Why Emergence May Never Be Fully Explained | Mark Bailey

[00:00:00]I tend to think that there might be some metaphysical limit that there might be some level of strong emergence and and some subset of emergent phenomena that are not explainable. Um simply because I think any formal language that we use to try and explain these things is inherently incomplete.

[00:00:15][Music] Mark, in science and in philosophy, emergence is a potentially powerful and certainly a controversial topic and we covered on Closer to Truth a great deal.

[00:00:34]How can AI in general and algorithmic complexity in specific contribute to the debate? Yeah, it's it's a fascinating question. I I often ponder whether or not emergence and and sort of the the idea that certain things are inherently unpredictable as to whether or not that's merely an epistemic problem where we really just don't have the the math or the science to to fully grasp these kinds of things or whether or not it's something more fundamental like a more of a metaphysical or onlogical aspect of the universe that significantly limits our ability to make these kinds of inferences about things.

[00:01:09]I sometimes think about you know how how there are certain phenomena that we can relatively easily map to different types of mathematical abstract. So like you could have the movement of an object through space that can map to Newton's laws or general relativity or these kinds of abstractions that allow us to gain insight into them. But then you have these kinds of emergent things that you can't necessarily map to these sorts of like very deterministic or or relatively easily solvable kinds of abstracta. So like you could have a complex system that you know of an economy and then you have like the emergence of a market or the dissolvance of a market and you can't necessarily predict those things unless you have uh you know either you run a simulation with a high enough fidelity to sort of see these emergent dynamics or you kind of watch it uh play out in real life. So I think you know studying emergence and sort of the nature of emergence in these kinds of systems um will give us will enable us to make better predictions about how things might function and might give us some insight into things like explanability uh when it comes to AI systems which in turn would help us understand AI alignment which is how AI you know whether or not AI uh comports with human expectations or aligns with human values and then ultimately control of AI systems that may go out of alignment. Yeah. And and so emergence is a a critical factor because on each level of the organizational hierarchy of how the world is created from physics, chemistry, biology, sociology, etc. Um there are regularities some people call laws that are operating that are um that seem to be unreducible irreducible to lower level. Now there could be two reasons for that. One is a lack of of knowledge and a lack of capacity in chaotic systems in which it's extremely difficult and nearing impossible to make predictions. Um but in principle it could it could be done. Um and then there's what's called strong emergence where in principle it is not possible even with ultimate knowledge to be able to make that prediction. Uh yeah, and the question is is there anything in that latter category or is that an empty set that you know sounds right but ultimately knowledge uh will be able to explain all of these so-called emergent phenomena. I don't know. I tend to I tend to think that there might be some metaphysical limit that there might be some level of strong emergence and and some subset of emergent phenomena that are not explainable. um simply because I think any formal language that we use to try and explain these things is inherently incomplete. You know, you think can think of like u girdles incompleteness theorem and these sorts of things where any formal language is never fully self-consistent. And so that might be an inherent limitation of any uh of any logic that we might derive to try and try and explain these sorts of things. um you know because we don't necessarily have access to like whatever the metallic of the universe might be which very well could exist outside of our understanding of you know of how the how the universe operates.

[00:04:20]Um in AI because of its unpredictability uh how do you articulate the concept of of difficult to predict with the concept of emergence because they there's a lot of similarities but they're not the same thing.

[00:04:37]Yeah, absolutely. I mean it's so you know the difficulty in predicting I think has to do with the overall complexity of certain types of AI models uh where you have maybe billions of parameters that you can't necessarily map to uh you know you can't necessarily map or interpret in some way. Um I think you know there might be sort of a weaker condition of interpretability where you can correlate certain types of activations or you know within certain types of AI models like for instance if you have a neural network kind of model maybe you map the uh you know the activations of different neurons within that model to some output but that still doesn't give you explanability which I I see as more of a a harder problem because that would require some fundamental knowledge as to what's actually going on under the hood. Mhm.

[00:05:23]Um so in these ways I think you know there there can be a connection between um you know between emergence and and unpredictability in these sorts of systems.

[00:05:33]And h how would that be expressed? How could that distinction be um if not proven at least demonstrated?

[00:05:43]Well, I mean, you might have, you know, you might have interesting properties or phase changes that might occur, uh, you know, within certain types of AI models where the behavior all of a sudden changes into something else. You know, that might be sort of an example of some emergent phenomena that you can't necessarily predict from, uh, you know, reliably from from previous behavior or other things. So that would be really interesting if you see phase changes in predict in uh in output that wasn't expected. Now how would you differentiate that from hallucination?

[00:06:20]It's hard to say. I mean I think hallucinations are are a feature and not a bug um for for different types of AI systems. Um it is interesting seeing some of the behavior where uh I I saw an example actually today where uh you know it was a an example of a a large language model that was having a regular conversation with someone and then all of a sudden it started to uh you know talk actually pretty darkly about about humanity and things like that and then when inquired about that it just started to it it basically said it was some sort of an anomaly. So, you get these kinds of weird uh I don't know these these interesting kind of bizarre emergent types of behaviors, these hallucinations that that tend to happen in ways that maybe we can't explain. I mean, it might be an interpolation issue or something, but I'm sorry. Yeah. I I mean, one approach is to put some metaphysical terror in there in terms of an emerging phenomenon. Another is just that you know the programming was um was had a bug in it or was inappropriate. I mean there could be very simplistic explanations for that. Absolutely. Yeah.

[00:07:30]And I think that gets to this, you know, this dichotomy between, you know, like epistemic issues with emergence and more metaphysical issues with emergence where, you know, one might simply be the fact that we just don't have the tools or the math to explain certain types of things yet. And the other might be we may never have the the correct or the appropriate tools to be able to explain certain types of things. I think when it comes to AI, it's probably more the epistemic issue and not necessarily a metaphysical issue. Um, but I think there's certainly room for for examining the the more metaphysical aspects of emergence and the the unpredictable set of things. So, I want to take advantage of our discussion of emergence to ask you about a theory that you have uh um put out with Susan Schneider called the protoime interpretation of quantum mechanics. Um, tell me what that is.

[00:08:20]Obviously the foundations of quantum mechanics is a very uh hot area that uh we cover a lot and this is certainly a different way to approach it. Yeah. So Susan and I have a diverse set of interests uh you know when it comes to things like AI and and other types of things and and one thing that we're really interested in is sort of the nature of emergence and and what that how that can be how that can potentially explain certain types of physical phenomena. Um so you know one thing that we sort of pondered is is uh you know things like entanglement in quantum mechanics where it would seem that you might have information transfer between two different particles that might be separated at at vast distances which would you would think would violate what they call the no signaling principle in in relativity which states that nothing no sort of transfer of information can travel faster than the speed of light.

[00:09:12]Mhm. So I think you know the standard view is that uh entanglement is sort of built on a correlation but Susan and I are postulating that perhaps there's some sort of underlying topology and I don't mean like a cosmological topology of the universe but more or less something that that's more like a series of connections that exist sort of sort of outside of spaceime from which space like spacetime as we understand it as a metric space could potentially emerge.

[00:09:40]And that's where we sort of that's where we have you know sort of thought about this idea of protoime as being some sort of temporal ordering without time if that makes sense. So protoime could be something where there's some fundamental feature of the universe where everything can be mapped out in some certain way and then you know the universe as we understand it could sort of play out on on top of that that topology.

[00:10:06]Thank you for watching. If you like this video, please like and comment below.

[00:10:10]You can support Closer to Truth by subscribing. Closer to Truth is now accepting your taxexempt donations.

[00:10:19]Please come to closerto.com/donate. Thank you very much for supporting us and thanks for watching.