What are brains?

It's clear to me that this is a simple case of map-territory confusion, though I feel a bit weird explaining it to a co-founder of CFAR.

"Thingness" comes from the map. Referents of "things" can be in the territory. "Realityfluid" - terrible name by the way, lets call it "fundamentals" - is the territory. Territory exists regardless whether it's interpreted by a subject of cognition or not, but subjects of cognition understand the territory only through a map. "Laws of physics" can be about properties of the fundamentals or about their reificated representation on the map and we need to be careful not to confuse these two [LW · GW]. 

So… when do we get to the place where we aren't using objects to explain how the impression of objects arises?

You're very clever, young man, very clever. But it's objects all the way down.

What you call "reification", I call "modeling".  This is what it feels like to be an algorithm (at least, THIS algorithm - who knows about others?) which performs compression-of-experience and predictive-description-based decisionmaking.  On many scales, it does seem to work to do the rough math of movement, behavior, and future local configurations based on aggregates and somewhat arbitrary clustering of perceptions.  

Brains, to the extent that they are useful to think of as a class (another non-real concept) of things (unreal, as you say), are local configurations of universe-stuff that do this compression and prediction.  When executing, they are processing information at a level different from the underlying quantum reality.  

The universe IS (as far as any subset of the universe, like a brain, can tell) swirling magical reality fluid.  A thing is it's own best model, and this includes both "the universe" and "the portion of the universe in any given entity's lightcone".  Brains are kind of amazing (aka magical aka sufficiently advanced technology) in that they make up models and stories which seem to make sense of at least part of what they experience, at some levels of abstraction.  Note that they ALSO hallucinate abstractions of configurations that don't occur (counterfactuals) or are outside their lightcones (for MWI, FAR outside).  

I think reductionism is a very useful model for many (MANY!) levels of abstraction, but I have to admit to believing (also a synonym of "modeling") that when taken far enough, there will be a mysterianism underlayer, where it's un-measurably and un-modelably complex.  It's unknown how far that is - current ability is probably seeing mysteries that can at some point become models, even if there are true un-modelable layers much deeper.  Scientific progress consists of peeling the onion one more level, but we will ALWAYS find mystery at the next level, which we will scratch at until we either dissolve the mystery into our models, or ... something.  It's unknown even whether we'll be able to know when we reach "the bottom".

[ Epistemic status: I think there's some valid modeling in what I say, but I don't claim it's complete nor perfectly coherent.  ]

What you are gesturing at is how to identify an embedded agent in "the real world", as far as I can tell. Then you keep asking deeper questions, like "what are these laws of physics". 

So, let's start from the basic assumptions, let me know if they make sense to you.

Assumption: there is a universe of which we are a small part (hence "embedded agency"). Basically "something exists". 

Assumption: from the point of view of Laplace's demon, we are identifiable and persistent features of the world, not Boltzmann brains.

Note that at this point we have not assumed the existence of "time" or any other familiar abstraction in our mental map. Just "externally identifiable structures".

Also note that the world might be a completely random instance of whatever. A bunch of rocks. You can even find loose patterns in white noise if you look hard enough. I wrote a couple of posts about it some years ago:

https://www.lesswrong.com/posts/aCuahwMSvbAsToK22/physics-has-laws-the-universe-might-not [LW · GW]

https://www.lesswrong.com/posts/2FZxTKTAtDs2bnfCh/order-from-randomness-ordering-the-universe-of-random [LW · GW]

Now, if we assume that this world contains something externally identifiable as "agents", it implies another very strong assumption: internal predictability. This is usually glossed over, but this point is crucial, and restricts possible worlds quite a bit. Again, it is a very very very strong constraint. The world must be such that some relevant large scale features of it can be found inside an incredibly tiny part of it. I do not necessarily mean "spatially tiny". We have not assumed the existence of space yet. Just that there are small subsets of the whole thing that have identifiable (at least to a Laplace demon) features of the whole thing.

Now, given this assumption, you can talk about the world being usefully lossily compressible, to an extremely large degree. "Usefully" here means that the compressed image of the world can fit into an agent and can be traced to be "used" by the agent. Actually the meaning of "usefully" and "used" is a separate can of worms deserving much more than a couple of sentences.

Now, at this point we got "physical laws": the distillation of the compression algorithm that fits into the agent. For some agents (bacteria) it is "identify sugar gradients and eat your way up the gradient". For others it is "quantum field theory that predicts the mass of the Higgs boson, given what we can measure". 

This is a crucial point. The world does not come with "matter" and "laws" separately. Physical laws are agent-size distillations of the world, and they are compatible but not unique, and depend on the agent. 

To recap, the chain of reasoning about the world goes like this: Something exists -> we exist in this "something" -> for us to persist the "something" must be compressible -> these compression algorithms are physical laws (and sometimes moral laws, or legal laws).

So when you say "brains exist as reifications of brains." what you probably mean is "the world is predictable from the inside".

I believe the Tractatus Logico-Philosophicus addresses this question. In a very important sense, philosophy and brain analysis pulls the rug out from under its own feet. Wittgenstein provides lots of interesting ideas but famously concludes with: "What we cannot speak about we must pass over in silence."

My informal chain of reasoning is as follows:

1. How could this paradoxical "thing before things" is be described in human language at all?
2. Rather, it must be pointed to.
3. Presumably some pointers would be more effective than others, but then what is the metric that determines which ones are more effective?
4. That must also be a "thing before things."
5. If we cannot reify the "thing before things" or the metric for it, we may still have hope of reifying it, since we can reify enough things around it until we have an extremely reliable pointer.

In my opinion, what this means in practice is that the best pointer to the question is basically every philosophical and religious response of humanity to Truth combined into a single whole. This is essentially a holistic combination of all pointers along with a metric for going towards better pointers.

In any case, I am pretty sure the Tractatus Logico-Philosophicus answers this question well, so maybe look there?

I like some of the other answers, but they aren't phrased how I would explain it, so I'll add my own. This is something like the cybernetics answer to your question.

The world is made of "stuff". This "stuff" is a mixed soup that has no inherent divisions. But then some of this stuff gets organized into negative feedback processes, and the telos of these feedback processes creates meaning when they extract information from sensors. This information extraction tells this from that in order for the stuff thus organized to do things. From this we get the basis of what we call minds: stuff arranged to perform negative feedback that generates information about the world via observation that models the world to change behavior. Stack enough of these minds up and you get interesting stuff like plants and animals and machines.

So the brain, though kind of weird to think about, is just this kind of control system, or rather an agglomeration of control systems, that is able to do things like map the territory it finds itself in.

I try to cover this topic in some depth in this chapter of my in-progress book, here [LW · GW].

Regarding the first part, here's what comes to mind: Long before brains evolved any higher capacities (for "conscious", "self-reflective", etc. thought), they evolved to make their hosts respond to situations in "evolutionarily useful" ways.  If you see food, some set of neurons fire and there's one group of responses; if you see a predator, a different set of neurons fire.

Then you might define "food (as perceived by this organism)" to be "what tends to make this set of neurons fire (when light reflects off it (for certain ranges of light) and reaches the eyes of this organism)".  Boundary conditions (like something having a color that's on the edge of what is recognized as food) are probably resolved "stochastically": whether something that's near the border of "food" actually fires the "food" neurons probably depends significantly on silly little environmental factors that normally don't make a difference; we tend to call this "random" and say that this almost-food thing has a 30% chance of making the "food" neurons fire.

There probably are some self-reinforcing things that happen, to try^[1] to make the neurons resolve one way or the other quickly, and to some extent quick resolution is more important than accuracy.  (See Buridan's principle: "A discrete decision based upon an input having a continuous range of values cannot [always] be made within a bounded length of time.")  Also, extremely rare situations are unimportant, evolutionarily speaking, so "the API does not specify the consequences" for exactly how the brain will respond to strange and contrived inputs.

("This set of neurons fires" is not a perfectly well-defined and uniform phenomenon either.  But that doesn't prevent evolution from successfully making organisms that make it happen.)

Before brains (and alongside brains), organisms could adapt in other ways.  I think the advantage of brains is that they increase your options, specifically by letting you choose and execute complex sequences of muscular responses to situations in a relatively cheap and sensitive way, compared to rigging up Rube Goldberg macroscopic-physical-event machines that could execute the same responses.

Having a brain with different groups of neurons that execute different responses, and having certain groups fire in response to certain kinds of situations, seems like a plausibly useful way to organize the brain.  It would mean that, when fine-tuning how group X of neurons responds to situation Y, you don't have to worry about what impacts your changes might have in completely different situations ABC that don't cause group X to fire.

I suspect language was ultimately built on top of the above.  First you have groups of organisms that recognize certain things (i.e. they have certain groups of neurons that fire in response to perceiving something in the range of that thing) and respond in predictable ways; then you have organisms that notice the predictable behavior of other organisms, and develop responses to that; then you have organisms noticing that others are responding to their behavior, and doing certain things for the sole purpose^[1] of signaling others to respond.

Learning plus parent-child stuff might be important here.  If your helpless baby responds (by crying) in different ways to different problems, and you notice this and learn the association, then you can do better at helping your baby.

Anyway, I think that at least the original notion of "a thing that I recognize to be an X" is ultimately derived from "a group of neurons that fire (reasonably reliably) when sensory input from something sufficiently like an X enters the brain".  Originally, the neuronal connections (and the concepts we might say they represented) were probably mostly hardcoded by DNA; later they probably developed a lot of "run-time configuration" (i.e. the DNA lays out processes for having the organism learn things, ranging from "what food looks like" [and having those neurons link into the hardcoded food circuit], through learning to associate mostly-arbitrary "language" tokens to concepts that existing neuron-groups recognize, to having general-purpose hardware for describing and pondering arbitrary new concepts).  But I suspect that the underlying "concept X <--> a group of neurons that fires in response to perceiving something like X, which gates the organism's responses to X" organization principle remains mostly intact.

1. ^{^}

   Anthropomorphic language shorthand for the outputs of evolutionary selection

There seem to be sharp negentropy gradients like cell walls, so I think having things as a structure in our map makes sense as a correspondence to structures out there. Nervous systems (and brains) seem to be loci of intense information processing.