Gilling theory of cortical layers

I am new here, and my name is Finn Gilling (from Copenhagen, Denmark)
I am impressed by the work of Jeff Hawkins and Numenta, and if I am permitted by staff, I think I can help promote the basic idea of using understanding from neuroscience, and in particular the neocortex, to promote the use of biological intelligence for building intelligent systems.
The subject here is “why the neocortex has layers and columns”: Yes in a broad perspective why?
The brain is a decision machine forced by evolution to always be able to answer the question “what to do now”. To do that, evolution has evolved the ability to use information from information channels separated in time each using different learning processes to maximize the value of the information in the NOW where it is used to make decisions about start/stopping movements.
The different information channels in time are the six different layers in the neocortex:
Now (6), moment (5), present (4), future(3), past (2), doubt (1).

In layer 6, the NOW, the system attends probably when to start/stop achieving a goal, start/stop engaging a target and start/stop moving.
In layer 5, in the moment, the system is timing when the goal is high on precision and the target high on accuracy (habituation) - this is the control layer
In layer 4, in the present, the system receives a continous inflow of sensedata (sensitization) - this is the regulation layer
In layer 3, the future, the system use sense data to predict based on assumptions about “if I do this, then this will happen” (operant conditioning) - this is the modification layer
In layer 2, the past, the system use sense data to predict based on anticipations about “what not to do” (conditioning) - this is the restriction layer
In layer 1, the endless time, the system tries to establish a link between initial doubt (as when htm blocks all 6 neurons in a column), to create qualified doubt by asking questions (the sequence is where, what, which, why, when, how) - this is the query layer…

There is thus good reasons to name the neocortex the decision system in the nervous system, since it decides what moves we are going to perform in the future. Also cells do that, and their “neocortex” is hidden in circuits. In a wider perspective the genetic functions (control, regulation) preserve information processing capabilities across species, the epigenetic functions (modification, restriction) preserves information processing capabilities across generations and finally the biological decision system (including the neocortex) preserves information processing capabilites across the individuals lifetime. You can see my proposed model of the functions in the human decision system in the model illustration here below/to the right. I have written a detailed book about it, and it is amazing how HTM and HDS are very close in concepts and functions, though HTM is at the bit processing level, and HDS is at the human decision level using a behavioral descriptive language. Combining these two languages in the future will give htm an interface to decision makers and in general humans operating in the world. This combination between man and technology we call ADI - automated decision intelligence.
Hope the here from some of you soon!
Finn Gilling

1 Like

I appreciate the interesting and novel idea, but this is a speculative expansion of our understanding of what the layers are doing. So I’m going to move this comment into #htm-theory:tangential-theories.

Hi Matt
thanks for answering - but it leaves me a little confused:
I understand what you are doing as (2+3+4 inference, 5 motor, 6 attention) as not contradictory but
rather the same:
In 2 past and 3 future we talk about inference, which is expressed as anticipation and assumption…
having consequences as restrictions and modifications in motor…since a data point from before can only have these consequences in the now (the system being always in start/stop or in between moving)…
So is it possible that you can tell me where I am contradictory? As you teach me: Be constructive to ideas not to people - so may be I did not understand htm correct, but I very much like to…

1 Like

sorry I am here again, but may be my explanation below can help you answering me…

How do you explain that the system needs to keep data from different time periods separated? can that all happen at the single neuron level? Or does the system at its top layer need to keep the effects of data from different time periods separated relative to the “now” it is always in:

In HDS the effect on motor output from each layer is:

  • start stop moving (layer 6)
  • control by repetition (layer 5)
  • regulation by difference (layer 4)
  • modification by assumption (layer 3) … which is exactly inference = self selection
  • restriction by anticipation (layer 2) … which another type inference = adaptation
  • doubt … which is some amount of entropy controlling the rest …(in HTM you “set” all column variables)

Anther question is how you treat doubt (=entropy) in HTM? Since the layers without doubt are handling entropy, giving commands to motor output…start/stop moving…

I hope you can clear up my mind about the differences and contradictions you see? Since I think that from my reading of HTM, that you in HTM exactly are using data on a timeline relative to the start and stopping of movements?

Kind regards

1 Like

Finn, I am totally open to your ideas, and I love that you are sharing them. You are taking HTM theory and expanding it to include additional theories. The place for this discussion is here. Please feel free to share your ideas as they are interesting and people here will read them. We are happy to have many different ideas, but we also cannot fully explore all community ideas to evaluate whether we necessarily agree with them or not. I hope you understand that we have our research team working in a specific direction.

I edited the post above because I was hasty in my generalization. It used to say that this was “directly contradictory to” HTM and I changed it to “a speculative expansion of”. I hope that makes sense and doesn’t mislead anyone.

hi Matt

Ok thanks, that a very different answer and I assume you are not just polite? smiling

“an expansion of”, is promising, since my view is that the dimension of top down view is important, and that searching for a biological top down view is an essence if one wants to reengineer what the brain is doing, after all it evolved from a single cell differentiation ending up in different types of neurons…ok bye for today