As a computer scientist, I was blown away the first time my friend explained to me that his research focused on the timing of neuron spikes, not their magnitude. After talking about it for a while I realized that machine learning neural networks are much closer to simple early models of how neuron's work (averages and all), not how neuron's actually signal. Makes sense when you consider how the latest LLM models have almost as many parameters as we have neurons, but we still seem pretty far from AGI.

All models are convenient fictions. I heard a neuroscientist once describe averaging as a low-pass filter. People know it hides high-frequency dynamics. But unless you have a way to interpret the high-frequency signal, it looks an awful lot like noise.

I've become increasingly convinced that the idea of averaging is one of the biggest obstacles to understanding things... it contains the insidious trap of feeling/sounding "rigorous" and "quantitative" while making huge assumptions that are extremely inappropriate for most real world situations.

Once I started noticing this I can't stop seeing this almost everywhere- almost every news article, scientific paper, etc. will make clearly inappropriate inferences about a phenomenon based on the exact same mistake of confusing the average for a complete description of a distribution, or a more nuanced context.

Just a simple common example, is the popular myth that ancient people died of old age in their 30s, based on an "average life span of ~33 years" or such. In reality the modal life expectancy of adults (most common age of death other than 0) has been pretty stable in the 70s-80s range for most of human history- the low average was almost entirely due to infant mortality.

The above example is a case where thinking in terms of averages causes you to grossly misunderstand simple things, in a way that would be impossible even with basic common sense in a person that had never encountered the idea of math... yet it is a mistake you can reliably expect people in modern times to make.

Very interesting how measurement limitations drive scientific consensus.

The author portrays this as a major flaw in neuroscience, but it seems like a natural consequence of Newton's flaming laser sword; why theorize about something that you can't directly measure?

Another convenient fiction is that neuronal communication is all spikes, 1's and 0's. In this fiction, neuromodulators are ignored. Glial cells are ignored. The immune system is ignored. The first neurons in the brain that select for different auditory frequencies are ignored - auditory hair cells release a steady stream of vesicles packed with glutamate, and postsynaptic glutamate receptors compute something like a moving average of the glutamate concentration in the synaptic cleft. But it's much, much more complicated than that. Sounds like something is holding back the field, but averaging is a pretty lousy description of what it might be.

There's an old case study from aerospace that shows up sometimes in UX discussions, where the US military tried to design an airplane that fit the 'average' pilot and found that they made a plane that was not comfortable for any pilots. They had to go back in and add margins to a bunch of things, so they were adjustable within some number of std deviations of 'average'.

There are even bigger problems. For example, the common “this region lights up more if this is done” type of fMRI studies are suspect because what the fMRI tool does may have no bearing to actual brain function. I read a book by a neuroscientist lamenting the abuses of fMRI in papers a while ago. Unfortunately, unable to locate the reference.

Great note Mark. I agree. Action potentials are noisy beasts but much may be hidden in spike time coding that is obscured by averaging.

There is an even lower level problem that deserves more thought. What timebase do we use to average, or not. There is no handy oscillator or clock embedded in the cortex or thalamus that allows a neuron or module or us to declare “these events are synchronous and in phase”.

Our notions of external wall-clock time have been reified and then causally imposed on brain activity. Since most higher order cognitive decisions take more than 20 to 200 milliseconds of wall clock time it is presumptuous to assume any neuron is necessarily working in a single network or module. There could be dozens or hundreds of temporally semi-independent modules spread out over wall clock-time that still manage to produce the right motor output.

Somewhat related book on how the concept of average can be misleading and/or detrimental, The End of Average

https://search.worldcat.org/title/The-end-of-average-:-how-w...

Reminds me of "When U.S. air force discovered the flaw of averages" [0]

[0]: https://www.thestar.com/news/insight/when-u-s-air-force-disc...

There's even an artist that made this point: Cartoon Neuron

https://www.redbubble.com/shop/ap/6229477

https://x.com/Cartoon_Neuron

The universe wants you to save energy so every intelligent life form allover runs flat copies of others, of which avg is just the scientific version thereof.

Thus neuro science is bad everywhere in the Universe.

There are three kinds of lies: lies, damned lies, and fake quotes.