> By the way, the idea that the universe only makes sense “no more than one observer at a time” is one of the central ideas in Amanda Gefter’s amazing book, “Trespassing on Einstein’s Lawn”, which I heartily recommend.
Ok, I bought the book. Hope it arrives soon!
> Thermodynamics and information theory also do --- the interpretation of thermodynamics turns out to be as relevant as the interpretation of quantum mechanics in the debate about the ultimate nature of reality!
Indeed! And I have not elaborated it further in my essay (frankly, because I am still trying to understand it fully), but the observer-relativity of thermodynamics does seem to have quite important experimental consequences, as for example, in
where, quite literally, depending on how much you know about a system, you can get more or less work out of it.
> (I love reading the interactions between the participants, I often learn more about their ideas by doing this than by reading the essays, where the more “maverick” ideas are often presented more “carefully” and less explicitly!)
So now that we are at it, I must say that I very much admire how you interact with people. Not just fishing for high marks, I mean, but making good questions, and being honest, wherever you think you should. That’s the useful way to talk...
Now to your questions. They are difficult, which is great, because it means you took my message, and elaborated further (I cannot hope for more!). I am not always 100% sure of my responses, but this is as far as I got today, I should better write it down, before (a) I forget, or (b) I elaborate even further, and then it becomes all to long to explain.
> 1) At the bottom of p. 6, you say that a local decrease in entropy suffices for an observer to be able to describe agency, but don't you first need the observer to define the system so that there is a local decrease in entropy? Is there some sort of strange loop in your hypothesis?
I do not see there is a loop, as long as we leave self-awareness out of the picture. True, for entropy to decrease, somebody (the observer) needs to have defined the borders of the system. But the existence of the observer does not depend on the properties of the observed agent – this is a bit different from what you study in your essay, where the observer is always an emergent property of the observed universe. In my essay, again, leaving self-awareness out, the observer is always external to the observed system. The extreme example would be us looking at a the emergent physics of a simulated universe running in our computer.
At the risk of boring you, I will rephrase these ideas again, a bit more slowly. I always find a bit of redundancy useful (the question is how much is a bit..).
If we assume there is a universe and no observer observing it, then (as you know better than I) the universe does not make sense to anybody, there is no agency, there is no nothing, apart from some abstract structure. One could even say that such a universe does not exist, or if you prefer, it exists in a much weaker sense than an observed universe. So yes, you do need an observer. But I do not think this necessary implies a strange loop (as the one you have in co-emergence) because for the agent to exist, it does not really matter who the observer is, nor it is determined by the observation act. The observer can be the brain of a human watching the entropy-reducing subsystem. Or the brain + the skull. Or the whole human. Or the human plus his/her clothes and a couple of relatives. Only if we observe the observer does it really matter how we cut out the observer out of the universe around it. But in this case, the observer has become the object of observation of a higher-order observer. As long as the observer is the subject, and some other entity is the object, it does not really matter how the observer is defined, as long as it carries a representation (a model) of the observed object. The existence of the agent is a consequence of the observer’s observation act, but the existence of the observer is not a consequence of the agent.
As I view it, leaving self-awareness aside, all observation acts are hierarchical. So anything in grade to learn the regularities in the entropy reducing mapping (that is, in the goal-seeking process) qualifies as an observer. Observers do not need to be conscious beings in my essay (they probably do need to be self-aware in yours). A robot may qualify as an observer, as well as a lizard. It needs to be able to observe a system, and choose the way to represent its operation in such a way that many initial states of the observed dynamics are mapped into few final states. That very observation act is equivalent to assigning agency. In this respect, observers do not need to describe themselves, so I see no strange loop. Of course, you and I are curious observers, and we enjoy observing observers. In doing so, we need to define also observers, so yes, in that sense, observers also need to be cut out of the continuum of nature. But as objects of observation, not as subjects. We can climb the hierarchical ladder as high as we want, and I think everything is still ok. Until we observe ourselves, and there, yes, a strange loop appears. Or not, I am not completely sure that the I observer is the same as the I observed. That is why I dedicated some time to consciousness. I do not truly understand the cognitive mechanisms involved. But I am trying to make an effort to understand self-awareness with the same learning mechanisms with which I understand all agency allocation. With the difference that, by being circular, this observation act actually feels different. I am not sure whether this is correct, for now, it is just a working hypothesis. And here, at this point, yes, the argument is circular. But of all people, you are not scared of circular arguments, are you?
> 2) Near the one third mark of p. 7, you say that observers perform computations that are liable to iteration. Can you elaborate on what you mean by “computation” in this context?
To arrogate agency, observers make a specific computation. I like to imagine it in terms of a neural network implementation, but the actual substrate is irrelevant. Observers observe a system, realize that certain aspects are predictable (the goal-seeking attitude), and represent the observed process somewhere inside their architecture keeping only the degrees of freedom that allow for a predictable outcome to emerge (they don’t track the garbage variables). These are the computational abilities of observers, and they involve a learning process. But the world has merits on its own. The world is such, that often observers may allocate agency to multiple agents, using always the same computation. Simply because the world contains many repetitions of certain agents, some of which appear at different locations, at different times, and/or at different hierarchical levels. For example, if we detect a certain entropy-reducing subsystem and define it as "prokaryote cell", it turns out that many such bugs exist. So if we look around, we can always use the same computational algorithm to process our observations, and assign agency to good many prokariotes, with no need to learn once and again which degrees of freedom constitute the prokaryote, and which the garbage variables. In a way, they are all pretty much the same. And it turns out that these prokariotes combine/evolve into other subsystems that can be called "eukaryote cell", that are also numerous. Observers find it easy to detect eukaryotes because their brains have already learned to detect prokaryotes, and eukaryotes are nested prokaryotes. These, in turn, combine/evolve into multicellular organisms, and so forth. These regularities of the world are exploited by observers, in using always the same algorithm to process what they see in the world. Agency allocation is most successful when it not only allows observers to predict the future (the goal of the agent) at low computational cost, but in addition, to use the notion of agent to create higher-order agents at again low computational cost. The best agents are the ones that allow us to model not only their own agency, but many higher-order agents. With them, not only the behavior of individual agents makes sense, the entire collection of agents make sense.
> 3) In the middle of p. 8, you say that if Maxwell’s demon takes the weekends off, it is no longer a good idea to ascribe purpose to the gas. I kind of understand what you mean, but could you elaborate? Is it that when the behavior of a system becomes too complex, you have to look beyond the system to explain it?
Yes, exactly. Reality has lots of degrees of freedom. Agency only appears when an observer tracks only a subset of all the degrees of freedom, and fit a model to their evolution: the tracked degrees of freedom tend to do this or that. If the behavior becomes more complex, so must the fitted model. Maybe more degrees of freedom need to be tracked, or maybe the model needs to have more fitting parameters – which is kind of equivalent.
> 4) Near the two-thirds mark of p. 8, you write that “observers explore the power set of the system (…) and search for some entropy-reducing subset from which an agent or goal can be defined. They then discard the superfluous degrees of freedom, thereby compressing information.” Can you elaborate a little? Do you think that something similar could help construct a strange loop that explains the lawfulness of the universe we observe without having to presuppose regular laws?
Yes! I do! I had thought about this before, but did not elaborate enough to include it in my essay - nor dared to speak it out loud, to be honest! And now you are dangerously leading me to your ontological essay. I believe that if a universe were completely chaotic, no observer would be able to emerge there to perceive it. Or if some observer emerged, that observer would have to be governed by a subset of all the chaotic changes taking place in the universe. A subset that is orderly enough to make observation possible. For example, if the laws of a pixeled universe where such that each pixel of spacetime were painted with a random color obtained from a certain probability distribution, the collection of colored pixeles would look pretty boring, and nothing interesting would seem to happen there. But then, perhaps it could be possible to define objects inside that universe, composed of collection of adjacent pixels colored with similar colors. Depending on the coloring probability distribution perhaps some structure could be found in the so-defined objects. If we focus on the objects and ignore all the surrounding randomness, perhaps some physical laws could emerge even out of a chaotic universe. And who knows, perhaps creatures with brains could develop, some of them with observation powers. But I do not want to keep typing hunches, I would really like to think about all this carefully. Which is more the topic of your essay, than of mine, and very much related to Einstein’s question about God’s choices, by the way. In my essay, I gave it for granted that the laws of the universe exist, and observers are only regularity-searching learning machines. Now you invite me to take this idea as far as to the construction of the universe itself. It’s a great idea, but I surely need more time to process it!
> you quite rightly wonder “whether [a] recursive hypothesis constitutes an actual explanation (…), or simply a way to bind two loose ends together and worry no more.” "Strange loop" or "infinite tower of turtles" --- will we ever find a meta-theory of all-that-exists that makes us "worry no more?" :)
Yep. Strange loops are quite thrilling. There might be something really deep in them, and I guess your essay builds from that idea. But we have to be careful not to wrap into them all what we do not understand by hiding the loose ends, we need to find a way of looking at them in which they truly constitute an explanation. It’s tricky, because explanations normally have the topology of a line, and strange loops are… well… circular. But I still have hope...
> I am still trying to fully come to term with these ideas, but I intuitively feel that you are on to something important. I certainly see how your ideas could help make sense of the strange loops that are needed to construct an ultimate “self-referential” ontology --- what I call co-emergentism. Who knows? By pooling our ideas and those of like-minded thinkers about these issues, we may eventually converge towards a fully-fledged observer-centered metaphysics that tackles successfully all the major dilemmas of foundational physics… thereby justifying FQXi’s existence! ;-)
Well, that would be fantastic... I will post this now, and then take some time to think about the ideas you left on your forum. Those are more difficult for me, but I feel truly motivated. If my observer ideas can contribute to your ontology, then we would not just be talking about the relativity of one point of view or another, but also about the very nature of … well… everything. Sounds appealing, doesn’t it?
Thanks a lot for all these ideas, I really appreciate it when someone simply hands out thoughts. Let’s keep thinking…
PS: Fortunately we both ended pretty well in the contest. I got some good votes in the last minute (!). I find the heterogeneity of essays and voting criteria quite amazing. If it were not because questions as the ones we are discussing already take the entire time I can dedicate to fun topics, it would be interesting to analyze these strategies systematically. Another life, maybe. Or universe?
view post as summary