My thoughts on The Unbearable Slowness of Being: Why do we live at 10 bits/s?

Note: I am not super well informed about this subject! My summary of the paper presented in this post could be very wrong.

Today’s paper isn’t about computing, but about human cognition.

As a kid, I hated writing essays in school by hand. I always felt like the pen just couldn’t move as fast as I thought. As I grew up a little and gained proficiency with a keyboard, I found that it felt slightly more possible for the words to keep pace with my ideas. Today, I like to play around with new keyboard layouts and input mechanisms to get closer to the ideal world of being able to have my thoughts manifest at the same rate I experience them. But how fast do I experience my thoughts? And how much slower is typing compared to that rate? This paper proposes that the throughput of the human brain is actually only around 10 bits per second.

The paper authors acknowledge that this is a surprisingly low number and draw comparisons to the bandwidth available via current networking equipment and highlight that network bandwidth is often several orders of magnitude larger. But what does 10 bits per second actually mean in this context, and how was it measured?

The authors provide a few examples of how they arrive at this number. First they examine typing speeds of professional typists, and show that the rate of characters being outputted corresponds to 10 bits per second. Note that bits here is not used in the sense of the number of bits required to represent the character, but rather the number of bits of new information gained within the context of the task (that is to say, a single character could be 1 bit or less and ignoring the fact that modeling the choice of a single character out of the set of possible output characters would require more than a single bit. In a sense, a bit is more like a single action, or a single decision taken by the brain - my understanding of this is definitely a bit sketchy and it took me a while to realize that it doesn’t literally mean 10 binary digits - or equivalent - streaming in and out of our neural pathways). This seems to hold across any output mechanism - even talking, and some back of the napkin math on my part seems to show that this even holds true for stenography.

However the paper also claims that it’s not just our output rate that’s capped at \(10b/s\), but also our input rate as well. This was demonstrated by memory games, such as showing a string of digits and measuring the number of memorized digits after a set interval. My initial reaction to this was to dismiss the claim - after all, my own memories of scenes are far more detailed than what 10 bits could afford. The paper addresses this by citing work on “subjective inflation”, a phenomenon where we perceive that we are perceiving far more information than we actually are. For example, when we take in a scene, we are normally only focused on a single object at a time, yet we feel as though we are perceiving the entire scene. While it’s true that we are aware of our peripheral vision, the amount of information gained is far less than our focus.

The paper also has a brief section on the theoretical capacity of our memories versus the amount of information that we could theoretically absorb at \(10b/s\). In short, they claim that the brain can store at most \(50TB\), but we can only input at most \(40GB\) over the course of a single lifetime. Another section I found interesting was on whether or not the brain can multitask. The paper cites prior work that shows that while processing incoming information is done via massively parallel computation, actual though processes are serial. This means that the \(10b/s\) limit cannot be enhanced by doing multiple things at once.

Thoughts

Again, I am not necessarily the intended audience for this publication. It took me a while to realize that my first read of this paper gave me an incorrect understanding of the claims presented. These thoughts are not well informed, but more like a collection of questions/thoughts that I hope will guide any future exploration I may attempt in this domain.

• Maybe the human brain isn’t good at acting as a bus that moves information from our perception to some action - maybe the processing capability is about understanding abstract concepts. How many bits represent the platonic ideal of something? The ability to absorb and synthesize information seems to be unrelated to pure information throughput, and could explain why humans are so biologically successful beyond the paper’s proposed idea that \(10b/s\) is simply more than sufficient for survival.
• It also seems to be that the human brain is amazingly good at compressing information into an efficient representation. I think I’ve read things before about how much information the brain filters out before delivering it to our consciousness, and it’s something I’d like to read more about in detail.
• We’ve established that 10b/s seems to hold for a single modality of information, but can that barrier be exceeded by combining multiple input modes? E.g. if we use multiple senses does that change how much information we perceive? It’s possible that perception is more than just the sum of our senses too - I’ve read about studies on improved recall in the presence of some stimulus like a memorable taste, or music, so maybe there’s some special processes involved with combining multiple modalities that allow for higher bandwidth?
• Is the rate adaptive? When I play games on an emulator with overclocking, my (perceived) reaction time increases as I adjust to the new speed - with repeated instances of testing at faster paces, do human improve at absorbing and replicating info?
  • from re-reading the paper, I think the answer to this is no. It’s more likely that even when things are “faster”, we might just be finding more efficient ways to only extract 10 salient bits of information.