the middle of the primate mentalizing regions.
We are capable of puppeteering other minds because language is, it seems, built right on top of a direct window to our inner simulation. Hearing sentences directly and automatically triggers specific mental imagery. This is why if someone is saying something that upsets us, we cannot simply "not listen"; we must cover our ears, otherwise the words directly trigger simulations whether we like it or not.
By predicting not just the answer but also the next step in reasoning about the answer, an LLM begins to exhibit emergent properties of thinking.
In the human brain, language is the window to our inner simulation. Language is the interface to our mental world.
If there is anything that truly makes humans unique, it is that the mind is no longer singular but is tethered to others through a long history of accumulated ideas.