Hacker News new | past | comments | ask | show | jobs | submit login
Cognition All the Way Down (aeon.co)
100 points by myth_drannon 5 days ago | hide | past | favorite | 25 comments

There's something called the Living Systems theory that postulates that there are significant parallels between subsystems, workings, failure modes and so on, between different levels of living systems. Where the levels are cells -> organs -> organisms -> groups of organisms -> organizations -> nations -> supra national institutions. The theory is not new actually.

So, it could be viewed as cognition all the way down and up. At least as a tool of analysis.

I've followed this thinking for awhile. I understand it as Complexity Science thinking, as espoused by the Santa Fe Institute. It's always struck me as the style of thinking that biomimicry is but a small subset of.

It feels like a way of thinking that elevates analogy between domains up to the level of partial knowledge. And I understand it as being more about generalized properties of networks -- seeing networks that "rhyme", even in totally different domains and strata.

Operating within these frameworks feels like looking at one of those 3D Magic Eye images. "Oh, if I cross my eyes a bit, I can see that death is simply a repeatedly discovered pattern that emerged in virtually all living networks, and death is just a process of collapsing/compressing a complex network agent into a lower dimensional representation that will be referential to other agents. And so how does this repeatedly successful pattern bear on how I should build a government? Or my organization? etc etc"

EDIT: I also find myself curious whether this truth is why we have this predisposition to see beauty in allegory and metaphor and poetics. Maybe our deep history and evolution knows that these things constitute a fuzzy sort of real knowledge. Maybe our biology knows better than us that networks share structure at various levels, and connecting them through linguistic bridges has always served us and made our collective will more powerful.

this all ties into things like systems dynamics and cybernetics, which look at feedback loops as the fundamental structure in a system. I've been looking for the kinds of communities/fields that recognise the properties you've described and study/discuss them more in-depth, as it's something I've noticed a few years ago and am deeply interested in. It seems the peak of investigation into system properties was in the 50s, and despite the idea (or at least the meme) of cybernetics being embedded into everything futuristic/high-tech, I haven't found much actual interest/community around these ideas outside of maybe academia (which I'm not involved in, or sure how I would even access).

I also believe this kind of thinking can have very real practical applications. As a trivial example, say you look at a criminal organization and treat it as a living system. Then you can identify what kind of failure modes there exist in living systems in general and how these can be forced, and then use these insights to incapacitate the organization. Granted, I'd expect that most of these forced failures, say destroying or overloading their communications channels, will already be known to practicioners of the field. However, there may be some methods that are overlooked due to how sciences dealing with other kinds of living systems are seen as completely unrelated fields living in their own silos.

What does death as a representation mean?

step 1 to macro-cognition is to get a lightspeed coms network linking every human to every other human in real-time ... complete!

I find it extremely fascinating - that the workings of our body and mind are actually encoded in the DNA. The secrets of AGI & consciousness are right there, and we know they are, but we just aren't able to read it so that we could replicate them. I've been wondering if there's some sort of computational method to extract the algorithms without having to simulate the biology.

They aren't. (The workings of body and mind are not encoded in DNA.)

The only thing that's encoded in DNA is how to make proteins. Everything else is emergent from the interactions of those proteins and "encoded" only in evolutionary history of how those interactions work when they do so in a repeatable fashion. DNA is not a "blueprint" for making a body or a mind... it is much more like the "seed" pattern for a run of Conway's Game of Life, except with a bit of indeterminism for environmental influences thrown in. There are things about the phenotype you can predict from the DNA because you've got past examples, but a small change in initial conditions or the environment during development can also throw you way off... you have to run the whole process (i.e. grow a complete animal) to ever be sure what the phenotype looks like or behaves like.

The blueprint analogy is too simplistic. Some genes called "transcription factors" activate and/or inhibit the expression of other genes, including other transcription factors. They form networks functionally equivalent to neuronal nets.

> The only thing that's encoded in DNA is how to make proteins.

Wikipedia on the human genome [1]:

> Protein-coding sequences account for only a very small fraction of the genome (approximately 1.5%), and the rest is associated with non-coding RNA genes, regulatory DNA sequences, LINEs, SINEs, introns, and sequences for which as yet no function has been determined.

Wikipedia cites [2].

[1] https://en.m.wikipedia.org/wiki/Human_genome

[2] https://www.nature.com/articles/35057062

Edit: Formatting


The point remains that the products of DNA self-assemble, in the correct environment, into the pattern that determines the organism. Eliminate that environment, and it is just a stew of RNA and proteins.

It would make sense if cells were only interested in talking about certain topics. In that case we could simulate the evolution of their communication and ignore what is causing the communication.

There should be an upper bound to the complexity of their communication because of game theory. Tit-for-tat only supports an error rate of up to 9 percent, so at the upper bound cells will need to differentiate.

This is a lot of domain knowledge to accelerate the computation by, provided the theory is correct.

Reading this as a biochemist in a past life, this was a WONDERFUL explanation.

> I've been wondering if there's some sort of computational method to extract the algorithms without having to simulate the biology.

That doesn't seem plausible to me. DNA, just like a compiled program, is a set of instructions for a specific computer. I don't believe you can extract the meaning of a program without understanding the computer that is supposed to run it (except perhaps by statistical analysis of its results when being run, which is generally how we are studying it today).

Now, if we did have a deep understanding of cellular machinery, of the way cells organize into organs and organisms, and of the functionality of each of these organs, including the brain, and the way they collaborate to produce cognition, THEN we could perhaps decompile the DNA into a high-level algorithm, where we could say something like 'these are instructions for programming a stem cell to specialize into a cell which feeds cells that react when this chemical is encountered in concentrations greater than X over a time period of Y'. But this is likely the opposite approach to what you were asking - you would need to have an almost complete understanding of the biochemistry of the entire body, and of the way it gives rise to intelligence, at which point the DNA itself may not be so important anymore in the hunt for AGI.

> I don't believe you can extract the meaning of a program without understanding the computer that is supposed to run it (except perhaps by statistical analysis of its results when being run, which is generally how we are studying it today).

This made me think of what forces of life are maybe engaging in within this universe: We can't know the structure of the machine running this simulation, but if we probe the results of the running simulation enough, we can understand everything of the system running it.

In the Einstein sense, maybe that's how we know God? :)

If the functional units of DNA are composable then the computational complexity of the problem decreases dramatically. We can use symbols instead of functions and derive an abstract algebra of life.

If we take the theory of natural evolution seriously then IMHO we should also take Gödel's incompleteness theorem seriously like Roger Penrose does.

Note that he proposes both a problem and a solution, which are two separate entities. There may exist other solutions to cognition of incompleteness.

I think that if the problem of and solution to incompleteness in cognition did not exist then we would experience these computational inconsistencies similar to migraine blind spots. However we don't, and we perceive the error in computation as a single entity and not as two boundaries.

That sounds like it'd be about as hard, as extracting the output of a complicated program without running it.

The authors are surely right that we are working on a too low level, and vulgar behaviourism has been slowly going out of fashion anyway.

However, they haven't yet convinced their approach is useful. I note that their approach can't (yet?) begin to answer their own worm question (which head will the worm have?) either, and their analogies to software and hardware seem simplistic.

As an article this is worthwhile, and their proposal may well turn out to be a useful research direction, but it's not there yet. Hopefully there'll be a part 2.

P.S. There's no need to convince us that multicell organisms are compatible with PD as the article tries to do - so what if they weren't?

Two days ago: https://news.ycombinator.com/item?id=24766148

Unsure how Aeon is doing this; some sort of URL trickery?

Hoh, "teleophobia", what a word for the aversion to ascribing will and purpose to living things.

> You think we shouldn’t anthropomorphise people?

The mechanistic/behaviorist way of looking at the world is fundamental to science as a discipline, and I suspect this originated from the crisis of religion - the split between matter and spirit, in terms of concepts. Consciousness and cognition have an uncomfortable resemblance to what we used to call "spirit".

It reminds me of the "uncanny valley", how some machines are starting to behave awfully close to living creatures; and machine learning ("artificial intelligence") producing results that seem as if there's an awareness, a mind of its own.

We know that they are just mechanisms, all matter and no spirit, but our guts sense a kinship - we have an emotional reaction to life or things that act like they're alive, like they're thinking. We look at an ant, moving around in the world, sensing and feeling, looking back at us with its eyes - and most of us would "anthropomorphise" it, to relate to it as "one of us". Our subjective experience tells us that we're not just objects, but something more, special, magical even.

But, having rid of magic, science must understand and explain convincingly the mystery of life and mind, the very strange material existence of ourselves.

Looking at how slime moulds crawl through mazes to find food, I get an uncanny feeling that it's alive and has a primitive mind. Or those sped-up films of plants reaching out their vines and tendrils, holding onto things and climbing toward the light - it has something in common with us at a fundamental level.

Science offers explanations for these phenomena, yes - but it seems there are still missing pieces, concepts, models, to tell the full story. Self-organization of matter, swarm behaviors, distributed intelligence, information and computation..

The problem of consciousness is divided into an easy and hard part. The hard part only requires an observer, but we can not observe the observer without a mind. This wrongly conflates the discussion.

A mind without an observer would indeed be uncanny valley stuff. I always ask people who claim consciousness does not exist if they are indeed undead, or merely confused by the terminology.

https://www.youtube.com/watch?v=LCvwyScn9jU briefly describes apparent learning by association in plants. For more, look up monica gagliano

Amazing article, that starts a little bit simplistic, but actually goes into the not so pop sci details

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact