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

[TJ Hooker]

Non-paywalled link for those interested: https://arxiv.org/pdf/2408.10234

I think the articles uploaded to arxiv are typically preprints, so there could be some differences from the formally published version.

 

[A Stoner]

You're treating each word as having the same information content? Clearly, that's not so. A word like "the" contains almost no information, which you can clearly tell by the fact that you can entirely remove them from most cases where they're used and meaning is neither changed nor lost. By contrast, a word like antidisestablishmentarianism is a complex and conceptually-rich term that loaded with meaning and context.

To look at it another way, which is more in line with the article, Claude Shannon defined information as surprise. So, when you can predict the next word, with high confidence, it's unsurprising and thus has low informational content. This was best exemplified in a demonstration Shannon did with some students, way back in 1951. The goal was to estimate the entropy of the English language by attempting to guess the next letter.

​

https://mathweb.ucsd.edu/~crypto/java/ENTROPY/

​

The relevance to the finding in this paper is that a human speaker isn't making a decision, word by word, of what to say next. The speaker is making decisions about words, phrases, and even sentences. For instance, if I choose to conclude a short anecdote with the phrase "all is well that ends well", that was essentially one decision I made. Yes, I plucked that phrase from a large space of possible options, but it's not as large as it may seem when you consider I might have only some hundreds of such phrases that I use in my everyday speech.


But, it's not. So, the place to start disputing the claim is by understand their definitions and methodology. It's a shame the paper is pay-walled, but I'm sure more can be gleaned about these details, given I've heard this same finding repeated from other news outlets.

You are right, when you start saying 'all is well that ends well' as a singular thought and we can read the first part of that and understand what the likely end is.

But there in is exactly what I was saying. These thoughts, these ideas, these phrases cannot be wrapped up in one single bit, yes or no, on or off.

But as I said, if you are looking at them as bytes of high density, then you could argue we might be thinking in only a few per second.

a byte of 64 bits can in fact represent any of 18,446,744,073,709,551,616 different ideas, thoughts, combined and so forth. With that, sure, 10 of those per second sounds about like what I expect from a human brain.

Words have meaning and the fact is that with 10 bits you could not even create a singular letter of a word.

 

[bit_user]

You are right, when you start saying 'all is well that ends well' as a singular thought and we can read the first part of that and understand what the likely end is.

But there in is exactly what I was saying. These thoughts, these ideas, these phrases cannot be wrapped up in one single bit, yes or no, on or off.

I never said it was a single bit. My point was that the choice of each successive word isn't independent of the priors. For any given word, there's a limited number of options that would fit. Therefore, the decision space is quite restricted to something well below what it would be, if you independently encoded each word as an index into the speaker's working vocabulary.

I'd also note that set of words people use in speech and writing tends to be much smaller than what they can recognize. So, when you quote a person's working vocabulary, it matters which side of the communication you're talking about.

a byte of 64 bits can in fact represent any of 18,446,744,073,709,551,616 different ideas, thoughts, combined and so forth. With that, sure, 10 of those per second sounds about like what I expect from a human brain.

Nobody I've ever known can have that many independent, uncorrelated thoughts or ideas per second.

From one thought to the next, there are consistent subjects, themes, ideas. In other words, there's a flow. This is a defining characteristic of conscious thinking, that it's serial and coherent in nature. Since each thought doesn't exist in isolation, the new information it introduces builds on the context of previous thoughts. This makes it very much like language, in that each additional word only introduces a limited amount of information.

I expect the speed of our conscious thought and speech are similar for reasons. I'm not convinced you can have a complete, fleshed-out idea in much less time than it would take to convey it. For some ideas, speech isn't the most efficient way to convey them, so you could substitute drawing or arranging objects as a means of conveyance.

Anyway, even if you're right, the authors of the paper tried to quantify the rate at which people can transmit information and found it's limited to 10 bits/s. So, maybe there's someone internally generating information at a higher rate, but we'd never know, because they didn't find any quantifiable example of someone communicating faster than that, in any medium or activity.

Words have meaning and the fact is that with 10 bits you could not even create a singular letter of a word.

We don't think in letters. Furthermore, my whole point about phrases is that a lot of thoughts don't even map to single words. So, if you have a thought which requires 64 bits to represent, it would take several words to convey and they take time for you to either speak or write.