What is your theorizing process like?

I’m wondering how people go about expanding HTM theory.

For example, what sort of information about neuroscience, HTM’s problems and capabilities, etc. do you look for and find most helpful? How do you keep track of your ideas? (I find it difficult to remember my hypotheses and even more difficult to write them down clearly.) Are there any information sources you find particularly useful?

Any tips are helpful.


Not a tip, just a guy in the same boat. I always found having a working model to experiment on is a requirement. After all this is a “thing” that you can only say it’s statistically working for some stuff and has a somewhat biological basis. It is so hard to theoretically grasp what leads to what. Most importantly by not having an experiment platform, you miss out on a solution that you would get from your somehow failed experiment. Personally, I just look at what’s going on and try to add or subtract things to accomplish some goal with it. So the workflow would be:

  1. Design a test case to observe learning behavior.
  2. Have an objective for that scenario
  3. Run it. Splits into two:
    • Increase the complexity of the task if things go well. Back to step 1.
    • Observe what is preventing it from accomplishing the task.
  4. Throw out a potential solution to it.
  5. Look for its biological plausibility on papers. Reference those papers. Split into two:
    • a: It seems plausible so implement it. Back to step 2.
    • b: It seems unlikely. Splits into two:
      • Implement it just for the sake of coming up with biologically plausible one while observing the “artificial” one.
      • Don’t implement it and search for a better solution. Back to step 3.

I find that working through David Marr’s 3 levels[1] is helpful in the evaluation phase.
I also find that going on long walks and dictating/explaining a given process into my cell phone to be helpful.

Each time I try to work up a slightly better version of the explanations and work through some of the areas I am struggling with. If you can’t explain it to yourself you don’t really understand it.

I do transcribe some of these sessions and it helps to reexamine some of the random threads and conjecture that pop up during these walks. This drives my internet searches and subsequent reading. This new reading leads to filling in some of the murky areas and a fuller picture of the whole AI landscape.

At latest count, I have about 6000 papers that I have downloaded or have links to. They are sorted by general topic. My electronics engineering, programming, neurology, and AI books fill most of a standard 6-foot high by 3-foot book shelve unit. Each book is read as it comes in. The PDP (parallel distributed processing)-David Rumelhart & James McClelland, Vision-David Marr, Gödel-Escher-Bach-Douglas Hofstadter, Neurophilosophy-Patricia Churchland, and the On Intelligence-Jeff Hawkins & Sandra Blakeslee books were some of the most influential when I encountered them. Each was a “whack on the side of the head” that caused me to think differently about what I thought I knew.

This return to the resources is a key part of the theorizing. I find it endlessly amazing that when it comes to studying the brain - people have been poking around on this for over 100 years. I have yet to ask a question and not find that someone has been studying that topic in great detail and published the results. This useful chunk of knowledge is just waiting for me find it and fit it to the question I am asking.

I think that the study of the brain is much like an earlier state of chemistry before the periodic table - everything is there just waiting for the right person to come along and rearrange the bits and bobs in a way that reveals the greater structure waiting to be discovered.

I have been doing this for most of 40 years now and some of this stuff is starting to make sense. I still have many questions about data format and processing algorithms.

If I had done one thing differently on this journey it would be to have made some sort of indexing system for interesting bits as I encountered them. I have a few things that I remember reading but I can’t remember where I read them. This was much worse before the internet when I was doing interlibrary loans and had to give the books back when I was done reading them.

[1] Marr’s Levels of Analysis


Reading lots and lots of material.
“A couple of months in the laboratory can frequently save a couple of hours in the library.” - Frank Westheimer

How to Retain More of Every Book You Read