An abstract language of thought for spatial sequences in humans

Interesting work and research by Dehaene et al. Is anyone at Numenta, or in this community, looking at his thesis in the light of HTM and TBT (specially location-based format of cognitive maipulations)?
His thesis is: “humans encode the transitions between sequence items as rotations and symmetries, and compress longer sequences using nested repetitions of those primitives”



My gut reaction to the title: “this is going to be a bunch of gibberish”

Is it worth a read?

Well I wouldn’t say its gibberish, but also I wouldn’t say its worth reading either. The author Dehaene previously wrote about the “Global Neuronal Workspace” theory, which is definitely worth reading.

I skimmed this current article and here are my thoughts:

  • This is science in the making. If this were food it would still be in the kitchen. I’d give this article a pass unless your willing to roll up your sleeves and stick your hands into a bowl of raw ingredients - so to speak.

  • There is some discussion of whether the task is only solvable by humans. The compare to (similar?) study which found that monkeys require 1000’s of practice iterations to succeed at the given task whereas humans need only need a few tries. This is a red flag for me.

  • From the abstract: “their brain activity was recorded using magneto-encephalography. The entire sequence could be decoded from brain signals.” There is a saying in neuroscience that goes like “if you look for it hard enough you will find it”. I’ll bet they could have decoded a lot of other things from the MEG data too, including things which validate other competing hypotheses.

  • All of their data comes from MEG & behavioral observations. MEG data is not really conclusive in the same way as direct electrical recordings from individual cells are. Here is an article which discusses issues with EEG recordings, and I think some of it also applies to MEG:

    • Herreras O (2016) Local Field Potentials: Myths and Misunderstandings. Front. Neural Circuits 10:101. doi: 10.3389/fncir.2016.00101

Thanks for the summary!
I’ll try to use less dismissive language in the future