How does causality work in the brain at a neuron level?

I think that causality is an interesting topic nowadays. I’ve opened this topic to discuss how is a causation instance possibly achieved in the brain. Any thoughts?

UPDATE: The context for this question is how this is achieved in the brain biologically using a known model for example HTM or Thousand Brains Theory.

If they told you the answer they would be doing it - so the only way they could is by example.

It is a common occurrance that ordinary educated people sometimes mix or confuse correlation versus causality – I think this fact implies that “achieving causality”, in whatever sense, is not a capability in the brain hardware, or wetware to be more precise.

I believe that the basic function of the brain, at the hardware level, is achieving (establishing) “association”, which is a type of “correlation”, which includes but is not limited to “causation or causality”. Lots of associations. Nested. Recursive. Composite.

Somewhere along the human history at the software (aka culture or collective knowledge) level, philosophy (specifically logic and reasoning) was developed, and a special kind of “association/correlation” was named “causation”, along the line of “2nd law of thermodynamics” (irreversibility of certain processes). So today we as educated human beings, understand “causation” so intuitively, as intuitively as we understand “happiness or meaning of life”, hence we might mistake it as something native to our brain.

I think causation or causality is purely on the “nurture” side when we divide things along the “nature vs nurture” line. It’s part of our belief system – collectively developed by cultures of societies, not a lower level brain function (where only association is available).

1 Like

I do not understand what you are talking about.

Yes, causation is nothing but a strong-enough correlation. But this confusion not cultural, it’s an artifact of winner-take-all competition in the brain, which picks one “cause” when in fact there are many likely influences. And that WTA itself is an artifact of cortex-to-body bottleneck, likely mediated by Thalamus / TRN. While the cortex can track many influences / correlations, it evolved to guide the body. Which on a macro-level can only do only one thing at a time, hence the simplistic single-mindedness of our choices.


What do you mean by “belief system”? But isn’t this part of our cognition?

This is what I can think of quickly (without thinking) :smiley: . What do you think makes a strong correlation that makes it a causation?

This is what I had in mind -

“belief system (of which causality is just a tiny component/element)” vs brain

is similar to

“software (of which file types are components/elements)” vs hardware (mainly CPU+storage)

How does the computer hardware achieves the magic of processing photos or videos? well, text or photos or video files are all just 0s and 1s, to the computer hardware.

The hardware is always doing simple operations on binary digits (basically additions), no matter the software running on it is doing simple math, or editing photos/videos.

P.S. Superstition is often about “false” causalities. Superstitions are typically non-trivial belief systems.

To me, causality is a subset of all types of associations. Establishing associations of patterns is the only thing that the brain does, stated as the main thesis of HTM/TBT on intelligence: memory + prediction

1 Like

In another perspective, “software” is just an illusion and everything really is just 1’s and 0’s or a transistor’s state. This perspective is where I was trying to get at, as to how at this level in the case of the brain forms causation possibly using “strong correlations”. Why causations intuitively are much stabler for example than mere correlations and how correlations reach this stable state in the brain where we most of the time call it a fact. Is it a case of overfitting for example? Or is it an evolutionary goal of our brains to select and learn an external fact that is causation? On the other hand, if causation is merely strong correlations, how can we explain the case of a delusion, does it mean in this case causation is agnostic to facts?

We can only know correlations, ever. Causation is not a knowledge, it’s what what we project, by choosing one among many correlations. And the reason we evolved to simplistically choose only one is that the body can only react in one way at a time. The brain evolved for action, not for objective contemplation of the way things really are.

I really can’t agree. Causation is an expression of mechanism, of the application of rules of behaviour to physical objects that cause them to change state. Causation is a state change mediated by mechanism; correlation is a state change with a temporal relationship but without an identified mechanism.

In the same way software is an expression of computation. Data represents state, but computation/algorithms/software represent changes in state over time. Data is the digital equivalent of physical state, software is the digital equivalent of mechanisms (forces) that cause changes in state.

The key concept that links them is the flow of time.


That’s a great observation … why/how we feel differently about causations vs mere correlations?

Thought experiment:
Pre-Newton era - apples fall to the ground, instead of flying to the outer space. Even a monkey knows that, but why? Who knows. It just happens. There is no answer … it’s a mystery. Maybe that is related to what you imply by “correlations feeling less stable in the brain”?

After Newton’s discovery, the theory of gravity gradually got accepted by the whole human society (it became a common belief thereafter), every school child’s brain learned that “gravity” caused the apple to fall to earth, instead of flying away. This particular causality “became” a fact, which makes brains feel … more certain? less puzzled? stabler? (The brain hates uncertainty…more or less)

Seems reasonable.

P.S. So it seems to me that causality may not be resolved at the brain biology level, with or without HTM/TBT theories, very similar to studying jpg file format will not give me insight on how a photo of beautiful sunset is possible.

I definitely agree, that Causality is a very interesting topic! And I even think its not that complex to think of it and understand how its implemented in the brain.

I suggest reading: Causal Inference in Statistics: A Primer (by Judea Pearl)

In the first chapter already, the difference is explained: Usually in statistics, there is data, that was collected. For example the two random variables C (for Cancer) and S (for Smoking) can take two values, and the data can be summarized by a table with 4 entries, showing the joint probabilities.
Based on this data, it is impossible to figure out, if Cancer causes Smoking, or Smoking causes Cancer.
We need something else to decide this question!

In the book, the author argues, that a directed graph is needed in order to solve the question, so it is known beforehand, what is a source and what is a effect.
However coming to the brain, the notion of time is very important, since when you first smoke, and then get cancer, it is pretty obvious which is the source and which is the effect! The brain therefore learns to build a causal model, that links from sources to their effects. And further I would argue, that is also why axons transmit signals in one direction: This way it is clear, what caused what, it is inherent to the model.

Actually almost done reading witht the Book of Why by Pearl. He defines causality in a non-common-sense manner. Great reading so far. It opens up a new way of thinking in machine learning IMO.


  1. A → B, B → C, C → D, D → E, E → F, F → G and so on until Z are causations.
  2. Based on strong evidence we know that a blackbox knows O → P, Q → R, R → S, S → T.
  3. Based on strong evidence we know that a the same blackbox knows A → B.

At some point Y → Z comes up, or H → I beomes a strong theory (e.g. gravity). BUT we cannot intuit or prove about how the blackbox learned them given the knowledge in #2 or #3. We then build a theory that they are all “correlated” and that the blackbox learned it. New facts or strong theories then are easy to classify as “strong correlations” and therefore there is no causation but just correlations.

I believe that we aren’t just learning correlations but instead our brain is etched with causations for example neuron J spiked due to neuron I’s signal, this is a causation bound to the laws of physics/biology/chemistry. Looking at them as a group and a group of connections, they did not form by chance, they formed based on these fundamental laws but were nurtured (etched) by external/internal signals/inputs ingested by chance.

Are you putting forward causation as purely a mental phenomenon with no counterpart in the real world of physics? Would gravity cease if we humans did not exist?

Pearl published his causation work in two main works, his seminal (2000). Causality: Models, reasoning and inference, CUP and the more lay-popular (with Mackenzie) The book of why.

This whole causality thing intrigued me, but I didn’t want to wade through “Why” and certainly not his main work, so I found this very excellent summary that is basically a Cliff Notes on Pearl’s thought–did the trick for me.

Link here.

1 Like

Came across this 2011 article … in terms of wording, the above quote is probably the closest direct answer to the original question of this thread.

It’s an elaborate way of saying “causality is achieved in the brain biologically through synaptic plasticity”, which seems to me to be modelled in TBT as “synapse permanence/strength (combined?)”

To me that is still “association”, which includes both “causation AND correlation”, at the neural circuitry level.


Great point. I like that causality is brought down to neuron-level rather than high-level events which are very hard to quantify its causal effects. Everything starts at the primitive level anyway.

IMO there is a danger in equating association with causation. Statistical association does not require underlying fundamental mechanism such as neuron A spiked due to an electro chemical event (allows spurious correlations). It just requires that something happens with a certain acceptable frequency when some previous event occurred or was observed.

Depending on perspective could it be that causality is represented in the brain as the current brain function f, where for simplicity f(x) = y. Y’s are independent events but are confounded by F and therefore we see them as Correlations but we cannot easily fathom their causal relationships because we don’t know f in the first place. The stable Y’s are maybe really due to the consensus of external and internal f’s, X’s and Y’s, and even though they are correlations, we thought of them as causations because they resemble a stable function such as f.