Stumbled across this article from last year, based on findings from Ruhr-University Bochum. Thought it supports HTM theory very well in this regard.
The researchers analysed the brains of 259 men and women using neurite orientation dispersion and density imaging. This method enabled them to measure the amount of dendrites in the cerebral cortex, i.e. extensions of nerve cells that are used by the cells to communicate with each other. In addition, all participants completed an IQ test. Subsequently, the researchers associated the gathered data with each other and found out: the more intelligent a person, the fewer dendrites there are in their cerebral cortex.
According to the HTM theory, neurons will add a lot of dendrite segments if theyāre constantly unable to predict whatās going to happen. Once the neurons can correctly predict things they stop growing more dendrite segments.
Itās interesting that theyāre able to measure the dendrites in a living person.
This is really interesting, and Iām looking forward to when we understand the upstream cause.
This end result of efficient dendrite connectivity in the cerebral cortex would only be created under the right global learning conditions, which I assume involves a lot of the chemistry related to attention, amongst other things.
The correlation of the test performance and dendrite count is interesting. I do not agree though on the measure of āintelligenceā it seems like the IQ test is not representative.
Well, whether an IQ test is representative of intelligence has always been contentious. But take away the term āintelligenceā, and take it for what it is - an IQ test consists of pattern prediction, sequence prediction, and reasoning. Then one can say that in these tasks subjects that perform better have fewer neurons activated. My hypothesis is that you would see the same results in other non-IQ based tasks that require similar processing in the brain, for example, coding and problem solving tests, which are usually given as part of recruitment/interviews. One can then argue whether those people that can quickly solve problems or write good code are āreally intelligentā. I still think these observations fit well within the context of HTM because HTM is essentially a pattern/sequence prediction framework.
IQ tests do not necessarily involve all the senses to contribute to the stimuli that are assumed to present a problem that tests an individualās intelligence. My point is each and everyoneās brain prefer a set of stimulus over another set of stimulus and that preference is what must intelligence tests must be targeted to, IOW intelligence tests must adapt to oneās intellectual preferences. In my experiments and simulations with HTM using nupic, one important thing, if not the most that Iāve learned from it is that the HTM regions prefer inputs (stimuli) at time t and because inputs naturally can be mutually exclusive, regions compromise the less preferred inputs (or vice versa), and eventually their corresponding neuronal connections are weakened (or strengthened). By mutually exclusive inputs I mean, due to how the region is wired and the order of the perceived inputs, some set of inputs cannot co-exist (stored in the region) with each other. I believe that time besides inputs (test) is an important factor to consider when observing neuronal connections.
In these particular article the important knowledge that I have learned here is highly agnostic to intelligence. And that this knowledge is that people who are good at X will have lesser activated neuronal connections when they encounter X. But classifying problem X as the intelligence tester is highly questionable to me. Being able to solve every type of game puzzles around the world (which is impossible) does not guarantee success in solving problems that are presented in his/her stimulus stream throughout his/her life.
Sure, but while youāre focusing on āintelligenceā, Iām not focusing on it at all, and instead my takeaway is that based on their experiments, people that perform well on certain cognitive tasks have less neurons firing. For a person not familiar with HTM this may seem counter-intuitive - people generally associate performance of something with āthroughputā, or āmanyā, but in this case itās the opposite. So one can say that the performance of an individual on certain cognitive tasks is measured in terms of neuron efficiency. But yeah one could tear apart results of any experiment when trying to generalize or extrapolate, which in this case I care very little about and am excited about the neural efficiency aspects.
It is important to mention that the assumption of intelligence is weak as this is the headline of the article, but my takeaway is similar as Iāve explained in my previous reply.
I feel this paper https://elifesciences.org/articles/41714 contradicts the paper you are discussing. Itās suggesting that people with high IQ have more complex and lengthier dendrites. Please correct me if I am wrong
Perhaps the difference is that this paper appears to be focused specifically on pyramidal neurons and dendrite length, whereas the other one being discussed above doesnāt appear to limited to the specific type of neuron, and is focused on dendrite density.
Current IQ tests are better than nothing and do have some correlation with performance in certain limited domains.
That said, the neuroscience of brain region functions and how they are connected suggests that āstandardā IQ tests only looks at a subset of brain regions to the exclusion of motor and sensory connections.
Knowing that functional regions in the brain can āborrowā capacity from others strongly suggests that people that have strong skills in other areas are likely to have achieved these abilities as a trade-off with the brain regions that contribute to high IQ scores.
If one must measure human abilities *(unavoidable if one is to measure progress to provide tailored teaching) a better approach may be to access the total person with the Multiple Intelligence model. I have not validated that there is a correlation between the proposed nine areas of intelligence and any particular clusters of brain areas but from a strictly intuitive stance - I think that there will be some correspondence that could be measured with modern imaging techniques. I suspect that the degrees of connectivity of fiber tracks would be the most productive area of investigation.
(*) There are multiple schools of thought about āmeasuring a personā revolving about personal dignity. While I think these concerns are well-intentioned the reality of teaching is that you do have to know what a person knows to adjust your delivered material to match the studentsā needs. A one-on-one tutor may not be using a formal test but the interaction with the student is actually a highly personalized and sensitive test when performed by a skilled teacher. You can misuse these measuring tools such as currently being done with āstandardized testingā as employed in the USA.
On IQ tests, I scored higher a decade ago than I did recently. A decade ago though, I also struggled to communicate in social situations, and had little social awareness overall. As Iāve made more effort over the past few years to learn how to interact with folks in a more āstandardā way and see better outcomes from my social interactions (this at the same time as learning Mandarin Chinese and living in mainland China for a few years), I feel like my IQ has fallen. The tradeoff is probably worth it though, as now itās slightly less likely that somebody will murder me out of conversational frustration. Some of this change may also have to do with decreasing amounts of sleep each night over the years as well.
So, IQ is useful, and it is a measurement of āsomethingā. But certainly everything needs context. Thereās a reason we have a stereotype of socially awkward smart people (i.e. Sheldon and friends from the tv show āBig Bang Theoryā)
I still think in general we have an overall problem with defining just what is āintelligenceā, and I donāt see any easy solution to that. Instead weāll keep having disparate sets of specific measurements in one area or another.
Possibly, was just pointing out the slight difference in focus of the two studies. Of course there are also a number of other neuron types in the cortex.
Interesting case. My case, I never have scored high in IQ tests, always average. I also do not think it is difficult to score high because I know which areas to expose myself to. Iām born with a perfect pitch in music and Iāve never been to any music school at all, today I think my skill alternates from perfect to partial. I did great scores in my Uni, had advance Maths both higher calculus and and CS ones almost been a key contributor to the company that Iāve worked on. However I still think that the most difficult intellectual area that Iāve encountered so far is in music. It doesnāt require a rocket scientist and a high IQ yet majority of us cannot play an instrument at a prof level. My point is the article is using a weak proof to present an interesting findings.
Everyone is still focusing on IQ aspects. Why donāt we propose an experiment using the same imaging technique, but with a different set of cognitive tasks - solving calculus problems, solving coding problems, playing a simple musical piece, etc? My hypothesis is that you would observe the same results, i.e. someone that can solve a calculus problem quickly would have smaller amount of neurons firing than those who struggle, or cannot solve the problem at all. In fact, I would argue that someone that knows exactly where to park their car in a parking lot would have lesser amount of neurons firing than someone who is struggling to find a parking space. I would say these results would line up quite well with what is observed on a āmacroā level - people that are exceptionally good at certain cognitive tasks, like driving a car through traffic, do so almost without even thinking about it, meaning less neural activity is taking place, and that gives them the ability to do additional āheavyā tasks like carry out complex conversations. Would Numenta be interested in such experiments?
What Iām curious about is how much of learning is active (i.e. āI want to learn this.ā) vs. passive (i.e. memorizing random facts and information in passing), and what that difference, if any, looks like in the brain.
I think itās relevant, because some people that might commonly be labeled āstupidā (in English, where the term is not as severe as in other European languages), might have the same dendritic density as an āintelligentā person, but merely specialized to the scope of their own daily challenges/context/existence.
My main point is that I suspect dendritic density might be a poor measure of intelligence on its own. The classical peasant/farmer may have the same number of dendrites as a leading expert in any academic field. Instead, maybe a better measure would be plasticity-to-permanence, where newer connections remain in place after one is exposed to new content. To that extent, Iāve personally seen professors, who when exposed to new information that may contradict their model of the world, react in ways that I would label as āstupidā.
I also have a suspicion that life circumstances would shape the observed patterns of brain activity; if someone on the lower rung of society is spending their time just trying survive each day, often through clever manipulation or complex social interactions full of nuance, maybe they would have more dendrites in order to process the variability of their circumstance. Somebody with a stable, routine lifestyle on the other hand might appear to have fewer, as they wouldnāt need to be in constant ālearning/survivalā mode to the extent a of a lesser well off personā¦ not to mention differences in hygiene and nutrition between them.
Max, it would be awesome to see how the learning process looks like from neural activity point of view, and yes, particularly for different ātypesā of learning methods, e.g. rote vs gradually building up from first principles, and if there is any difference at all, and if there is difference between different people (e.g. those with so-called photographic memory). And also how it looks like when youāre learning completely foreign concepts vs concepts that you can associate with something you learnt earlier.
Agree here. And to add the rate also of the lost connections vs the permanence would be useful. How important is a piece of memory over the other? Which memories may co-exist? Knowing these can be useful to our daily lives.