News Human thought runs at just 10 bits per second, say Caltech scientists — that's why we are mostly single-taskers

[Admin]

Humans process thoughts at just 10 bits per second, according to a recent paper from Caltech researchers.

Human thought runs at just 10 bits per second, say Caltech scientists — that's why we are mostly single-taskers : Read more

 

[hotaru251]

10 bits per second?
Eat it NES!

Do wonder if its just thought is considered a "less important" task and its slowness is by design to keep more important data (i.e. senses that can tell you about world/dangers) from ever being slowed down as a survival thing.

 

[newtechldtech]

This is not Science. this is complete utter rubbish . no one "knows" how thoughts happens in the brain. Full Stop .

 

[Dementoss]

It's hard to understand how hey have come to the conclusion, that human thought processes are so slow. If our thought processes are really so slow, it would be interesting to know how they believe we can do things like, driving a car in a busy town centre whilst listening to the radio and talking talking to a passenger, riding a mountain bike at speed on a difficult trail, or learning to fly a plane, etc...

 

[acadia11]

Always been said we are serial processing pattern inference machines. Maybe, over billions (really 500 million or so since Cambrian expansion) of years we’ve derived an extremely complex digital compression algorithm and it’s far more efficient / optimized to use as little energy as possible. Consider we are creating “thinking machines” today but it takes an ungodly amount of energy to do so … if we learned and processed information like our AI inventions how would the human body power the brain? I’d trust nature on this one otherwise we’d probably need to eat candy by the tanker load?

 

[A Stoner]

If I actively try to think of an image in my mind it is going to give me something in billions of colors with nearly infinite pixels and I can summon that up in a fraction of a second. I can even think in moving images. I can control the image for what it is composed up of.

That all requires much more bandwidth than 10 bits per second.

Talking would be far more than 10 bits per second. A fast talker can bring forth hundreds of words per minute. Each word is many bits in size. I checked my calendar and it is not April 1.

 

[Notton]

I can kind of see it being true.
If you've looked into any sort of how the human brain works, you quickly realize it makes shortcuts everywhere.
We have handedness because it saves brain power (calories)
What we see with our eyes isn't actually what we see, and our brain fills a huge portion of the missing bits. (This is why we fall for visual tricks)
The brain doesn't process a large part of the functions, like reflexes, breathing, digestion, etc.
Humans are awful at math without writing it down on something.

I'd imagine we would have to consume a million calories per day to run our brains at full power.

 

[Alvar "Miles" Udell]

It makes sense from an evolutionary perspective, using the least amount of energy to accomplish a task. But there's also research into if the brain processes information in a quantum way such that it causes consciousness as well.

But as it relates to machines being able to stimulate sensory inputs, if it will only take a low bit rate to do it then it bodes well for more affordable, power efficient bionics.

 

[pjmelect]

Utter rubbish, I can easily read and understand what I am reading at 300 baud and although it is a strain I can read at 1200 baud without a problem which makes nonsense of the 10 bits per second.

 

[spongiemaster]

If anyone ever needs an example of I read the headline and not the article then went straight to the comment section to argue. This thread is it.

 

[adamXpeter]

I can even think in moving images. I can control the image for what it is composed up of.

That all requires much more bandwidth than 10 bits per second.

They weren't speaking about sensory information, but about the speed of thinking. Your vision misleads you, it is hard to see with the single lens we had. So the brain avoids looking through the eyes and prefers it's internal model of the environment. I am sure it happened to you that you did not find the keys of the car before leaving the house - the more stressed you were because of the potential of being late, the less you found the key. Because more CPU power went to counter the danger than remained to correct the physically distorted picture, the brain used more the inner model where the places for the keys were limited by rationality

 

[mitch074]

Utter rubbish, I can easily read and understand what I am reading at 300 baud and although it is a strain I can read at 1200 baud without a problem which makes nonsense of the 10 bits per second.

You couldn't at first. You needed to train your neural pathways to see the signs, interprete them and then make sense of them. Becoming conscious of them (the 'slow' part the paper is about) is much slower.
You can read a whole book's page in 20 seconds; no problem. Making complete sense of it in that time ? Less likely.

 

[newtechldtech]

scientists cant even explain where thoughts and ideas come from ... and how they appear from nowhere ... what 10bit per sec ? it is almost infinite per sec.

I challenge ANY scientist to explain how a new idea comes suddenly in human mind from NOTHING. they just "light" in the brain.

 

[palladin9479]

This is dumb, like real dumb. Our consciousness operates much faster then that, and more importantly our subconscious is where most "thought" happens. That part of our brain rocks at ridiculous speeds.

 

[bit_user]

This is not Science. this is complete utter rubbish . no one "knows" how thoughts happens in the brain. Full Stop .

Do you ever check the source links, to see what the article is talking about? It can often give more insight and answer some questions or uncertainty not sufficiently addressed by the author.

Here's from the paper's abstract (because the actual paper is behind a paywall):

"We review measurements spanning the better part of a century that involve all aspects of human cognition: perception, action, or—as in the example above—imagination. The general approach is to assess the range of possible actions that a person may execute in a given time. Along the way, one needs a clear criterion to distinguish the action from its noisy variations. This distinction of “signal” and “noise” is quantified by Shannon’s entropy and ultimately leads to an information rate, expressed in bits/s.

This information-theoretic approach allows us to compare the speed of processing across different mental tasks and processes, between different neural structures in the same brain, across different species, and between brains and machines."
​

So, one way to put it is that they found the decision rate to be equal to about 10 bits/s (using a very loose definition of "decision"). Obviously, they can only measure what's externally observable, so that's what it's talking about - externally observable decisions, across a wide variety of physical and mental tasks.

 

[bit_user]

it would be interesting to know how they believe we can do things like, driving a car in a busy town centre whilst listening to the radio and talking talking to a passenger, riding a mountain bike at speed on a difficult trail, or learning to fly a plane, etc...

If you actually look at the rate of decisions you make, while driving a car, riding a mountain bike, or flying a plane, it's not very high. If you were having to make many conscious decisions every single second of these activities, they'd be too exhausting to keep up for very long.

As for radio listening and conversations, these are just using idle cycles in your brain. When not much is happening and your autonomic nervous system can pretty much handle it, you can focus on them. However, when you reach a busy intersection, with a lot of cars or people doing different things, you'll find that you stop talking/listening and focus your attention on the task of driving.

 

[bit_user]

If I actively try to think of an image in my mind it is going to give me something in billions of colors with nearly infinite pixels and I can summon that up in a fraction of a second. I can even think in moving images. I can control the image for what it is composed up of.

That all requires much more bandwidth than 10 bits per second.

I'm not saying the 10 bits/s metric applies to this activity, but I do think the task of mental imaging is deceptive.

You believe you have the whole image in your mind at once, just like you believe you see objects in the world all at once. However, the photoreceptors in your retina are massively concentrated in the center of your vision and what's really happening when you look at an object is that you're scanning it with your eyeballs and your brain is noticing things about the object and building a representation of it over time. It also knows what certain objects look like, and can fill in other details from memory. It can't focus on all aspects of the object, all at once.

In a similar vein, I think mental imaging works by making a set of decisions about the object and building up the image as you mentally examine it. You get the impression of having the entire image there at once, because whenever you mentally inspect one part or aspect of it, it's there. However, what you can't know is how much your brain is generating that imagery just-in-time, basically as you're mentally focusing on it.

In a similar way, I think this explains the act of dreaming, because your brain doesn't need to build an entire world. It just needs to generate the parts you're thinking about, as you think about them. Cognitively, this is a much simpler task.

Each word is many bits in size.

Here's where I'm not sure about their precise terminology. However, the way to think about the information content of a word is probably more like if you built a Huffman table of all the words in your lexicon. The linguistically simplest and most frequently-used words would have the lowest information content.

 

[bit_user]

Utter rubbish, I can easily read and understand what I am reading at 300 baud and although it is a strain I can read at 1200 baud without a problem which makes nonsense of the 10 bits per second.

But the brain doesn't use ASCII. The authors acknowledge that senses like your eyes are taking in "gigabits" of information per second. But, this quickly gets distilled down by your visual cortex, as it recognizes words and phrases, which can be represented much more efficiently.

 

[kjfatl]

10 bits per second?
Eat it NES!

Do wonder if its just thought is considered a "less important" task and its slowness is by design to keep more important data (i.e. senses that can tell you about world/dangers) from ever being slowed down as a survival thing.

But the brain doesn't use ASCII. The authors acknowledge that senses like your eyes are taking in "gigabits" of information per second. But, this quickly gets distilled down by your visual cortex, as it recognizes words and phrases, which can be represented much more efficiently.

10 bits per second is probably accurate. If so, it is poorly defined in the article.
I suspect it means 10 conclusions per second. Behind each bit might be 500 billion neural actions done in parallel using about 20W.

 

[A Stoner]

They weren't speaking about sensory information, but about the speed of thinking. Your vision misleads you, it is hard to see with the single lens we had. So the brain avoids looking through the eyes and prefers it's internal model of the environment. I am sure it happened to you that you did not find the keys of the car before leaving the house - the more stressed you were because of the potential of being late, the less you found the key. Because more CPU power went to counter the danger than remained to correct the physically distorted picture, the brain used more the inner model where the places for the keys were limited by rationality

Bit, 0 or 1. Not much information there. 2, 4, 8, 16, 32, 64, 128, 256, 512, 1024...

English is made up of about 170,000 total words, and most people know 20,000 to 30,000 of those words. Effectively, at 10 bits per second it would take several seconds to simply parse out a single word you want. And while, yes, I have a condition that causes me to lose words, for the most part I can clear many words every second consistently. That is far greater than 10 bits per second.

Now, if they would argue that we are processing at 10 64-bit bytes per second, I could possible believe something like that. Maybe even as low as 32-bit bytes. But simply 10 bits, eff no. We think and actively process far more than 10 bits per second actively and consciously.

 

[kjfatl]

Bit, 0 or 1. Not much information there. 2, 4, 8, 16, 32, 64, 128, 256, 512, 1024...

English is made up of about 170,000 total words, and most people know 20,000 to 30,000 of those words. Effectively, at 10 bits per second it would take several seconds to simply parse out a single word you want. And while, yes, I have a condition that causes me to lose words, for the most part I can clear many words every second consistently. That is far greater than 10 bits per second.

Now, if they would argue that we are processing at 10 64-bit bytes per second, I could possible believe something like that. Maybe even as low as 32-bit bytes. But simply 10 bits, eff no. We think and actively process far more than 10 bits per second actively and consciously.

Don't think of the brain as a von Neuman machine. Think of it as a slow, massively parallel computer with 85 billion 'simple' parallel processors called neurons. The results of all of your thoughts are fed into a 'table'. In this table representing the current state of your mind, 10 bits change per thought at a rate of 10 thoughts per second. These 10 bits are fed back into your 85billion neurons with the next thought being processed 100ms seconds later. Of course the input of all you senses is also being fed into your neurons as well and there is certainly latency involved particularly in the visual cortex.

Using your logic above, the brain has approximately 1000 neural connections per neuron allowing it to make 60 trillion decisions every 100ms to get a single thought, or 600 trillion decisions per second.
All of this happens while consuming about 20 watts.

 

[bit_user]

Dear Caltech researchers, please develop a 10-bit NPU to replace your brains so we can laugh at you cripples

That's not remotely what they're saying.

If you look at the tokens/s generated by cutting-edge LLMs, it's only a handful, yet the models themselves have up to hundreds of billions of parameters. It's this processing rate that they're focusing on.

 

[Steve Nord_]

This is not Science. this is complete utter rubbish . no one "knows" how thoughts happens in the brain. Full Stop .

Check out Allen Brain Center and some threads of Woods Hole Research Inst. to see how much is visible for the looking. There are a few jarring bits like 'We can cure your mind, let us pick a sample out' in treatment.

 

[Kamen Rider Blade]

It's hard to understand how hey have come to the conclusion, that human thought processes are so slow. If our thought processes are really so slow, it would be interesting to know how they believe we can do things like, driving a car in a busy town centre whilst listening to the radio and talking talking to a passenger, riding a mountain bike at speed on a difficult trail, or learning to fly a plane, etc...

Most likely through muscle memory and training.
There's a reason why humans need to train to do complex tasks.
Repetition builds muscle memory, and it helps to make it easier to perform said tasks at a subconscious level.

 

[JRStern]

That's not remotely what they're saying.

If you look at the tokens/s generated by cutting-edge LLMs, it's only a handful, yet the models themselves have up to hundreds of billions of parameters. It's this processing rate that they're focusing on.

Tokens per second is not a measure of activity, just ask OpenAI, they'll tell you how many megawatt/seconds they used to produce a few tokens, a few dozen characters, not even counting amortizing the immense training time.