Practice
if you ask a chatbot like ChatGPT a question, it will answer quickly and probably correctly. If you
ask it many questions, it will seem to recall everything you told it earlier. If you challenge any of its
answers, it may report, in fluent sentences, that it finds your response frustrating. Plainly a great
deal is happening behind the screen. But is any of it understanding? Does the program want
anything? Does it feel anything? Or is it an intricate mechanism arranging the right words in the
right order, with no one home inside? You may have a firm intuition here. Consider, how you would
go about proving it to someone who disagreed?
Now ask the same kind of question about yourself. Take a single moment of fear. You might be
out hiking in the forest when suddenly you notice a large animal moving nearby in your peripheral
vision. In that instant your heart pounds. Adrenaline pours into your blood. A small structure buried
in the brain called the amygdala drives the whole cascade. That is one description of the moment. It
is a completely physical description. There is a second description we can use. The fear feels like
something. There is the lurch, the jolt, the urgent wish for it to stop. Both descriptions are about the
very same instant. Are they two different accounts of one event? Or one event rendered in two
languages? And would the experience be different, in principle, if the system doing the fearing were
built from silicon instead of cells? Or if it were built from a brain organized very differently from your
own?
These puzzles are old, but they are not solved or trivial. A lot is at stake based on how we
answer them. In artificial intelligence, if a machine grows fluent, does it thereby have a mind? And if
so, what kind? In ethics and law, we hold people responsible for what they do. Could we ever hold a
machine responsible, and what would have to be true of it before that made any sense? In
neuroscience, what should even count as a good explanation of a thought or a feeling? In
psychology, can mental notions like belief and desire be translated into the language of biology? Or
do they name something biology will never fully reach?
Two general questions sit underneath all of these, and this module is built around those
questions. The first is, what is the relationship between the mind and the brain? The second is, what
would it take for something that is not a human brain (whether an animal, an alien, or a machine) to
have a mind? These are not two separate subjects. The second question is a version of the first,
asked about a different kind of stuff. Whatever we conclude about how minds relate to brains will, at
the same time, tell us what else besides a brain might have a mind.
1. Mind and Brain
Page 1 of 28
It is worth pausing on why a course largely about how minds and brains evolved and how they
work should open with questions this abstract. The rest of this book treats minds and brains as
natural objects with histories. They are something that arose in living things, took different forms in
different lineages, and can be studied piece by piece like any other product of biology. But that
whole project takes something for granted. It assumes there is a mind there to have a history, and
that a mind is the kind of thing the methods of science can reach. Both assumptions can be
doubted, and thoughtful people have doubted them, in more than one way. If the mind were a non-
physical soul, its story would not be a chapter of biology at all. But the doubts do not end with souls,
and the ones this chapter dwells on longest are more modern. Even granting that the mind is fully
physical, we can wonder whether our everyday mental words for things like belief, desire, and fear
name anything real, or whether a mature brain science will find nothing there for them to pick out.
And even granting that they do name something real, one can still ask whether the mind can be
understood in its own terms at all, or whether any real understanding of it must run through the
biological details of the brain beneath. Each of these doubts, taken seriously, changes what a
science of the mind could be. So before we can ask how a mind evolved, how a brain produces
one, or whether a machine could ever have one, we have to know what a mind is, and whether it
sits within the reach of science. This chapter clears that ground. It is what makes “the evolution of
the mind” a coherent subject rather than a figure of speech.
There is a second thing to flag before we begin. This chapter is philosophy more than it is
science, and that is not an accident or an apology. Science answers questions by observation and
experiment: you form a hypothesis, you test it against the world, and the world gives you
information you use to judge the value of your hypothesis. But some questions have to be settled
before that process can even start. What do we mean by “mind”? What would count as a physical
explanation of one? Is the mind the sort of thing an experiment could detect at all? No
measurement answers these, partly because they are questions about which measurements would
count. This is the work of philosophy. It sharpens concepts and draws distinctions. And for
questions that observation cannot now (or ever) settle, it evaluates whether a set of commitments
hangs together. Far from being opposed to science, this work establishes the foundation that
science is built on.
Because these questions are not settled by experiment, you will meet several competing
answers in the pages ahead, and no one of them can simply be proven the way a claim in the lab
can. That is easy to misread, so it is worth being careful. “Cannot be settled by experiment” does
not mean there is no fact of the matter, or that any answer is as good as any other. It means the
tools that decide the question are different ones. Philosophers and scientists reason toward these
views, and between them, by asking which account is internally consistent, which fits best with
everything else we know about the brain and the world, and which explains the most while
assuming the least. Judged that way, the answers are not equal. As we will see, most contemporary
brain and cognitive scientists have converged on one broad family of views, for reasons this chapter
will lay out. But argument rarely compels the way a decisive experiment does. A coherent, well-
supported position can almost always be resisted at some cost, by giving something up elsewhere.
That is why thoughtful people still disagree, and why this stays a live conversation rather than a
closed case.
1. Mind and Brain
Page 2 of 28
Mind
perceiving · recognizing · learning
remembering · reasoning · deciding
communicating
Consciousness
awake, aware, knowing that one knows
Qualia
the raw feel of an experience
The dashed boundaries are contested: some views treat qualia as the whole of consciousness, and
some hold that mind simply is consciousness.
Figure 1.1 — Mind, consciousness, and qualia, with the boundaries the chapter is careful about drawn as
contested rather than crisp.
People have been offering answers to these questions for as long as they have been asking
them. The answers pile up into a bewildering heap: souls and spirits, brain states, patterns of
behavior, flows of information. There are many named perspectives attempting to explain the mind–
body relationship, such as dualism, materialism, idealism, and emergentism. But rather than march
through different answers one at a time, we will try to do something more helpful. First we will pin
down what we even mean by “mind.” Then we will ask a short sequence of questions about how the
mind might relate to the brain, and show how the famous positions fall out of how one answers
these questions.
§1.1.1 — What We Mean by “Mind”
Before asking how the mind relates to the brain, we had better say what we mean by the word
mind. Loose terms often lead to misunderstandings or debates that wouldn’t exist with clearer
definitions. And few terms are looser than mind. In this book, we use mind to refer to the broad set
of capacities a system has for taking in information about the world, holding onto it in some usable
form, and putting it to work in guiding what the system does. Perceiving, recognizing, learning,
remembering, reasoning, deciding, communicating: each is a way of turning information into
something useful. Together, they are what we will mean by a mind. This is the same information-
processing picture the previous module introduced, now serving as our working definition.
Stated that broadly, our definition of mind has blurry edges, and these edges are where
interesting fights sometimes break out. Consider a motion sensor on a door. It takes in information
that something moved, and acts on it, sliding a door open. Or consider a smart thermostat that
senses the temperature, compares it against a goal, and switches the air conditioner on or off.
Almost no one wants to grant that a thermostat has a mind. Yet it is doing, in miniature, the very
thing our definition names. Consider a much more complex artificial entity. Large language models
like ChatGPT, Claude, and Gemini produce strikingly fluent and apt responses. Many nonetheless
insist they do not and cannot have minds. One common dismissal is that they are “only predicting
the next word,” and therefore are not really thinking at all. Is that right? Does how a system
processes information, or how much of it, decide whether the word mind applies? We will not settle
these questions here. The point for now is that “mind” has no crisp boundary you can check a
1. Mind and Brain
Page 3 of 28
system against. Where the line falls is itself one of the field’s live debates, and it is the very problem
the machine cases keep forcing us to think about. Our definition is a starting place, not a test.
One aspect of the mind sits apart from the rest, and it is worth setting aside for now. When we
imagined that jolt of fear a moment ago, part of what we noted was that there is something it is like
to feel afraid. That felt, qualitative character of an experience is what philosophers call qualia.
Qualia are closely tied to consciousness, though the two are not simply the same thing
(Figure 1.1). Consciousness is used broadly, covering everything from being awake and aware to
knowing that one knows. Qualia name something narrower: the raw feel of an experience. Some
philosophers treat qualia as the defining core of consciousness, while others keep the wider sense
in view. We need not adjudicate between those views right now. What matters here is that this
cluster of problems is hard and distinctive enough that a longer version of this course would give it a
week to itself. Consciousness is a real part of the mind, or at least of the human mind. But it is only
one part. And at least descriptively it can be pried away from the rest. A system might perceive,
remember, and reason at a high level while it stays a separate question whether anything is
experienced along the way. Some views resist that separation, holding that mind simply is
consciousness. But this is a book about the mind in the broad sense, and consciousness is one
subtopic within it rather than its center. Whatever one finally concludes about consciousness, the
relationship between mind and the physical stays a rich and largely tractable question across the
whole territory of mental life. So we flag consciousness here, and then mostly set it aside for now.
With “mind” defined well enough to work with, we can take up the relationship we came for.
Philosophers call it the mind–body problem: how does the mind relate to the physical stuff it
seems to arise from? We will often put it even more neutrally, as the relation between mind and the
physical, because part of what is in question is whether that physical stuff has to be a brain at all.
We will build our understanding of it from a short sequence of questions, and the familiar named
positions will fall out along the way as answers we can label, rather than as boxes we start from.
§1.1.2 — What Kind of Thing Is a Mind?
Our first question is the most basic one there is: what kind of thing is a mind? This is a question
about the mind’s ontology, or its basic nature. Three answers have serious defenders. The first is
that mind is a distinct kind of thing in its own right, separate from physical stuff. The second is that
mind is nothing but the physical stuff of the body, so that “mind” is only another word for what the
brain is doing. And the third is that mind is a real property of physical stuff, as genuine as the stuff
itself but not identical to any particular piece of it. We will take the three in turn.
One quick test sorts the first answer from the other two. Imagine two systems physically identical
down to the last particle, in exactly the same state at the same instant. Must they have exactly the
same mind, the same thoughts and the same feelings? If you think they could differ, that one could
have a mind, or an experience, the other lacks, then you hold the first answer, that the mind is
something over and above the physical. If you think the physical facts settle the mental ones, so
that identical bodies must house identical minds, then you hold that the mind is physical, and your
quarrel is only between the second answer and the third. Most scientists take the physical side; the
general commitment that the physical fixes the mental is called materialism.
Start with the first answer, that the mind is a thing in its own right, separate from physical stuff.
Its classic defense (though by no means the first) comes from 17th-century philosopher René
Descartes. Descartes was wondering what he could actually be certain about. Perhaps some
powerful deceiver was tricking him at every turn. Maybe all of his experiences were a dream, or a
hallucination, or a fabrication. Resolving to doubt everything he possibly could, Descartes found that
he could doubt that he even had a body, or that the world outside him even existed. The one thing
he could not doubt was that he was thinking. The very act of doubting proved there was a mind
doing the thinking. Descartes argued that the mind’s existence itself was a certainty, entirely
separate from the body, which could be doubted away. This intuition is far older than Descartes,
1. Mind and Brain
Page 4 of 28
running back through the medieval philosopher Avicenna to Augustine to Plato’s eternal soul, and it
carries deep religious roots. Descartes simply gave the argument its sharpest edge.
Now aim Descartes’ intuition at a machine. Picture a being physically identical to a person.
Wired the same, behaving the same, flinching from pain, saying “that hurts,” and seeming to mean it
as far as anyone could tell. But this being has no inner experience of any kind. No actual feeling of
pain. No light on inside. Nobody home. The philosopher David Chalmers made this thought
experiment famous and named its subject a philosophical zombie. Notice that what the zombie is
missing is exactly the consciousness we set aside a moment ago. This is no accident: the
independent view draws most of its force from felt experience. That we can apparently imagine
such a being at all is what gives the view its grip. If the physical facts leave it open whether
experience is present, then experience looks like something over and above the physical facts.
At the center of this answer to the mind–body problem sits the view called substance dualism,
sometimes just called dualism. Dualism is the view that mind and body are two different kinds of
substance, and the non-physical mind, though it interacts with the brain, cannot be reduced to it.
Around dualism sit related views that also refuse to ground mind in the physical. Idealism is the odd
cousin that keeps only the mental, and demotes the physical. Idealism holds that the physical world
is in fact all an illusion in some entity’s mind (often God’s). Panpsychism is another view related to
dualism that holds that qualia (experience) are basic features of all matter, present in some dim
degree even in a rock or an electron. Neutral monism proposes that mind and matter are two
arrangements of a more basic stuff that is itself neither.
This commitment about the nature of mind fixes what dualism says about machine minds. If
having a mind requires a share of non-physical mind-stuff, then no feat of engineering can ever
produce one, because engineering only arranges physical parts. This is the philosophical zombie
from a moment ago, now turned on real hardware. However fluent or intelligent-seeming a
computer is, it is an empty mechanism, saying all the right things with no one inside. Panpsychism
is the family’s lone exception to this. If experience saturates all matter, then a chip has its faint
share too, strange as that sounds, though it holds no more of it for being a computer than a stone
does.
Why is this first answer, that mind and the physical world are distinct kinds of things, so
tempting? And if it is so tempting, why do so few scientists hold it? Its appeal is real. It sidesteps the
hardest puzzle of all, how mere matter could ever give rise to felt experience, by denying that matter
has to. It honors the strong sense that consciousness is special and not just more chemistry. And it
leaves room for the existence of an immaterial soul that might continue to exist independent of the
body. What makes it problematic as a scientific view is twofold. First is the interaction problem: if the
mind is genuinely non-physical, how does it affect the physical neurons that move your hand? Every
answer seems to smuggle the mind back into the physical world it was meant to transcend. Second
is untestability: a mind that leaves no physical trace is a mind no experiment can find. It is by its
very nature outside of science.
For all its intuitive pull, the first answer (mind is a separate kind of stuff from the physical) keeps
few working scientists in its camp, for the reasons just given. The great majority hold that the mind
is not a separate thing at all, but physical through and through. Yet that agreement hides a deep
disagreement, because there are two very different ways to be a materialist about the mind. They
are our second and third answers, and the difference between them shapes everything that follows.
§1.1.3 — The Mind as Nothing but the Physical
The second popular answer to the mind–body problem is quite blunt: the mind just is the brain, and
a separate mental vocabulary earns its keep only as a convenient shorthand, if at all. On this
answer, mental talk adds nothing that a physical vocabulary could not, in principle, replace.
The starkest historical form was radical behaviorism. John Watson and B. F. Skinner urged
psychology to study only what could be observed, namely behavior, and to banish talk of inner
1. Mind and Brain
Page 5 of 28
mental states as unscientific ghost-chasing. For a time (roughly from 1920-1960) behaviorism
dominated. What broke it was its inability to account for the very things minds most obviously do:
understand and produce language, reason, remember, and plan. The cognitive revolution of the
1960s and 1970s reopened the door to a science of inner mental processes.
Another variant of this answer is identity theory. Identity theory grants that mental states are
perfectly real, but insists each one simply is a physical brain state. Its slogan is that “pain is the
firing of C-fibers.” C-fibers are a particular kind of nerve fiber, the ones that carry the slow, dull,
aching kind of pain from the body toward the brain. The claim is not just that this firing causes pain,
or reliably comes with it, but that the pain is that firing. One event, two names. The model is the
earlier discovery that heat is the motion of molecules, not caused by it or merely correlated with it,
but the very same thing. Identity theory bets that every mental state will turn out this way, each one
revealed to be some specific state of the nervous system. But pinning a mental state to one specific
physical state carries a cost. If pain simply is the firing of C-fibers, then anything lacking C-fibers, an
octopus, an alien, or a silicon computer, could not be in pain no matter how it behaved.
Philosophers call this the chauvinism problem. Escaping it will be one of the attractions of the third
answer.
The boldest version of the second answer goes further still. Eliminative materialism agrees
that the mind is physical, but adds that our everyday way of talking about it is not merely
incomplete, it is mistaken. We typically explain someone’s behavior by saying she opened the fridge
because she wanted a snack and believed there was food inside. When we do this, we are leaning
on everyday language about beliefs, desires, hopes, and fears. The argument from eliminative
materialism is that these words are actually unscientific and misleading. Philosophers refer to these
kinds of words and the explanations behind them as folk psychology. The eliminativist’s claim is
that a mature brain science will no more find beliefs and desires inside the head than modern
medicine found the four humors it once blamed for disease. Concepts like those were not refined
and kept; they were dropped, because there was nothing there for them to name. Folk psychology,
the eliminativist says, is headed for the same fate. It will not be tidied up and matched to brain
states, but replaced outright. The best-known defenders of eliminative materialism are the
philosophers Patricia Churchland and Paul Churchland.
The reductive family’s strengths are genuine. It fits neuroscience’s methods hand in glove. It is
parsimonious, and posits no vague mental extras. But its troubles are equally real. There is the
chauvinism problem noted above, which implies that any difference in brain structure would make
for a different mental state. There is eliminativism’s air of self-refutation, since to assert “there are
no beliefs” is to express one. There is the stubborn fact of felt experience, the “what it is like” that a
list of brain states seems to leave out. And there is the working practice of psychology itself, which
explains and predicts behavior well in terms of memory, attention, and emotion. On the “do
machines have minds” question, identity theory says no by way of chauvinism. Eliminativism does
something stranger: it eliminates the question rather than answering it. If there are no folk “minds”
for us to have either, then asking whether a machine has one is as confused as asking it about a
person.
1. Mind and Brain
Page 6 of 28
Eliminated
The four humors named nothing, and
were dropped rather than refined.
Eliminative materialism’s bet
about beliefs and desires.
Identified
Heat turned out to be molecular
motion itself. One thing, two
names.
Identity theory’s bet: each mental
state is one particular brain
state.
Grounded, and kept
Temperature is explained by
statistical mechanics, and
thermodynamics still does real
work.
Where most of the field sits:
mental terms name something real,
explained by the brain without
being replaced by it.
Figure 1.2 — Three things that can happen to a higher-level concept, with the historical case above and the
matching claim about our mental vocabulary below.
Stepping back, these versions of the second answer echo two things that can happen to a
higher-level idea in the history of science. It can be discarded as naming nothing, the way the four
humors were, which is the eliminativist’s bet about folk psychology. Or it can be kept but shown to
be identical to something lower down, the way heat is equated with molecular motion. This is the
identity theorist’s view. But a higher-level idea can meet a third fate, unlike either of the first two
(Figure 1.2). In this case it is grounded in a deeper theory without being replaced by or equated with
it. Most brain and cognitive scientists think the mind fits this pattern best. That third pattern we
explain next.
§1.1.4 — The Mind as a Real Property of the Physical
The third answer is the one most working brain and cognitive scientists hold. Contrary to the
eliminativist, it says our mental terms refer to something real. But the mind is neither a thing in its
own right (as a dualist would say), nor a redundant name for the brain (as an identity theorist would
say). The dualist’s deepest mistake, on this view, is a kind of category error. The dualist asks what
kind of thing the mind is, and reaches for a special, non-physical kind, when the mind is not a thing
at all. It is a property of the physical stuff we already have, not a further item standing beside it.
Compare mass. Mass is not an object tucked inside a rock; it is something the rock has, a property
of physical matter. Temperature is another such property, being alive is a third. In Aristotle’s older
language, mind is the form a lump of matter takes when it is organized a certain way. The mind, this
answer says, is a property of just that kind. It is a property a physical system has when its parts are
organized and active in the right way. Mind is as real as the matter itself, but it is not a separate
ingredient added to it.
Under this view, temperature repays a closer look, because it also marks the borders this
answer shares with its neighbors. We earlier noted that physicists came to say that heat is the
motion of molecules. But even knowing this, thermodynamics, the science of heat, temperature, and
pressure, still does real work above and beyond theories of molecular motion. Its laws tell an
engineer how an engine or a refrigerator must behave. A deeper theory, statistical mechanics, later
explained where those laws come from, by treating heat and pressure as the combined motion of
countless molecules. Yet no one throws temperature away to track molecules one by one. It is
grounded in molecular motion without being replaced by it, and without being identical to any one
arrangement of it, since countless different arrangements count as the same temperature. That is
the line between this answer and identity theory, which instead equates a mental state with one
specific physical state. A different line separates it from panpsychism: this property surfaces only
1. Mind and Brain
Page 7 of 28
where matter is organized and active in the right way, as life does. It is not, as the panpsychist
supposes, a basic ingredient present in every scrap of matter. A rock has a mass and a temperature
but no mind, because it is not arranged to have one. On the third answer, a mind is what suitably
organized matter has and does: a real, higher-level property, not a ghost added to it.
Let us zoom back out and contrast how scientists tend to evaluate the third view compared with
the other two. Why not the first answer? Because materialism keeps explaining things. There has
never been a scientifically successful account of how a mind, or anything else, could exist apart
from a physical instantiation. And we understand the mind better every time we learn more about
the brain. The second answer is therefore half right. It is right that the mind is physical and that the
brain carries real explanatory weight. Its error, as the next step argues, is the further leap that
mental-level explanation can then be thrown away.
Why not stop at reduction to the brain? Because psychological explanations work, and they add
something a neuron-by-neuron story does not. You cannot in practice predict or explain a person’s
decisions in terms of movement of molecules or atoms. The calculation is hopeless, and the higher-
level account is how the science actually gets done. Most sciences, moreover, reduce without
eliminating, as thermodynamics and the gene remind us. And there is a deeper reason many
cognitive scientists take seriously, one that deserves its own name.
That reason is multiple realizability, the idea that one and the same mental state can be
realized in different physical materials, just as one temperature is realized by countless different
arrangements of molecules (Figure 1.3). The idea has an intuitive starting point. When you and a
friend both feel hunger, both grow angry, both recognize an apple, or both believe that the Earth
goes around the Sun, we find a lot of value in saying the two of you are in the same mental state.
Yet no two brains are wired exactly alike, and it is very unlikely that the state is realized in precisely
the same physical pattern in both of you, which is just what a strict identity theory would need. If the
mental state is shared while the physical detail is not, that state cannot be identical to the detail.
The point only sharpens as the systems grow more different. Pain in a person, pain in an octopus,
pain in some hypothetical alien, perhaps pain in a machine, may be functionally equivalent while
implemented in vastly different ways or even materials. If that is right, then no single physical state,
no particular firing of C-fibers, can be pain, if pain turns up in systems that have no C-fibers at all.
Multiple realizability is what defeats the identity theorist’s chauvinism. It is also what turns “could a
non-brain have a mind?” from a confusion into a real scientific question. It ties the mind–body
question to the machine question, which is why it deserves more than a passing mention. It is not
beyond dispute. And we will see the multiple realizability perspective challenged later, especially
with regard to machines. But its plausibility carries much of the weight behind this view. Multiple
realizability, then, lets the mind be a real, physical property without chaining it to one kind of matter.
Our third main kind of answer to the mind–body problem is mind as a real property of the
physical: mind is a property of physical matter arranged in a certain way or doing certain things. It
is shared by a family of more specific views, which differ in how they spell it out. Emergentism
holds that mental properties genuinely arise from physical organization the way wetness arises from
many H₂O molecules even though no single molecule is wet. Functionalism defines a mental state
by the causal role it plays, by what brings it about and what it brings about in turn, rather than by
what it is made of. A calculator’s “add” operation is whatever plays the adding role, on a phone or a
laptop alike. And by the same logic a mental state is whatever plays its role, in a brain or perhaps
elsewhere. This is the position that opens the door to machine minds. If a mental state simply is a
functional role, then anything that fills the role has the state, brain or no brain. It is, more or less, the
working assumption of most people involved in artificial intelligence. In its sharpest contemporary
form, this family treats a mental state as a real pattern of causal organization, something a physical
system genuinely has, discoverable by experiment yet reducible to no single neuron, so that a mind
is real the way an organization is real rather than the way an object is. A third perspective,
Embodied or Enactive Cognition, enters as a pointed caveat. Perhaps thinking is not only
something the brain does, but something a whole body does in an environment. If this is true, it is
1. Mind and Brain
Page 8 of 28
the sharpest check on a naive functionalism that expects a disembodied program to suffice for
mind. A fourth perspective, the levels of analysis approach, we hold back for the whole of §1.2,
since it is the framework the rest of the course runs on.
One temperature, many arrangements of molecules
20 °C 20 °C 20 °C
One mental state, many kinds of matter — the claim of multiple realizability
a human brain an octopus a silicon chip
the claim is that all three could be in the same functional state
Figure 1.3 — One property, many realizations, drawn twice in the same grammar. The top row is settled
physics; the bottom row is the claim this section is making.
What does this view say about machines? It is the home of “yes, in principle” (Table 1.1).
Functionalism and the levels approach make a machine mind conceptually possible without
claiming that any machine yet built actually has one. Two thinkers who agree the mind is a real
physical property can still part ways on machines, because a further question stays open: does a
mind’s physical basis have to be biological? For the cognitive capacities, most are willing to say a
machine could in principle have them. It is usually over consciousness, the part we set aside, where
many philosophers and scientists are most likely to believe that biological matter is necessary. One
can hold the mind fully physical and still suspect that living matter has features, in how it organizes
and maintains itself, that a simulation would copy without reproducing. The neuroscientist Anil Seth
argues in this spirit; it is a bet about substrate, not a lapse back into dualism. The course returns to
weigh all of this in Module 2. Here we set the question up; we do not answer it.
§1.1.5 — How Freely Can We Study the Mind?
Settling that the mind is a real property of the physical still leaves one more question, which is as
contentious today as any in the field. How freely can we study the mind in its own terms? Grant that
a mental state is a real property of some physical system. Must a good theory of it keep the physical
details in view, or can it abstract them away and work at the mental level alone? Philosophers call
this the mind’s explanatory separability. It is a question about method, not about what the mind is
made of.
1. Mind and Brain
Page 9 of 28
Table 1.1 — Eight named positions, set beside one another. The machine column is the one to read across:
what each position says about minds and brains fixes what it can say about anything else that might have a
mind.
Position What a mind is
made of
Do mental terms
name anything
real?
Could a machine
have a mind? Its main problem
Substance
dualism
A non-physical
substance,
interacting with
the body
Yes
No — engineering
only arranges
matter
How anything
non-physical
could move a
neuron, and what
experiment could
ever find it
Idealism
Only the mental;
the physical is an
appearance within
a mind
Yes No
Why the physical
world is so
orderly, and so
reliably shared
Panpsychism
Ordinary matter,
with experience
basic to all of it
Yes
Faintly — but no
more than a stone
does
Why organized
matter has one
unified mind
rather than a dust
of tiny ones
Radical
behaviorism
The body; only
behavior is fit to
study
No — set aside as
unscientific
The question is
not asked
Cannot account
for language,
reasoning,
memory, or
planning
Identity theory
Brain states, one
for each mental
state
Yes No — no C-fibers,
no pain
Chauvinism: it
rules out any
differently built
system
Eliminative
materialism
Brain states; folk
psychology
names nothing
No
The question
dissolves rather
than gets an
answer
Asserting that
there are no
beliefs is itself
expressing one
Emergentism
Physical matter,
organized; mind is
a property it has
Yes Yes, in principle
Saying precisely
what emergence
amounts to
Functionalism
Whatever fills the
causal role, of any
material
Yes
Yes — the role is
what matters, not
the stuff
Leaves out felt
experience, and
perhaps the body
At one pole sits strong autonomy. Psychology, on this view, is a science in its own right, standing
to neuroscience roughly as chemistry stands to physics: grounded in it, but not replaceable by it.
You can describe the procedure a mind runs without much caring what it runs on, exactly as you
can describe a chess program without describing transistors, or for that matter describe chemical
1. Mind and Brain
Page 10 of 28
reactions without describing the micro-details of the underlying protons, neutrons, and electrons.
This is the liberal functionalist’s confidence, and it is one wing of the third answer. At the other pole
sits a more cautious view. The mind is real and physical, but the details of how it is realized shape
what it can be. A theory that abstracts away from the body and the brain will miss something, quite
possibly something important or critical. Embodied cognition pulls this way, and so does the view
that mental kinds are realization-sensitive. Two systems that seem to behave alike when we view
them from a zoomed out or abstract perspective may not actually share a mental state unless they
are organized alike underneath. And those differences in implementation will be noticeable at the
functional level if we look closely enough. On this view the levels of a mind are separable but not
independent. You can ask about each on its own, yet the answers constrain one another.
Q1. Do the physical facts fix the mental
facts?
No Mind is a distinct kind of thing
substance dualism · idealism · panpsychism ·
neutral monism
Yes
Q2. Do our mental terms name anything
real?
No Mind is nothing but the physical
eliminative materialism · radical behaviorism
Yes
Q3. Is a mental state identical to one
specific physical state?
Yes Each mental state is one brain state
identity theory
No
Mind is a real property of the physical
emergentism · functionalism · levels of analysis — where most of the field sits
Q4. How freely can such a mind be studied in its own terms?
a matter of degree, not another fork — and the field is genuinely split along it
Strong autonomy: psychology as chemistry is to physics Realization-sensitive: embodiment, the speed of neurons
Figure 1.4 — The positions as answers, arrived at by asking four questions in order rather than by sorting the
world into boxes.
Neither pole is obviously right, and the disagreement is not without consequences. It decides
how far a science of the mind can float free of a science of the brain. To make progress on it we
need a way to lay the levels of a mind side by side, so that we can see which questions belong to
which and how far each constrains the others.
§1.1.6 — Where the Field Sits
Step back, and the section has been a short sequence of questions. Before anything else we fixed
what “mind” means, and set consciousness aside as a subtopic of its own. Then we asked what
kind of thing a mind is, and found three serious answers: a distinct kind of thing (dualism, idealism),
1. Mind and Brain
Page 11 of 28
nothing but the physical (behaviorism, identity theory, eliminative materialism), or a real property of
the physical (emergentism and functionalism, including the view of a mind as a real pattern of
causal organization). Most people in the field tend to endorse a perspective in the third category.
Then we asked how freely such a mind can be studied in its own terms, and found the field
genuinely split, from strong autonomy at one end to realization-sensitivity at the other (Figure 1.4).
The famous positions are not boxes we sorted the world into. They are just names for the different
ways of answering these and related questions.
So this is where most contemporary brain and cognitive science sits. The mind is a real property
of the physical, dependent on it and yet demanding explanation in its own terms, with the exact
degree of that independence still under active debate. That is a good description, but so far it is still
quite vague. “Study the mind at its own level” has a satisfying ring, yet we have not said what those
levels are or how they relate to the brain beneath them. In §1.2 we turn this into a concrete, working
framework, Marr’s levels of analysis, which gives us a way to lay a mind’s levels side by side and
see how each constrains the others. This is the structure the rest of the course is built on.
§1.2 — Levels of Analysis
§1.2.1 — One System, Many Kinds of Explanation
We ended the last section by stating that, to most brain and cognitive scientists, the mind is a real
property of physical systems. It is an entity that can be studied, at least in part, in its own terms. But
it is also answerable to the brain beneath it. That has a satisfying ring. But it stays empty until we
say what “its own terms” are, and how those terms relate to one another. Before we reach for the
mind, it helps to see the same layered situation in something simpler and fully physical, where
nothing about souls or consciousness can cloud the view.
Take the heart. Suppose you want to understand it. One good explanation describes what it is
made of and how the parts move. The heart is muscle fibers contracting in sequence, valves
opening and closing, and the electrical pulses that set the rhythm. That is a real kind of explanation,
of the kind that a physiologist can give in exquisite detail. But notice that it leaves a different
question untouched. What is the heart for? The answer, that it pumps blood to carry oxygen and
nutrients through the body, is not a fact about muscle tissue at all. It is a fact about a problem the
heart solves. And there is a third question sitting between the other two. How, by what strategy,
does it solve that problem? Pushing fluid one way through a system of one-way valves is a strategy,
and you could describe it without naming a single protein, or even knowing the pump was built from
biological tissue rather than metal.
We have identified three questions, then, about one object: 1) what problem is being solved, 2)
by what strategy, and 3) in what physical stuff (Figure 1.5). These are not competing questions with
mutually exclusive answers. The molecular story does not make the “what for” false, and knowing
the job does not tell you the tissue. Each is a real question with its own answer, and you have not
fully understood the heart until you can answer all three. This is the heart (so to speak) of the
multiple levels of analysis perspective, so it is worth stating plainly: a single system can demand
several different kinds of explanation at once, and they complete one another rather than compete
with each other.
There is a reflex worth resisting here, because it returns throughout the course. It says: surely
the physical story is the real explanation, and the talk of “problems” and “strategies” is loose
shorthand we will drop once we understand the tissue. We met the reasons against this in §1.1,
when we saw why the mind is not thrown out in favor of the brain. The same reasons apply one
level up, to explanation itself. A complete account of the muscle, every fiber and every firing
mapped, would still leave two questions open. It would not tell you what the heart is for, that it
1. Mind and Brain
Page 12 of 28
solves the problem of moving oxygen through a body. And it would not tell you how it does the job,
that its strategy is to push blood one way through a system of one-way valves. Those are further,
real facts about the system, and no amount of detail about the tissue will hand them to you. The
same holds when we turn from hearts to minds. A complete wiring diagram of the brain, every
neuron and every connection, would still not tell you what problem a given circuit is solving, say,
recognizing a face, or how it solves it, by what series of steps it turns a wash of light into a name. A
map of the stuff is not yet an account of the problem or the method.
The Heart at Three Levels
Computational
What problem is being
solved, and why?
Move oxygen and nutrients to every tissue in a body.
Algorithmic
By what strategy? Push fluid one way, through a system of one-way valves.
Implementational
In what physical stuff,
and how?
Muscle fibers contracting in sequence, one-way valves, and
electrical pulses setting the rhythm.
Figure 1.5 — The heart, answered three ways. Each band is one kind of question, and the three answers
complete one another rather than compete.
This insight, that one system supports several complementary kinds of explanation, is old, and it
has been carved up in more than one way. Aristotle held that fully explaining a thing takes answers
to four different questions about it: what it is made of, how its parts are arranged, what brought it
into being, and what it is for. The philosopher Daniel Dennett describes three “stances” one can
take toward a system, treating it as mere physics, as a piece of design, or as an agent with goals. A
long tradition in the philosophy of biology explains a phenomenon by laying out the mechanism that
produces it: its parts, their operations, and their organization. These frameworks overlap and mostly
agree. This course adopts a different one, a system of three levels of analysis proposed by the
vision scientist David Marr. We adopt it not because it is uniquely correct but because its three
levels line up cleanly with how brain and cognitive science already divides into disciplines, and with
the evolutionary spine of the course. Building that alignment is the work of the rest of this chapter.
We turn to Marr’s three levels now.
§1.2.2 — Marr’s Three Levels
David Marr’s framework was born from a complaint. Marr studied vision in the 1970s, at a time
when neuroscience was learning an enormous amount about individual visual neurons: which ones
fired, when, and in response to what. And yet, Marr felt, the field was not much closer to
understanding how seeing works. Piling up facts about neurons, he argued, was like trying to
understand the flight of a bird by studying only its feathers. You could catalog every barb and
1. Mind and Brain
Page 13 of 28
filament and still have no idea why birds can fly. Something essential was being missed, and no
amount of the same kind of detail would supply it.
Computational
What problem is being
solved, and why?
evolutionary biology,
behavioral ecology,
anthropology
Algorithmic
By what strategy?
cognitive psychology,
artificial intelligence
Implementational
In what physical stuff,
and how?
neuroscience
Addition: combine quantities into a total.
Order does not matter; adding nothing changes nothing.
Decimal columns,
carrying Binary place value Beads on an abacus Counting on
fingers
A child with a
pencil Brass gears Silicon circuits the same expansion
could be drawn under
any of the four
Figure 1.6 — Addition at three levels, drawn as the branching it is. One problem admits several algorithms,
and each algorithm admits several implementations, so fixing any one level leaves the others open.
What was missing, Marr proposed, is that fully understanding any information-processing
system takes answers to three different kinds of question, at three levels. Marr’s contribution was to
name them, sharpen them, and insist that all three are needed.
The first is the computational level. It asks what problem the system solves, and what anything
solving that problem must respect. Notice that this is not a question about the machinery. It is a
question about the task itself, considered on its own. Take basic addition. The computational
description of addition says what the operation has to do: combine quantities into a total, in a way
that does not depend on the order the parts arrive (three apples and then two is the same as two
and then three), and where adding nothing leaves the total unchanged. These are facts about
addition, true no matter who or what is doing the adding. If described this way, you have specified
what addition is at the computational level, without yet saying anything about how the sum actually
gets computed.
The second is the algorithmic level. It asks how the system solves the problem: by what
procedure, working over what representations. Here the choices multiply, because one and the
same computation can be carried out in many ways. To add, you might line the numbers up as
decimal digits and work column by column, carrying as you go, the way many learn to do so in
school. You might instead represent the numbers in binary, as a computer does, and follow a
different procedure that reaches the very same totals. You might slide beads on an abacus, count
on your fingers, or manipulate Roman numerals by rules that look nothing like the digit-column
method. Each of these is a distinct algorithm, and all of them compute addition. The algorithmic
level is where we say which one a given system actually uses.
1. Mind and Brain
Page 14 of 28
The third is the implementational level. It asks what physical stuff carries out the algorithm,
and how. The column-by-column method can be run by a child with a pencil, by the gears of a
mechanical adding machine, or by the circuits of an electronic calculator. Same problem, same
algorithm, entirely different matter. The implementational level describes that actual substrate:
neurons and their firing, silicon and its voltages, brass gears and their teeth.
The three levels are genuinely distinct, and it is worth seeing why, because the reason is
structural rather than a matter of taste. The relations between the levels are many-to-one, in both
directions (Figure 1.6). A single computational problem, such as addition, can be solved by many
different algorithms. And a single algorithm can be implemented in many different physical
materials. So fixing one level leaves the others open. Knowing the problem does not tell you the
procedure, and knowing the procedure does not tell you the hardware. That is exactly why an
answer at one level cannot stand in for an answer at another. The three are not three descriptions
of a single fact. They are answers to three different questions, and a full understanding needs all
three.
Marr added a further, stronger, and more controversial claim, which we will mention now and
examine later. He held that the levels are not merely distinct but ordered, with the computational
level coming first. You cannot understand how a system works, he argued, until you know what
problem it is solving. This means analysis should run from the top down. This top-down priority is
central to Marr’s own view, and it is the natural reading of his complaint that studying feathers and
muscles doesn’t add up to understanding flight. But it is a claim, not an obvious truth. It has been
challenged, and we will discuss that challenge in §1.2.5. For now we have what we came for. We
have three distinct levels, each answering a question the others cannot. The next step is to see why
these particular three levels make a good backbone for this course.
§1.2.3 — The Course Backbone: Levels, Disciplines, and the Evolution of the
Human Mind
We now have three distinct levels. What turns them from a piece of philosophy into the backbone of
this course? You may have already noticed that the three levels line up with the way brain and
cognitive science already divides itself into fields. That alignment is what lets a framework guide our
curriculum.
Look at how the questions sort out. The computational question, what problem is being solved,
and why, is the home turf of the disciplines that study why organisms are built the way they are.
Evolutionary biology, behavioral ecology, anthropology, and parts of psychology ask what a capacity
is for. The algorithmic question, by what procedure, over what representations, is the province of
cognitive psychology and of artificial intelligence. These fields both describe the steps a mind runs
abstracted away from the physical materials they are built from. The implementational question, in
what physical stuff, and how, belongs to neuroscience. So we have levels corresponding to three
broad families of academic disciplines, each asking a different one of Marr’s questions about the
same object. This is not a coincidence. The fields grew up around different questions in the first
place. This is why they so often talk past one another, and why laying the levels side by side lets
their findings be combined rather than ranked.
One of these alignments between levels and academic disciplines needs more than a label,
because it is what gives this course its particular shape. Return to the computational level, the
question of what problem a system solves and why. For a human-made object, the “why” is easy.
The why is explainable in terms of the purpose of its creator or designer. Ask why a calculator does
addition and the answer is that its makers wanted a tool for doing arithmetic. But a heart has no
designer. Neither does an eye, or a memory system, or the circuit that recognizes a face. So what
grounds the “why” for a living thing? The primary answer for most organisms is natural selection.
For a biological system, the computational question usually becomes a question about ancestry and
survival. What problem did this organism’s ancestors face, such that solving it helped them survive
1. Mind and Brain
Page 15 of 28
and reproduce, and pass on whatever machinery did the solving? The “why” of a biological trait is
an adaptive why. This is the module’s evolutionary perspective, and it makes evolution the spine of
the whole course rather than one week’s topic. Again and again in this course, asking at the
computational level what a mental capacity is for will mean asking what problem it solved when it
evolved.
A word of caution comes with this approach. An adaptive why is a hypothesis to be tested, not a
story to be assumed. Not every trait is an adaptation shaped for a purpose. Some are byproducts of
other traits, some are the residue of chance, some are accidents of history that stuck around. The
computational level asks about the adaptive problem a capacity solves. It does not license inventing
a flattering tale about a trait or property. Being open-minded about adaptive hypotheses, and
demanding evidence for them, is critical.
There is a further reason to be careful about adaptive stories, and for our own species it is the
largest of all: culture. Many of the problems a modern human mind solves were set neither by a
designer nor directly by natural selection. Reading, symbolic arithmetic, handling money, driving a
car: natural selection has not had nearly enough time to adapt the human species to solve these
problems. And yet, they are among the most demanding things we do. Their what for bottoms out in
a goal set by a human practice and handed down socially, not in ancestral survival and
reproduction. Culture completes a triad of sources for a computational goal, one to match each kind
of system we have discussed: a designer’s intent for an artifact, natural selection for a biological
trait, and a cultural practice for learned human skills (Figure 1.7).
A designer’s intent
A calculator does addition
because its makers wanted a tool
for arithmetic.
for an artifact
Natural selection
An eye recovers the world because
ancestors who saw it survived and
reproduced.
for a biological trait — the
adaptive why
A cultural practice
Reading and handling money are
goals set by a practice and
handed down socially.
for a learned human skill
A human mind draws on all three at once.
Figure 1.7 — Three sources of a computational goal, one for each kind of system — and a human mind
draws on all three at once.
Culture is a critical source for the human mind’s computational goals, but it does not float free of
biology. The very capacity to acquire knowledge through culture, our unusual flexibility in learning,
our language, our habit of teaching one another, these abilities are themselves among our strongest
biological adaptations. This means cultural goals rest on an evolved foundation rather than
replacing it. And cultural skills tend to recycle machinery that evolved for something else. Reading is
only a few thousand years old, far too recent for a dedicated brain system. But the ability to read
makes heavy use of circuitry that evolved for recognizing objects. Similarly, symbolic arithmetic is a
cultural practice that is built atop an approximate sense of number that we share with other animals.
One word deserves care in all this, because it names two different things. We say a biological trait
is adapted across generations, by selection acting on a lineage. And we say a skill is adapted within
a single lifetime, by a person learning and practicing. Both shape what a mind computes, on entirely
1. Mind and Brain
Page 16 of 28
different timescales and by entirely different means, and telling them apart is one of the difficult
tasks of the field of brain and cognitive science (Table 1.2).
Table 1.2 — One word, two processes. Both leave a mind better matched to a problem, and almost nothing
else about them is the same.
Adapted across generations Adapted within a lifetime
Timescale Many generations Seconds to years — one telling
can be enough
Mechanism Natural selection acting on a
lineage
Learning and practice in one
individual
What changes The inherited machinery a
species is built with
One individual’s skills and
knowledge
What is passed on Genes Nothing biologically — culture
may carry it socially
Example
An eye shaped over evolutionary
time to recover the world from
light
Reading, or doing arithmetic on
paper
One more relationship between the three levels and the themes of our course deserves to be
mentioned. The algorithmic level is where minds and machines become comparable in a disciplined
way. Because an algorithm abstracts away from the stuff that runs it, the same procedure can in
principle run on neurons or on silicon. That is exactly what makes a comparison between a brain
and a machine a scientific claim rather than a loose metaphor. The algorithmic level is about
comparing procedures, not materials. This is the connection to the multiple realizability issue we
discussed in §1.1. It is also why artificial intelligence research tends to sit at the algorithmic level in
Marr’s framework. And it is why the question of machine minds, which §1.1 posed and left open, will
keep returning as a question about shared algorithms.
Because Marr’s levels map onto disciplines this cleanly, they give the course its structure. Most
weeks ahead take up one capacity: perception, memory, decision-making, and examine it at all
three levels in turn. What problem did it evolve to solve, what algorithms and representations carry
out the solution, and what neural machinery implements it. You will see the same three-part scaffold
return again and again, with the discipline labels attached, until the habit of asking “which level is
this explanation working at?” becomes second nature. To see the scaffold in action before it
becomes routine, in the next section we will examine a single capacity at all three levels.
§1.2.4 — Vision at All Three Levels
The best way to understand the framework, and to see its value, is to watch it work on a real case.
We will use vision as our demonstration, since it was Marr’s own example. It is also a capacity we
understand unusually well at all three levels, so the framework can be shown working rather than
merely asserted.
Begin at the computational level, which asks what problem vision solves. The obvious answer is
that vision uses the information in light to sense the objects and events around us, so that we can
respond to them appropriately. That answer is not wrong, but it hides a real difficulty, one that
appears only when we look closely at what the eye actually has to work with.
1. Mind and Brain
Page 17 of 28
What actually reaches the eye is far less than the world itself. At the back of the eye sits the
retina. The retina is a thin sheet of light-sensitive cells. Light bouncing off the things around an
organism passes through the lens, and is focused onto the retina, much as a camera focuses an
image onto its sensor. What the retina registers is simply how much light, and of what color, falls on
each point across its surface. That surface has only two dimensions: up-and-down and side-to-side,
with no depth of its own. This means the raw input to vision is in effect a flat picture, a two-
dimensional array of light.
Yet a flat picture is not what the animal needs. What it presumably needs to know about is the
three-dimensional world that produced the image. What surfaces are present and where are they?
What objects are present? What is food, and what is a predator? One conceptualization of vision’s
task, then, is to recover that solid world from the flat pattern of light.
But the big problem with recovering the 3D world from a 2D image is that, for a given 2D image,
there are an infinite number of different 3D scenes that could have generated that image. So the 2D
image doesn’t tell you all you need to know about what is actually out there. Different 3D scenes,
from different distances and under different lighting conditions, could cast the very same pattern of
light on the retina. Recovering one three-dimensional world from that image is what is called an ill-
posed problem. The data alone cannot precisely specify which scene generated the image
(Figure 1.8).
How does the visual system solve this ill-posed problem? The visual system is able to determine
what 3D world most likely generated the image by making assumptions about how the world usually
is. These include assumptions such as that light tends to come from above, and that surfaces tend
to be continuous. These assumptions help the mind single out the most likely scene from the
endless candidates. Discovering that the visual system makes these assumptions would be very
difficult, if not impossible, if we were just studying the visual system’s neurons without also thinking
about the nature of the problem that it needs to solve.
One of those assumptions can be caught in the act (Figure 1.9). Every disc in the grid is shaded
identically. The only difference among them is that the middle row has been turned upside down.
Yet one set reads as raised bumps and the other as hollow dimples. Turn the page around, and the
two sets exchange appearances. Nothing about the ink changed, so the bumps and the dimples
were never on the page at all. They were supplied by a visual syst