What's the difference between a brain and a computer?

What's the difference between a brain and a computer?

Show Video

My grandmother was a computer, and I don’t  mean there was a keypad on her chest.   My grandmother calculated orbits of stars, with  logarithmic tables and a slide ruler. But in   which sense are brains similar to the devices we  currently call computers, and in which sense not?   What’s the difference between what they can do?  And is Roger Penrose right in saying that Gödel’s   theorem tells us human thought can’t just be  computation? That’s what we’ll talk about today. Before we talk about computers, some quick news.  We now have a newsletter for science without the   gobbledygook. You can sign up for it on my  website sabinehossenfelder dot com. It goes  

out once per week, is completely free, and you  can unsubscribe at any time. Many thanks to our   supporters on Patreon, who make all this possible,  especially those of you in tier four and higher.   So now let’s talk about how  brains are like computers. If you have five apples and I give you  two, how many apples do you have in total?   Seven. That’s right. You just did a computation.  Does that mean your brain is a computer? Well,   that depends on what you mean by  “computer” but it does mean that I   have two fewer apples than I did before.  Which I am starting to regret. Because I   could really go for an apple right now.  Could you give me one of them back?

So whether your brain is a computer depends  on what you mean by “computer”. A first   attempt at answering the question may be to say  a computer is something that does a computation,   and a computation, according to Google is  “the action of mathematical calculation”.   So in that sense the human brain is a computer. But if you ask Google what a computer is, it  says it’s “an electronic device for storing   and processing data, typically in binary  form, according to instructions given to it   in a variable program”. The definition  on Wikipedia is pretty much the same   and I think this indeed captures what most of  us mean by “computer”. It’s those things we   carry around to brush up selfies, but that  can also be used for, well, calculations.

Let’s look at this definition again in more  detail. It’s an electronic device. It stores   and processes data. The data are typically in  binary form. And you can give it instructions   in a variable program. Now the second and  last points, storing and processing data,   and that you can give it instructions,  also apply to the human brain.  

This leaves the two properties: it’s an electronic  device and it typically uses binary data,   which makes a computer different to the  human brain. So let’s look at these two. That an electronic computer is “digital” just  means that it works with discrete data, so data   whose values are separated by steps, commonly  in a binary basis. The neurons in the brain,   on the contra-ry, behave very differently.  Here’s a picture of a nerve ending. In orange   and blue you see the parts of the synapse that  release molecules called “neurotransmitters”.   Neurotransmitters encode differ-ent signals,  and neurons respond to those signals gradually   and in many different ways. So a neuron is not  like a binary switch that’s either on or off. But maybe this isn’t a very important difference.  For one thing, you can simulate a gradual response  

to input on a binary computer just by  giving weights to variables. Indeed,   there’s an entire branch of mathematics  for reasoning with such inputs.   It’s called fuzzy logic and it’s the best logic to  pet of all the logic. Trust me, I’m a physicist. Neural networks which are used for artificial  intelligence use a similar idea by giving   weights to nodes and sometimes also the links  of the network. Of course these algorithms   still use a physical basis that is ultimately  discrete and digital in binary. It’s just that  

on that binary basis you can mimic the gradual  behavior of neurons very well. This already shows   that saying that a computer is digital whereas  neurons aren’t may not be all that relevant. Another reason this isn’t a particularly  strong distinction is that digital computers   aren’t the only computers that exist. Besides  digital computers there are analog computers   which work with con-tinuous data, often in  electric, mechanical, or even hydraulic form.   An example is the slide  ruler that my grandma used.  

But you can also use currents, voltages and  resistors to multiply numbers using Ohm’s law. Analog computers are currently having somewhat  of a comeback, and it’s not because millennials   want to take selfies with their record players.  It’s because you can use analog computers for   matrix multiplications in neural networks. In an  entirely digital neural network, a lot of energy   is wasted in storing and accessing memory, and  that can be bypassed by coding the multiplication   directly into an analog element. But analog  computers are only used for rather special cases   exactly because you need to find a physical  system that does the computation for you. Is the brain analog or digital? That’s  a difficult question. On the one hand  

you could say that the brain works with  continuous currents in a continuous space,   so that’s analog. On the other hand thresholds  effects can turn on and off suddenly   and basically make continuous input discrete.  And the currents in the brain are ultimately   subject of quantum mechanics, so  maybe they’re partly dis-crete. But your brain is not a good place  for serious quantum computing.   For one thing, that’s because it’s  too busy trying to remember how many   seasons of Doctor Who there are just in  case anyone stops you on the street and   asks. But more importantly it’s because quantum  effects get destroyed too easily. They don’t  

survive in warm and wiggly environments. It  is possible that some neurological processes   require quantum effects, but just how much is  currently unclear, I’ll come back to this later. Personally I would say that  the distinction that the brain   isn’t digital whereas typical  computers that we currently use   are, isn’t particularly meaningful. The reason  we currently mostly use digital com-puters is   because the discrete data prevent errors and the  working of the machines is highly repro-ducible. Saying that a computer is an electronic device  whereas the brain isn’t, seems to me likewise a   dis-tinction that we make in every-day language,  alright, but that isn’t operationally relevant.   For one thing, the brain also uses  electric signals, but more importantly,   I think when we wonder what’s the difference  between a brain and a computer we really wonder   about what they can do and how they do it, not  about what they’re made of or how they are made.

So let us therefore look a little closer at  what brains and computers do and how they do it,   starting with the latter: What’s the difference  between how computers and brains do their thing? Computers outperform humans in many tasks,  for example just in doing calculations.   This is why my grandmother used  those tables and slide-rulers.   We can do calculations if we have to, but  it takes a long time and it’s tedious and   it’s pretty clear that human brains aren’t all  that great at mul-tiplying 20 digit numbers. But hey, we did manage to build machines  that can do these calculations for us!   And along the way we discovered electricity  and semi-conductors and programming and so on.   So in some sense, you could say, we  actually did learn to do those calculations.   Just not with our own brains, because those are  tired from memorizing facts about Doctor Who.  

But in case you are good at multiplying  20 digit numbers, you should totally bring   that up at dinner parties. That way, you’ll  finally will have something to talk about. This example captures the key difference between  computers and human brains. The human brain took   a long time to evolve. Natural selection has given  us a multi-tasking machine for solving prob-lems,   a machine that’s really good in adapting  to new situations with new problems.   Present-day computers, on the contrary, are  built for very specific purposes and that’s   what they’re good at. Even neural nets haven’t  changed all that much about this specialization. Don’t get me wrong, I think artificial  intelligence is really interesting.  

There’s a lot we can do with it, and  we’ve only just scratched the surface.   Maybe one day it’ll actually be intelligent.  But it doesn’t work like the human brain. This is for several reasons. One reason is  what we already mentioned above, that in   the human brain the neural structure  is physical whereas in a neural net   it’s software coded on another physical basis.

But this might change soon. There are some  companies which are producing computer chips   similar to neurons. The devices made of  them are called “neuromorphic computers”.   These chips have “neurons” that fire  independently, so they are not synchronized   by a clock, like in normal processors. An example  of this technology is Intel’s Loihi 2 which has   one million “neurons” interconnected via  120 million synapses. So maybe soon we’ll   have computers with a physical basis similar to  brains. Maybe I’ll finally be able to switch mine  

for one that hasn’t forgotten why it went  to the kitchen by the time it gets there. Another difference which may  soon fade away is memory storage.   At present, memory storage works very  differently for computers and brains.   In computers, memories are stored in specific  places, for example your hard drive, where   electronic voltages change the magnetization  of small units called memory cells between two   different states. You can then read it out again  or override it, if you get tired of Kate Bush.

But in the brain, memories aren’t stored in  just one place, and maybe not in places at   all. Just exactly how we remember things is still  subject of much research. But we know for example   that motor memories like riding a bike uses brain  regions called the basal ganglia and cerebellum.   Short-term working memory, on the other  hand, heavily uses the prefrontal cortex.   Then again, autobiographical memories  from specific events in our lives,   use the hippocampus and can, over the course  of time, be transferred to the neocortex. As you see memory storage in the brain  is extremely complex and differentiated,   which is probably why mine sometimes misplace the  information about why I went into the kitchen.  

And not only are there many different types of  memory, it’s also that neurons both process and   store information, whereas computers  use different hardware for both. However, on this account too, researchers are  trying to make computers more similar to brains.   For example, researchers from the University  of California in San Diego are working in   something called memcomputers, which combines data  processing and memory storage in the same chip. Maybe more importantly, the human brain has  much more structure than the computers we   currently use. It has areas which specialize in  specific functions. For example, the so called  

Broca's area in the frontal lobe specializes  in language processing and speech production;   the hypothalamus controls, among other things,  body temperature, hunger and the circadian rhythm.   We are also born with certain types of knowledge  already, for example a fear of dangerous animals   like spiders, snakes, or circus clowns. We  also have brain circuits for stereo vision. If   your eyes work correctly, your brain should be  able to produce 3-d information automatically,   it’s not like you have to first  calculate it and then program your brain. Another example of pre-coded knowledge  is a basic understanding of natural laws.  

Even infants understand, for example, that  objects don’t normally just disappear.   We could maybe say it’s a notion of basic  locality. We’re born with it. And we also   intuitively understand that things which  move will take some time to come to a halt.  

The heavier they are, the longer it will take.  So, basically Newton’s laws. They’re hardwired.   The reason for this is probably that it benefits  survival if infants don’t have to learn literally   everything from scratch. I was upset to learn,  though, that infants aren’t born knowing Gödel’s   theorem. I want to talk to them about it,  and I think nature needs to work on this. That some of our knowledge is pre-coded  into structure is probably also partly the   reason why brains are vastly more energy  efficient than today’s supercomputers.   The human brain consumes on the average 20 Watts   whereas a supercomputer typically consumes  a million times as much, sometimes more. For example, Frontier, hosted at the  Oak Ridge Leadership Computing Facility   and currently the fastest supercomputer in the  world consumes 21MWatt on average and 29MW at peak   performance. To run the thing, they had to build  a new power line and a cooling system that pumps  

around 6000 gallons of water. For those of you  who don’t know what a gallon is, that’s a lot of   water. The US department of energy is currently  building a new supercomputer, Aurora, which is   expected to become the world’s fastest computer  by the end of the year. It will need about 60MW. Again the reason that the human brain is  so much more efficient is almost certainly   natural selection, because saving energy  benefits survival. Which is also what I   tell my kids when they forget to turn  the lights off when leaving a room.

Another item we can add to the list of differences  is that the brain adapts and repairs itself,   at least to some extent. This is why,  if you think about it, brains are much   more durable than computers. Brains work  reasonably well for 80 years on average,   sometimes as long as 120 years. No existing  computer would last remotely as long.   One particularly mind blowing case (no pun  intended) is that of Carlos Rodriguez, who had   a bad car accident when he was 14. He had stolen  the car, was on drugs, and crashed head first.  

Here he is in his own words. God damn that's  crazy though. Boys come on, that’s always no good,   drinking and driving, or drug, drug,  drugness and driving, no good kids. Not only did he survive, he is in reasonably  good health. Your computer is less likely to  

survive a crash than you, even if it remembered  to wear its seatbelt. Sometimes it just takes a   single circuit to fail and it’ll become useless.  Supercomputing clusters need to be constantly   repaired and maintained. A typical supercomputer  cluster has more than a hundred maintenance stops   a year and requires a staff of several  hundred people. Just to keep working.

To name a final difference between the ways  that brains and computers currently work:   brains are still much better at parallel  processing. The brain has about 80 billion   neurons, and each of them can process more than  one thing at a time. Even for so-called massively   parallel supercomputers these numbers are still  science fiction. The current record for parallel   processing is the Chinese supercomputer Sunway  TaihuLight. It has 40,960 processing modules,  

each with 260 processor cores, which means a total  of 10,649,600 processor cores! That’s of course   very impressive, but still many orders of  magnitude from the 80 billion that your brain   has. And maybe it would have 90 billion if you  stopped wasting all your time watching Doctor Who. So those are some key differences between  how brains and computers do things,   now let us talk about the  remaining point, what they can do. Current computers, as we’ve seen,  represent everything in bits,   but not everything we know can be represented  this way. It’s impossible, for example,   to write down the number pi or any other  irrational number in a sequence of bits.   This means that not even the best  supercomputer in the world can compute   the area of a circle of radius 1, exactly, it  can only approximate it. If we wanted to get pi  

exactly, it would take an infinite amount of  time, like me trying to properly speak English.   Fun fact: The current record for calculating  digits of pi is 62.8 trillion digits. But even though we can’t write  down all the digits of pi,   we can work with pi. We do this all the time,   though, just among us, it isn’t all that uncommon  for theoretical physicists to set pi equal to 1.

In any case, we can deal with pi as  an abstract transcendental number,   whereas computers are constrained  to finitely many digits.   So this looks like the human brain  can do something that computers can’t. However, this would be jumping to conclusions.   The human brain can’t hold all the digits of  pi any more than a computer. We just deal with  

pi as a mathematical definition with certain  properties. And computers can do the same.   With suitable software they are capable  of abstract reasoning just like we are.   If you ask your computer software if pi is a  rational number it’ll hopefully say no. Unless   it’s kidding in which case maybe you can think  of more interesting conversation to have with it.

This brings us to an argument that Penrose has  made, that human thought can’t be described   by any computer algorithm. Penrose’s argument is  basically this. Gödel showed that any sufficiently   complex set of mathematical axioms can be  used to construct statements which are true,   but their truth is unprovable within that system  of axioms. The fact that we can see the truth of   any Gödel sentence, by virtue of Gödel’s theorem,  tells us that no algorithm can beat human thought.

Now, if you look at all that we  know about classical mechanics,   then you can capture this  very well in an algorithm.   Therefore, Penrose says, quantum mechanics is the  key ingredient for human consciousness. It’s not   that he says consciousness affects quantum  processes. It’s rather the other way round,   quantum processes create consciousness.  According to Penrose, at least. But does this argument about  Gödel’s theorem actually work?   Think back to what I said earlier, computers  are perfectly capable of abstract reasoning   if programmed suitably. Indeed, Gödel’s theorem  itself has been proved algorithmically by a  

computer. So I think it’s fair to say that  computers understand Gödel’s theorem as much   or as little as we do. You can start  worrying if they understand it better. This leaves open the question of course   whether a computer would ever have been able  to come up with Gödel’s proof to begin with.   The computer that proved Gödel’s  theorem was basically told what to do.   Gödel wasn’t. Tim Palmer has argued that indeed  this is where quantum mechanics becomes relevant.

By the way, I explain Penrose’s argument about  Gödel’s theorem and consciousness in more detail   in my new book existential physics. The book also  has interviews with Roger Penrose and Tim Palmer. So let’s wrap up. Current computers still  differ from brains in a number of ways.   Notably it’s that the brain is a highly  efficient multi-purpose apparatus whereas,   in comparison, computers are special  purpose machines. The hardware of computers   is currently very different from neurons in  the brain, memory storage works differently,   and the brain is still much better at parallel  processing, but current technological developments   will soon allow building computers that are  more similar to brains in these regards. When it comes to the question if there’s anything  that brains can do which computers will not one   day also be able to do, the answer is  that we don’t know. And the reason is,  

once again, that we don’t really  understand quantum mechanics. Yet another reason humans are different from  computers is that we’re social learners.   Most of what we learn, we learn from others.  And while you can learn some thing on YouTube,   most of the time you really need someone  to talk to, especially if you’re not even   sure what your questions are in the first place.  Today’s sponsor, Wyzant will help you with that.

Their business idea is really simple: connect  people who want to learn with people who want   to teach. Their platform makes it easy to  find tutors and schedule sessions with them,   online or in real life. They also take care  of the financial part of the transaction,   so you don’t have to worry about that. Wyzant has  lots of tutors for physics and related topics,   but also pretty much everything else you might  want to learn, whether that’s Japanese or cooking.  To me, knowledge sharing is really the  best thing that’s come out of the internet,   and Wyzant is such a great example for this.  If you’re curious about Wyzant and just want   to experiment with the platform, you can  get 25 dollars off your first lesson when   you click the link in the description box! Thanks for watching, see you next week.

2022-08-02 07:32

Show Video

Other news