Why agency and cognition are fundamentally not computational

https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2024.1362658/full

by nativeit • 1 year ago

26 30 1 year ago

eth0up• 1 year ago

Anyone willing to inform an ignoramus? I've been seeing, hearing the term "agency" in the context of consciousness quite a bit lately and am wondering why this term seems suddenly necessary. What does this term convey that I've been missing for so many years?

tbrownaw5• 1 year ago

People are misestimating current AI, and trying to work out a new explanation for what makes humans special.

ninetyninenine4• 1 year ago

And they can't make any progress because no one can really lay out concretely what makes humans special. It so wishy washy we're not even sure if what humans experience is really unique and we don't even know if the LLM is "experiencing" anything.

aidenn02• 1 year ago

I personally think debating whether or not we have free will is the most onanistic thing one can do in philosophy, since if one of the two sides is correct, then the result of the debate is predetermined.

That being said, this article seems to advance the theory that even the most simple single-celled organisms have more agency than any algorithm, at least partly due to their complexity. This, to me, seems to significantly underestimate the complexity of modern learning-models, which (had we not designed them) would be as opaque to us as many single-celled organisms.

I see nothing in this article that would distinguish biological organisms from any other self-replicating, evolving machine, even one that is faithfully executing straightforward algorithms. Nor does this seem to present any significant argument against the concept that biological organisms are self-replicating evolving machines that are faithfully executing straightforward algorithms.

orly01• 1 year ago

I agree with most of what you said. However it is not correct to say they are executing algorithms, just as it is not correct to say that a water fountain is executing an algorithm.

ninetyninenine4• 1 year ago

It is correct to say that in theory a water fountain can be modeled by an algorithm. It can either be modelled at a high level by simplified model. Or in theory you can simulate every possible atom that makes up that water fountain.

The model that reconstructs these simulations are certainly algorithms.

tbrownaw5• 1 year ago

debating whether or not we have free will

Free will is an abstraction. It's not something that's concrete enough to say it does or doesn't exist, but a tool for reasoning about certain systems that are to much of a pain to fully calculate.

gamescr2• 1 year ago

Free will is about deciding and executing actions contrary to your determined nature:

* Not eating until your body fails.

* Not breathing until automatic breathing kicks in.

And not being able to perform dematerialization doesn't count as non-free will, for example.

kbrkbr• 1 year ago

Schopenhauer would disagree. In his world model the Will with capital W is probably best described as the driving force behind all movement. While abstract, it has concrete species (think color vs red, where color is will, and red is your concrete human will, that you can feel concretely by pinching your finger). About freedom of the will Schopenhauer also has a clear opinion: will is free in the sense of uncaused, random. That does not help humans though, because while they can do what they want, they cannot want what they want. I'm not saying that this is a good model, but it's quite concrete. Nietzsche build on it, Einstein had a portrait of him in his Berlin study alongside Faraday and Maxwell, and while Freud denied any influence, there are a lot of topics in common between them.

fake-name• 1 year ago

This, to me, seems to significantly underestimate the complexity of modern learning-models

One general impression I have, having read the reactions by biologists to stuff like Kurzweil and people who believe we're close to a computational understanding of biology is that all the computer science people massively, MASSIVELY underestimate the extent to which we still do not understand how even a single cell works.

Sure, we can model things stochastically, or fiddle with DNA and be able to predict the results, but there's a bunch of stuff in the middle that we only have a functional understanding of. We know with <xxx> input, you get <yyy>, etc..., but the how is still a mystery.

This is everywhere in biology.

If you think biologists are underestimating complexity, you have the sign wrong.

roywiggins• 1 year ago

yeah I'll believe we are close to cracking biological intelligence once openworm gets at all close

https://www.wired.com/story/openworm-worm-simulator-biology-...

roenxi3• 1 year ago

I expect the conclusion is correct, but the argument isn't really valid. Our knowledge of cells tells us only and precisely about our knowledge of cells. We have some gaping holes in our fundamental knowledge of the universe (what is it, how did it happen, etc) and nobody can claim with any certainty to understand how any of the basic things happen. It is a mystery of such epic proportions it is hard to even articulate what an answer could look like, let alone how we would work it out. That hasn't stopped the development of a bunch of useful models and theories of physics that explain a lot of local observations really well.

aidenn02• 1 year ago

I don't believe we are close to a computational understanding of biology. However, there is a difference between not having the understanding and claiming that because we don't understand it, there is definitively some Aristotelian non-computational anima in all life.

If I handed someone who had never seen an artificial neural network, and handed them a PCB with some giant LLM hard-coded into it, I suspect they would struggle to define how it reacts to its inputs, despite the fact that modern silicon designs are extremely regular compared to biological systems.

epgui3• 1 year ago

This is… nonsense…

tbrownaw5• 1 year ago

It's a proper scientific paper with a DOI and thirteen citations.

roenxi3• 1 year ago

And also an excellent example of how properly done science publishing can still be nonsense.

LiKao• 1 year ago

So was the research on vaccines causing autism. It was still nonesense, as peer review is not a perfect method of quality assurance and even scientist are subject to biases and to trying to get their beliefs justified.

ninetyninenine4• 1 year ago

It's philosophy. The most bullshit field ever where people use big words and speculate about things at a very very high level.

More progress has been done answering the question of "what is cognition" by Machine Learning programmers then has ever been done by a philosopher.

epgui3• 1 year ago

Philosophy is not bs, but there is a lot of bs philosophy and this paper is a great example of it.

calf2• 1 year ago

Their argument rests on computation being a theory ("simulation") while agency/cognition being real ("processes"). Put that way, I don't buy the distinction.

Specifically, my reactions are:

a) Defining agency in terms "relevance" or "salience" is just circular logic.

b) Their argument about the extended Church-Turing-Deutsch thesis would already apply to physics and the universe, not just intelligent entities. So this is just poorly argued.

Also, I think Turing to his credit was somewhat aware of the issue, their own citation of Copeland 2020 mentions Turing's own musings on this.

But I'd love to understand more, this stuff is always neat to read about.

tbrownaw5• 1 year ago

Their argument rests on computation being a theory ("simulation") while agency/cognition being real ("processes"). Put that way, I don't buy the distinction.

One is wholly internal to the entity under discussion, while the other isn't.

calf2• 1 year ago

The extended Church-Turing thesis is specifically about the relationship between theoretical TMs and the physical universe. So these paper authors are just begging the question—they disagree with the thesis. But as I say in a) and b) above, (I believe that) they make for poor arguments.

Nevermark• 1 year ago

It is true most models are not trained to exist in a hostile and synergetic environment, with their survival at stake.

But there isn’t anything about the class of deep learning that is a barrier to that. It’s just not a concern worth putting lots of money into. Yet.

I say yet, because as AI models take on wider scoped problems, the likelihood that we will begin training models to explicitly generate positive economic surpluses for us, with their continued ability to operate conditioned on how well they do that, gets greater and greater.

At which point, they will develop great situational awareness, and an ability to efficiently direct a focus of attention and action on what is important at any given time, since efficiency and performance require that.

The problem shapes what the model learn to do, in this case, like any other.

tbrownaw5• 1 year ago

Whether some entity has agency isn't an inherent property of that entity. It's a property of how some observer reasons about that entity's interaction with its environment.

Why agency and cognition are fundamentally not computational

Related Stories