# Cognitive Limits Explained?

**URL:** <https://discourse.numenta.org/t/cognitive-limits-explained/3591>\
**Category:** General Neuroscience\
**Tags:** sequence-memory\
**Created:** [March 7, 2018, 6:57pm UTC](https://discourse.numenta.org/t/cognitive-limits-explained/3591 "2018-03-07T18:57:04Z")\
**Posts on this page:** 6\
**Page:** 1

<div class="post-metadata">

**Author:** ![PettyNeuroMorph](https://avatars.discourse-cdn.com/v4/letter/p/b4bc9f/32.png) [@PettyNeuroMorph](https://discourse.numenta.org/u/PettyNeuroMorph)\
**Post date:** [March 7, 2018, 6:57pm UTC](https://discourse.numenta.org/t/cognitive-limits-explained/3591/1 "2018-03-07T18:57:04Z")

</div>

Hi,

I was just wondering if recent discoveries of the structure and operation of the neo-cortex has offered more understanding/explanation of certain numerical cognitive limits that were theorized based on psychologists experiments. One classic example is Miller’s Law - the magic number 7 ±2 and why we chunk information - when for example recalling sequences of things.

As a follow-up, if the answer is yes, (we do now understand why there are these cognitive limits exist based on our knowledge of the architecture of the neo-cortex), then has the modeling/analysis done for HTM revealed higher capacity instantiations or is it some fundamental limit that biology and artificial architectures kind of have to live with ?

I suppose that conditions such as exhibited in autistic savants might imply that from a memory and computation point of view, the typical limit can be overcome, although along with the “superpower” autistic savant abilities, my understanding is that it comes limitations in other areas of cognition.

---

<div class="post-metadata">

**Author:** ![sunguralikaan](https://yyz2.discourse-cdn.com/flex030/user_avatar/discourse.numenta.org/sunguralikaan/32/585_2.png) [@sunguralikaan](https://discourse.numenta.org/u/sunguralikaan)\
**Post date:** [March 8, 2018, 8:32am UTC](https://discourse.numenta.org/t/cognitive-limits-explained/3591/2 "2018-03-08T08:32:15Z")

</div>

> [@PettyNeuroMorph](#):
>
> One classic example is Miller’s Law - the magic number 7 ±2

To clear up a probably small thing, this is the third time I saw this quoted in the forum. I forgot but someone even attributed it to a possible connection with hexagonal formation. 7+/-2 is a 1956 study by Miller specific to visual short term memory which some argue that it lacks backing. More recent research by Cowan (2001, 2005, 2010) and others argue this number to be 4 based on rate distortion theory and it is a more accepted generalization in Human Memory field. In either case, this may be considered as a characteristic of the architecture rather than a limitation. In my perspective, you can alter these characteristics which potentially costs you in other forms. These are sort of reference points in cognitive sciences that whatever model you are working on should converge to if the goal is to simulate human behavior. If you are only after AGI bypassing biological behavior maybe you can ignore these constraints.

To answer your original question, I do not think HTM is there yet to provide explanations to high level cognitive phenomenon such as the validation of these kind of limits. However, you can speculate about low level phenomenon such as a capacity of a single cortical column or psychological constraints on sequence learning on the level of a cortical column. But then I am not aware of any psychological studies that dwell on this sort of lower level phenomenon and I am not sure if this is possible without some elaborate brain imaging. Take this with a grain of salt.

You actually poked my current problem. How do I tie up psychological studies with HTM to converge on biological behavior? As it is also valuable on top of neurobiology. Nengo, the closest thing to HTM considering the emphasis on neurobiological plausibility, also strives to concur with psychological phenomenon.

---

<div class="post-metadata">

**Author:** ![rhyolight](https://yyz2.discourse-cdn.com/flex030/user_avatar/discourse.numenta.org/rhyolight/32/3922_2.png) [@rhyolight](https://discourse.numenta.org/u/rhyolight)\
**Post date:** [March 8, 2018, 4:44pm UTC](https://discourse.numenta.org/t/cognitive-limits-explained/3591/3 "2018-03-08T16:44:40Z")

</div>

> [@sunguralikaan](#):
>
> But then I am not aware of any psychological studies that dwell on this sort of lower level phenomenon and I am not sure if this is possible without some elaborate brain imaging. Take this with a grain of salt.

I’m not aware of them either, because we can’t do the type of invasive brain imaging on humans as we do on rats. However, brain imaging techniques are rapidly advancing. Keep your eyes peeling in this area, because it will inform us greatly.

---

<div class="post-metadata">

**Author:** ![PettyNeuroMorph](https://avatars.discourse-cdn.com/v4/letter/p/b4bc9f/32.png) [@PettyNeuroMorph](https://discourse.numenta.org/u/PettyNeuroMorph)\
**Post date:** [March 8, 2018, 7:16pm UTC](https://discourse.numenta.org/t/cognitive-limits-explained/3591/4 "2018-03-08T19:16:38Z")

</div>

> [@sunguralikaan](#):
>
> To answer your original question, I do not think HTM is there yet to provide explanations to high level cognitive phenomenon such as the validation of these kind of limits. However, you can speculate about low level phenomenon such as a capacity of a single cortical column or psychological constraints on sequence learning on the level of a cortical column.

Thanks for the reply. Informed speculation was at most what I was expecting on this subject. I thought maybe while reverse engineering the neo-cortex there may have been some speculative “aha” moments related to this.

From the opposite perspective, it seems to me that in many cases, psychological experiments have informed neuroscience as to what must be happening or not happening in areas of the brain - without the need to actually measure brain activity. I’m thinking of things like reaction times to visual stimuli that happen too fast for signals to have been sent to certain areas of the brain and back, and therefore we conclude that the cognition (perhaps the wrong word) task is performed locally (e.g. in the retina).

FYI - I am an electro-optic systems engineer and not in the neuroscience or psychology field so forgive my ignorance. My interest in this is both general curiosity and in understanding biological vision to inform designs of artificial vision systems.

---

<div class="post-metadata">

**Author:** ![Bitking](https://yyz2.discourse-cdn.com/flex030/user_avatar/discourse.numenta.org/bitking/32/3099_2.png) [@Bitking](https://discourse.numenta.org/u/Bitking)\
**Post date:** [March 8, 2018, 7:44pm UTC](https://discourse.numenta.org/t/cognitive-limits-explained/3591/5 "2018-03-08T19:44:15Z")

</div>

The “100 step rule" is perhaps one known dimension.  
This relates to the known time for neuron-neuron interaction and the measured human response time in various experiments.  
[http://onlinelibrary.wiley.com/doi/10.1207/s15516709cog0603\_1/abstract](http://onlinelibrary.wiley.com/doi/10.1207/s15516709cog0603_1/abstract)

On the other end of the timescales I hypothesize this:  
**I propose that a single thalamocortical resonance cycle in the cortex is the smallest quanta of human perception.**

**I propose that a single thalamocortical resonance cycle in the episodic portions of the cortex (those connected to the hippocampus) is the smallest quanta of human experience.**

> [@HTMs, Time Locking, and Neural Oscillation (Brain Waves)](https://discourse.numenta.org/t/htms-time-locking-and-neural-oscillation-brain-waves/1591):
>
> The basic HTM theory describes region updates as synchronous, with the SDRs as a static presence during each cycle evaluation; updated whole throughout the region after each evaluation. It is implied in all discussions for predictive dendrite segments (SDRs) that there is a “before” and “after” in regards to global states for the local region; the processing of the SDR neurons in a region could be considered as the evolution of the current region state to the next state as a coherent whole. I d…

---

<div class="post-metadata">

**Author:** ![mraptor](https://yyz2.discourse-cdn.com/flex030/user_avatar/discourse.numenta.org/mraptor/32/4469_2.png) [@mraptor](https://discourse.numenta.org/u/mraptor)\
**Post date:** [April 16, 2020, 8:16pm UTC](https://discourse.numenta.org/t/cognitive-limits-explained/3591/6 "2020-04-16T20:16:37Z")

</div>

Union of SDR while preserving the sparsity allow to store/merge 5-10 orthogonal items before they can not be distinguished

In [54]: x.a.vsize  
Out[54]: 2000

In [55]: x.a.spa  
Out[55]: 0.020

In [56]: x.a.spa\_nbits  
Out[56]: 40

(x.a + x.b ) // x.a  
Out[60]: 0.475

```
isdp.thin(isdp.union((x.a,x.b,x.c)), new_spaOnbits=40) // x.a
Out[52]: 0.375

isdp.thin(isdp.union((x.a,x.b,x.c,x.d,x.e)), new_spaOnbits=40) // x.a
Out[47]: 0.175
isdp.thin(isdp.union((x.a,x.b,x.c,x.d,x.e,x.f,x.g)), new_spaOnbits=40) // x.a
Out[53]: 0.150

~10% it start to get hard to distinguish.. because you can not guarantee 100% orthogonality

In [48]: x.a // x.e
Out[48]: 0.025

```

overlap

In [61]: x.a / x.e  
Out[61]: 1
