Back to Categories
Science20 min read
A Thousand Brains
by Jeff Hawkins
A New Theory of Intelligence
Published: June 9, 2023
4.5 (153 ratings)
Book Summary
This is a comprehensive summary of “A Thousand Brains” by Jeff Hawkins. The book explores a new theory of intelligence.
what’s in it for me? blow your mind with a stunning new theory on how the brain operates.#
Introduction
jeff hawkins' a thousand brains – a new theory of intelligence ever look at yourself in the mirror and wonder, what's going on in there?
it's a rhetorical question, of course you have.
the brain is one of our greatest mysteries, not just of science, but of humanity itself.
it has stumped us for millennia.
not that we haven't tried to figure out its inner workings.
neuroscientists have accumulated plenty of weird and wonderful observations about the brain.
but still, the fundamental question remains.
how do a bunch of little individual cells, called neurons, combine to create intelligence, creativity, and the whole theatre of consciousness itself?
enter the thousand brains theory.
it's a sprawling, majestic theory.
one that biologist richard dawkins says is so exhilarating, so stimulating, it'll turn your mind into a whirling maelstrom that'll keep you from sleeping at night.
so strap in, because in this chapter to jeff hawkins' a thousand brains, we're going to attempt to explain this theory – heaven help us – in about 15 minutes.
mysteries of the neocortex#
mysteries of the neocortex let's begin by visualising the structure of the brain.
first, imagine making a pot of spaghetti.
now picture cutting each piece of spaghetti into lengths a tenth of an inch long – that's 2.5mm – and holding one of these little guys upright, like a tiny roman column.
now, virtually stick these miniature spaghetti pillars side by side until you make a sheet of 150,000 pieces, about the size of a dinner cloth.
are you with us so far?
good.
so now, imagine that every single thing you know or could know about the world – everything you've ever thought, seen, heard or imagined in your entire life – is inside that sheet.
this symbolises your neocortex, and it's a wrinkly, folded piece of brain matter which takes up about 70% of the space inside your skull.
it's called neocortex because it's believed to have evolved relatively recently.
it wraps around brain regions that are older, like the limbic system – the so-called reptile brain.
the little spaghetti pieces are what hawkins calls cortical columns.
these columns aren't visible to the naked eye.
the cortex just looks like one big, crinkly sheet, but they're there if you look under a microscope.
they're patterns of how neurons connect – tiny, column-like structures of neural wiring.
the neocortex lets you see, hear, touch, talk and think.
it lets you learn languages, do maths, paint watercolours and ponder philosophy.
but here's the puzzle.
despite governing all these totally different functions, your neocortex looks pretty much the same everywhere.
strange, right?
and not only that, it closely resembles the neocortex of other mammals – animals who can't speak languages, solve rubik's cubes or learn quantum physics.
so how is this possible?
how can this one type of brain tissue – this one repeating structure that is the cortical column – do so many different things?
that's the question we're going to answer.
the brain is a prediction machine#
the brain is a prediction machine.
picture a brain in a vat.
just a brain, connected to nothing, lying there in total darkness.
now hook this brain up to some sort of sensory input – a visual feed from a camera somewhere, with a signal delivered by little spikes of neuron activation.
the brain takes these signals, incomprehensible at first, and begins detecting patterns.
and then patterns of patterns.
soon enough, it starts to anticipate what will come next, modelling and predicting the video input the same way supercomputers run models that forecast the weather.
every time it gets a prediction wrong, the brain updates this model a little bit, refining it to yield better predictions.
that is, fewer surprises about what images will come next.
now connect this brain to a couple of hands.
with hands, it can learn about the world in a new way – by manipulating it.
the brain vat holds an unfamiliar object – a stapler.
it rotates this thing around, looking at it from all angles.
it presses down on the thing, and a staple pops out.
it pulls the thing open at the hinge and sees a hundred staples inside, all lined up in a neat row.
now the brain isn't just passively waiting for data, it's actively creating it.
and the whole time it's using this new data to create a better, more refined, more accurate model of its world.
as humans – heck, as organisms – we model to survive.
models are essential because they give us predictions, and predictions give us control.
a prediction might be, if i reach out, grab this doorknob, turn it clockwise and pull on it, the door will open for me.
or it might be, if i'm nice to this person in front of me, they'll smile and share their fries with me.
intelligence is the ability to generate accurate models of the world, one little piece at a time, and use the models to get stuff.
and here's the thing.
you are that brain in a vat.
the world that you experience is a simulation, an hallucination that's running inside your neocortex as it models the weird world outside.
your neocortex is essentially a prediction machine whose function is to generate models that work, so that you can shape your environment to let you survive and pass on your genes.
but how do we build these models?
how does this prediction engine work?
and how does this insight help us understand the mysterious structure of the neocortex?
our 150,000 cortical columns.
your split personality#
your split personality.
to understand hawkins' view, we need to first look at some older assumptions of neuroscience.
assumptions we're going to overturn.
the traditional view describes the brain as having distinct functional modules.
for instance, it says we have a dedicated sensory cortex that processes input from the senses, and a motor cortex that controls motion.
in this view, raw sensory data comes in from things like our eyes, ears and skin in the form of simple features that get combined into complex ones.
nerve signals from rods and cones in our eyes are first processed in terms of simple shapes, like straight edges, curved edges, little blobs of colour and so on.
then these primitive features are combined, like building blocks, to make larger and more complex mental structures and objects.
chairs, tables and other objects as we know them.
after that, some other part of the cortex processes this sensory information, thinks about it, and makes a decision about what action to take.
sending instructions to the motor cortex, a brain centre which controls our muscles.
but, hawkins says, this traditional view is outdated.
neuroscientists no longer conceive of the brain as being divided up into these neat functional units.
instead, they found something much weirder.
it turns out that each of these cortical columns – our thousands of little spaghetti pieces – each have their own connections to sensory inputs, as well as their own motor connections.
for example, you have cortical columns that are hooked up to the retina and process visual input.
but these same columns are also connected to the muscles around the eye – the ones the eye uses to flit around and scan the visual field.
it's as if each of these virtual columns – and remember, you have 150,000 of them – is like a little brain all by itself.
a little brain in a vat.
a little prediction engine containing its own world model, with its own sensory and motor attachments.
in other words, we have thousands of tiny brains, each sensing the world, modelling it and acting on it.
it sounds bizarre, but stay with us, because it turns out this theory helps explain some long-standing mysteries of the brain.
let's look at three of them.
first up, there's the fact that we can see the same repeating structure throughout the neocortex, despite its incredible variety of functions – sight, sound, touch, language, abstract thought, etc.
second, experiments with young animals have shown that optic and auditory nerves can be swapped, switching which parts of the brain the eyes and ears are hooked up to.
and the animal will still grow up to see and hear relatively well.
third, the brain is remarkably adaptable and resilient.
if someone has a traumatic brain injury that damages part of the cortex, the brain simply rewires the remaining cortical tissue around the damage to restore function.
the explanation for these mysteries is this.
the neocortex isn't fundamentally divided up by uniquely specialised functional units.
rather, every part is essentially the same.
it has the same kind of circuit diagram.
the parts differ not in terms of their fundamental structure, but in what they're hooked up to.
the brain is astoundingly complex.
it has to be, because our reality is too.
we're not just dealing with staplers and staples.
reality is love, hate, trees, flowers, failure, shakespeare, democracy, music, the higgs boson, you name it.
but it seems that evolution built up this complexity by implementing a much simpler learning algorithm.
it created a single circuit that our dna builds many copies of.
the circuit found in each cortical column.
this has huge implications for understanding the brain.
if we can figure out how just one cortical column – a single little spaghetti slice – works, we can understand the mind and everything it's capable of.
crazy, right?
so what's going on in there, inside these 150,000 mini-yous?
hawkins and his lab think they have the answer.
thinking in motion#
thinking in motion the fundamental unit of cognition, the essential building block that all intelligence and perception is based on, is this.
the prediction of sensory input after motor movement.
that is, we do something, and then we see, or hear, or feel, or smell, or taste, something.
and this something is either expected or unexpected.
at its core, cognition is anticipating – and learning to better anticipate – which outputs yield which inputs.
this process is what's happening inside every single cortical column, each with a different slice of reality.
when we see, hear and touch objects in the world, we take sequences of information about movement and sensation and relate them together.
an object like a stapler is actually stored in hundreds of cortical columns at once.
so then, what separates one of our mini-brains from another?
well, each cortical column has its own reference frame.
what's a reference frame, you ask?
think of a geographical map.
a map is a kind of model.
maps have lines and colours and shapes that correspond to features of the particular territory they represent.
but all of this stuff, this detail, happens within a grid.
the lines that represent longitude and latitude.
longitude and latitude are the map's reference frame.
the frame of reference within which everything in the model is specified.
in the same way, you can think of every cortical column as having a coordinate system.
but in this case, it's based on different slices of sensory inputs – say, the sensations on the tip of one finger – and their relationship of these sensations to one another and to little slices of action.
if somebody blindfolded you and handed you a familiar object, like a coffee cup, you could probably work out what it was by running one finger around it.
you'd do that by tracking the shape of finger sensations in space – that is, using a spatial frame of reference to analyse the motion of your finger relative to the cup.
hawkins believes that this basic ability evolved from a mechanism called grid cells, which allowed simple organisms to navigate a map.
evolution then repurposed this same mapping mechanism to let us store models of objects.
as you ran your finger over that coffee cup, your brain would be making predictions about what your finger should feel next.
if you felt something unexpected, like a crack in the porcelain, your brain would take note and update its sensory inputs.
the idea of a grid is that you can use a map to map objects.
if you felt something unexpected, like a crack in the porcelain, your brain would take note and update its model.
your fingertip's model of a coffee cup is like a region on a map – a map of sensations that it can feel across space.
almost like the map an ant could use if it were crawling over the cup's surface.
on that note, how are your cortical columns doing?
has your spaghetti overcooked and turned to mush?
no?
good.
then let's try and finish the job by looking at two final pieces of the puzzle.
democracy in the brain#
as we know, our brains deal with more than just staplers and coffee cups.
so what about higher level cognition – things like language and maths?
remember the ingredients we're working with here – movement, sensations, models and predictions.
we've also said that these prediction models work by using cellular machinery that evolved to help us navigate through space.
so now close your eyes – if you can safely do that – and imagine walking through your home.
you can probably picture it pretty well – your front door, how it opens to the hallway, then your kitchen.
what's your brain doing while you imagine?
it's rehearsing, based on its stored model, which sensations it would anticipate in navigating that space.
you're basically walking through a simulation your brain has built of your home.
the same as if you were to physically walk through it.
and guess what?
any line of thinking, reasoning or imagining is basically like this.
it's a form of navigation and traversal.
like taking a pleasant hike through an abstract space of concepts and features instead of a physical, literal environment.
everything we experience – from the mona lisa to maths to justice, fomo and reggaeton – is part of a model constructed through the same architecture of overlapping reference frames.
abstract shapes which are stored, used and constantly revised in thousands and thousands of little world models.
just try and wrap your neocortex around that.
and now for the last puzzle piece.
how do we get these 150,000 little slices, these models of reality, to combine into one big model?
how do we get this so-called society of mind to come together and cooperate?
by voting, of course.
it turns out that each cortical column has certain neurons which are connected over longer distances, veering off from their home column and reaching out to other columns and regions of the brain.
hawkins calls these voting neurons.
voting neurons combine the results of different cortical columns, process them and converge to a uniform result.
when you feel a coffee cup and recognise it, it's because a bunch of cortical columns think, oh hey, a coffee cup, and outvote any other competing interpretations.
hawkins believes that your conscious experience, your life as you experience it, is the process of these voting neurons tallying their votes.
now that's a functional democracy.
so there you have it.
final summary#
Conclusion
our chapter to a thousand brains by jeff hawkins, which outlines the thousand brains theory.
the mind-bending idea that keeps scientists like richard dawkins awake at night.
the neocortex, the brain's thinking engine, is an assembly of repeating structures of neurons known as cortical columns, which are little mini-brains.
they're each based on the same type of neural circuit, a prediction engine with the ability to model a piece of sensory input and act on it.
each mini-brain is organised around a particular reference frame, which is like a coordinate system that maps a slice of reality.
thinking and reasoning are essentially navigation through this simulated landscape of abstract thoughts and concepts.
and to create unified perceptions and actions, our cortical columns use a process of convergence akin to voting.
simple, right?
anyway, thanks so much for listening.
if you can, please leave us a rating, as we always appreciate your feedback.
hope to catch you again in the next chapter.
You Might Also Like
Discover more book summaries in the same category or by the same author.