A Parable on Visual Impact

A long time ago, you could get the biggest positive visual impact for your money by generating art, and if you wanted awe you could fund gardens and cathedrals. And lo, these areas were well-funded!

The printing press arrived. Now, you could get massive numbers of pamphlets and woodcuts for a fraction of the price of paintings. And lo, these areas were well-funded!

Then tv appeared. Now, if you wanted the greatest awe and the biggest positive visual impact for your money, you crafted something suitable for the new medium. And in that time, there were massive made-for-tv propaganda campaigns, and money poured into developing spacecraft, and we got our first awe-inspiring images of Earth from the moon.

Some even claim the Soviet Union was defeated by the view through a television screen of a better life in America.

And then we developed CGI and 3D MMORPGs. And lo, the space program defunded, as people built entire cities, entire planets in CGI for a tiny fraction of the cost!

Thinking about value-drift:

It seems correct to me to term the earlier set of values a more "naive set.*" To keep a semi-consistent emotional tone across classes, I'll call the values you get later the "jaded set" (although I think jadedness is only one of several possible later-life attractor states, which I'll go over later).

I believe it's unreasonable to assume that any person's set of naive values are, or were, perfect values. But it's worth noting that there are several useful properties that you can generally associate with them, which I'll list below.

My internal model of how value drift has worked within myself looks a lot more like "updating payoff matrices" and "re-binning things." Its direction feels determined by not purely by drift, but by some weird mix of deliberate steering, getting dragged by currents, random walk, and more accurately mapping the surrounding environment.

That said, here's some generalizations...

Naive:

• Often oversimplified
  • This is useful in generating ideals, but bad for murphyjitsu
• Some zones are sparsely-populated
  • A single strong value (positive or negative) anywhere near those zones will tend to color the assumed value of a large nearby area. Sometimes it does this incorrectly.
• Generally stronger emotional affects (utility assignments), and larger step sizes
  • Fast, large-step, more likely to see erratic changes
• Closer to black-and-white thinking; some zones have intense positive or negative associations, and if you perform nearest neighbor on sparse data, the cutoff can be steep

Jaded:

• Usually, your direct experience of reality has been fed more data and is more useful.
  • Things might shift so the experientially-determined system plays a larger role determining what actions you take relative to theory
    • Experiential ~ Sys1, Theoretical ~ Sys2, but the match isn't perfect. The former handles the bulk of the data right as it comes in, the later sees heavily-filtered useful stuff and makes cleaner models.**
• Speaking generally... while a select set of things may be represented in a more complete and complex way (ex: aspects of jobs, some matters of practical and realistic politics, most of the things we call "wisdom"), other axes of value/development/learning may have gotten stuck.
  • Here are names for some of those sticky states: jaded, bitter, stuck, and conformist. See below for more detail.
  • Related: Freedomspotting

** What are some problems with each? Theoretical can be bad at speed and time-discounting, while Experimental is bad at noticing the possibility of dangerous unknown-unknowns?

Jaded modes in more detail

Here are several standard archetypical jaded/stuck error-modes. I tend to think about this in terms of neural nets and markov chains, so I also put down the difficult-learning-terrain structures that I tend to associate with them.

Jaded: Flat low-affect zones offer no discernible learning slope. If you have lost the ability to assign different values to different states, that spells Game Over for learning along that axis.

Bitter: Overlearned/overgeneralized negatives are assigned to what would otherwise have been inferred to be a positive action by naive theory. Thenceforth, you are unable to collect data from that zone due to the strong expectations of intense negative-utility. Keep in mind that this is not always wrong. But when it is wrong, this can be massively frustrating.

(Related: The problems with reasoning about counterfactual actions as an embedded agent with a self-model [LW · GW])

Stuck: Occupying an eccentric zones on the wrong side of an uncanny valley. One common example is people who specialized into a dead field, and now refuse to switch.

Conformist: Situated in some strongly self-reinforcing zone, often at the peak of some locally-maximal hill. If the hill has steep moat-like drop-off, this is a common subtype of Stuck. In some sense, hill-climbing is just correct neural net behavior; this is only a problem because you might be missing a better global maximum, or might have different endorsed values than your felt ones.

An added element that can make this particularly problematic is that if you stay on a steep, tiny hill for long enough, the gains from hill-climbing can set up a positive incentive for learning small step size. If you believe that different hills have radically different maximums, and you aspire for global maximization, then this is something you very strongly want to avoid.

*From context, I'm inferring that we're discussing not completely naive values, but something closer to the values people have in their late-teens early-twenties.

Naive modes in more detail

Naive values are often simplified values; younger people have many zones of sparse or merely theorized data, and are more prone to black-and-white thinking over large areas . They are also perhaps notable sometimes for their lack of stability, and large step-size (hence the phrase "it's just a phase"). Someone in their early 20s is less likely than someone in their 40s to be in a stable holding pattern. There's been less time to get stuck in a self-perpetuating rut, and your environment and mind are often changing fast enough that any holding patterns will often get shaken up when the outside environment alters if you give it enough time.

The pattern often does seem to be towards jaded (flat low-affect zones that offer no discernible learning slope), bitter (overlearned/overgeneralized negatives assigned to what would otherwise have been inferred to be a positive action), stuck (occupying eccentric zones on the wrong side of an uncanny valley), or conformist (strongly self-reinforcing zones, often at the peak of some locally-maximal hill).

Many people I know have at east one internal-measurement instrument that is so far skewed that you may as well consider it broken. Take, for instance, people who never believe that other people like them despite all evidence to the contrary. This is not always a given, but some lucky jaded people will have identified some of the zones where they cannot trust their own internal measuring devices, and may even be capable of having a bit of perspective about that in ways that are far less common among the naive. They might also have the epistemic humility to have a designated friend filling in that particular gap in their perception, something that requires trust built over time.

I stumbled into a satisfying visual description:

3D "Renderers" of Random Sampling/Monte Carlo estimation

This came out of thinking about explaining sparseness in higher dimensions*, and I started getting into visuals (as I am known to do). I've see most of the things below described in 2D (sometimes incredibly well!), but 3D is a tad trickier, while not being nearly as painfully-unintuitive as 4D. Here's what I came up with!

Also, someone's probably already done this, but... shrug?

The Setup

In a simple case of Monte Carlo/random-sampling from a three-dimensional space with a one-dimensional binary or float output...

You're trying to work out what is in a sealed box using only a thin needle (analogy for the function that produces your results). The needle has a limited number of draws, but you can aim the needle really well. You can use it to find out what color the object in the box is at each of the exact points you decided to measure. You also have an open box in which to mark off the exact spot in space at which you did the draw, using Unobtainium to keep it in place.

Each of your samples is a fixed point in space, with a color set by the needle(function/sampler's) output/result. These are going to be the cores that you base your render around. From there, you can...

Nearest Neighbor Render

For Nearest Neighbor (NN), you can see what happens when you situate colored-or-clear balloons at each of the sample points, and start simultaneously blowing up all of them up with their sample point as a loose center. Gradually, this will give you a bad "rendering" of whatever the color-results were describing.

(Technically it's Voronoi regions; balloons are more evocative)

Decision Trees

Decision trees are when you fill in the space using colored rectangular blocks of varying sizes, based on the sample points.

Gradients

Most other methods involve creating a gradient between 2 different points that yielded different colors. It could be a linear gradient, an exponential gradient, a step function, a sine wave, some other complicated gradient... as you get more complicated, you better hope that your priors were really good!

Neural Net

A human with lots of prior experience looking at similar objects tries to infer what the rest looks like. (This is kinda a troll answer, I know. Thinking about it.)

Sample All The Things!

3D Pointillism even just sounds like a really bad idea. Pointillism on a 2D surface in a 3D space, not as bad of an idea! Sparsity at work?

Bonus

The Blind Men and an Elephant fable is an example of why you don't want to do a render from a single datapoint, or else you might decide that the entire world is "pure rope." If they'd known from sound where each other were located when they touched the elephant, and had listened to what the others were observing, they might have actually Nearest Neighbor'd themselves a pretty decent approximate elephant.

*Specifically, how it's dramatically more effort to get a high-resolution pixel-by-pixel picture of a high-dimensional space. But if you're willing to sacrifice accuracy for faster coverage, in high dimensions a single point's Nearest Neighbor zone can sloppily cover a whole lot of ground (or multidimensional "volume").

Out-of-context quote: I'd like for things I do to be either ignored or deeply-understood and I don't know what to do with the in-betweens.

Give a man a fish, feed him for a day. Teach a man to fish, feed him for a lifetime.

Cultivate someone with an earnest and heartfelt interest in fishing, until he learns how to grow his own skills and teach himself... and you can probably feed at least 12 people.

Which just might finally allow the other 11 to specialize into something more appealing to them than subsistence fishing.

On Community Coordinator Roles

Some things I think help are:

• Personal fit to the particular job demands (which can include some subset of: lots of people-time, navigating conflict, frequent task-switching, personal initiative, weighing values (or performing triage), etc.)
• Personal investment in appropriate values, and a lasting commitment to doing the role well.
• Having it be fairly clear what they're biting off, what resources they can use, and what additional resources they can petition for. Having these things be actually fairly realistic.

(Minimum expertise to thrive is probably set by needing to find someone who will advocate for these things well enough to make them clear, or who can at least learn how to. Nebulous roles can be navigated, but it is harder, and in practice means they need to be able to survive redefining the role for themselves several times.)

• Having some connections they (and others) respect who will help support what (and how) they're doing, lend some consistent sense of meaning to the work, and people who will offer reliable outside judgement. Quite possibly, these things come from different people.

(Public opinion is often too noisy and fickle for most human brains to learn off of it alone.)

• For roles that require the buy-in of others, it's helpful if they have enough general respect or backing that they will mostly be treated as "legitimate" in the role. But some roles really only require a small subset of high-buy-in people.
• Also, some sort of step-down procedure.

Extracted from a FB thread, where I was thinking about burnout around nebulously-defined community leader roles, and what preparedness and a good role would look like. Parts of this answer feel like I'm being too vague and obvious, but I thought it was worth making the list slightly more findable.

After seeing a particularly overt form of morality-of-proximity argued (poorly) in a book...

Something interesting is going on in my head in forming a better understanding of what the best steelman of morality-of-proximity looks like? The standard forms are obviously super-broken (there's a lot of good reasons why EA partially builds itself as a strong reaction against that; a lot of us cringe at "local is better" charity speak unless it gets tied into "capacity building"). But I'm noticing that if I squint, the intuition bears a trace of similarity to my attitude on "What if we're part of an "all possible worlds" multiverse, in which the best and worst and every in-between happens, and where on a multiversal scale... anything you do doesn't matter, because deterministically all spaces of possibility will be inevitably filled."

Obviously, the response is: "Then try to have a positive influence on the proximal. Try to have a positive influence on the things you can control." And sure, part of that comes from giving some weight to the possibility that the universe is not wired that way, but it's also just a very broadly-useful heuristic to adopt if you want to do useful things in the world. "Focus on acting on the things you can control" is generally pretty sound advice, and I think it forms a core part of the (Stoic?) concept of rating your morals on the basis of your personal actions and not the noisy environmental feedback you're getting in reply. Which is... at least half-right, but runs the risk of disconnecting you from the rest of the world too much.

If we actually can have the largest positive impact on the proximal, then absolutely help the proximal. This intuition just breaks down in a world where money is incredibly liquid and transportable and is worth dramatically more in value/utility in different places.

(Of course, another element of that POV might center around a stronger sense that sys1 empathetic feelings are a terminal good in-and-of-itself, which is something I can't possibly agree with. I've seen more than enough of emotional contagion and epidemics-of-stress for a lifetime. Some empathy is useful and good, but too much sounds awful. Moving on.)

I guess there needs to be a distinction made between "quantity of power" and "finesse of power," and while quantity clearly can be put to a lot more use if you're willing to cross state (country, continent...) lines, finesse at a distance is still hard. And you need at least some of both (at this point) to make a lasting positive impact (convo for later: what those important components of finesse are, which I think might trigger some interesting disagreements). Finding specialists you trust to handle the finesse (read: steering) of large amounts of power is often hard, and it's even harder to judge for people who don't have that finesse themselves.

And "Finesse of power" re: technical impact, re: social impact, re: group empowerment/disempowerment (whether by choice or de-facto), re: PR or symbology... they can be really different skills.

One of my favorite little tidbits from working on this post [LW · GW]: realizing that idea innoculation and the Streisand effect are opposite sides of the same heuristic.

While I could rattle off the benefits of "delegating" or "ops people", I don't think I've seen a highly-discrete TAP + flowchart for realizing when you're at the point where you should ask yourself "Have my easy-to-delegate annoyances added up to enough that I should hire a full-time ops person? (or more)."

Many people whose time is valuable seem likely to put off making this call until they reach the point where it's glaringly obvious. Proposing an easy TAP-like decision-boundary seems like a potentially high-value post? Not my area of specialty, though.

If The Ocean was a DM:

Player: I want to play a brainless incompletely-multiceullular glob with no separation between its anus and its mouth and who might have an ability like "glow in the dark" or "eject organs and then grow them back." It eats tiny rocks in the water column.

The Land DM: No! At least have some sensible distribution-of-nutrients gimmick! Like, you're not even trying! Like, I'm not anti-slimes, I'll allow some of this for the Slime Mold, but they had a clever mechanic for succinctly generating wider pathways for nutrient transfer between high-nutrition areas that I'd be happy to...

The Ocean DM: OMG THAT'S MY FAVORITE THING. Do you have any other character concepts? Like, maybe it can also eject its brain and grow it back? Or maybe its gonads are also in the single hole where the mouth and anus are?

The Land DM: ...

Live Parsers and Quines

An unusual mathematical-leaning definition of a living thing.

(Or, to be more precise... a potential living immortal? A replicon? Whatever.)

A self-replicating physical entity with...

3 Core Components:

• Quine: Contains the code to produce itself
• Parser: A parser for that code
  • Code includes instructions for the parser; ideally, compressed instructions
• Power: Actively running (probably on some substrate)
  • Not actively running is death, although for some the death is temporary.
  • Access to resources may be a subcomponent of this?

Additional components:

• Substrate: A material that can be converted into more self
  • Possibly a multi-step process
  • Access to resources is probably a component of this
• Translator: Converts quine into power, or vice-versa
  • Not always necessary; sometimes a quine is held as power, not substrate

Parser and Code: the information can actually be stored on either; you can extract the correct complex signal from complete randomness using an arbitrarily-complex parser chosen for that purpose. There are analogies that can be drawn in the other direction, too: a fairly dumb parser can make fairly complicated things, given enough instructions. (Something something Turing Machines)

Ideally, though, they're well-paired and the compression method is something sensible, to reduce complexity.

A quine by itself has only a partial life; it is just its own code, it requires a parser to replicate.

(If you allow for arbitrarily complex parsers, any code could be a quine, if you were willing to search for the right parser.)

Compilers are... parsers? (Or translators?)

It is possible for what was "code" information to be embedded into the parser. I think this is part of what happens when you completely integrate parts in IFS.

I'm less clear on what's important about power? Things to poke: Power, Electicity, Energy (& time-as-the-entropy-direction?). Animating. Continuity, Boltzmann brains, and "Is it frozen/discontinuous relative to this frame of time/reference?" (Permutation City). What is Death? (vs Life, or vs a frozen/cryonic quine)

Examples

Example replicons: A bacterium, a cell, a clonal organism, a pair of opposite-sexed humans (but not a single), self-contained repeating Game of Life automata, the eventual goal of RepRap (a 3D printer fully producing a copy of itself)

Viruses: Sometimes quines, sometimes pre-quine and translator

A scholar is just a library's way of making another library.

-- Daniel Dennett

The distinctive thing about Lisp is that its core is a language defined by writing an interpreter in itself. It wasn't originally intended as a programming language in the ordinary sense. It was meant to be a formal model of computation, an alternative to the Turing machine. If you want to write an interpreter for a language in itself, what's the minimum set of predefined operators you need? The Lisp that John McCarthy invented, or more accurately discovered, is an answer to that question.

-- What I Worked On, Paul Graham

Related: Self-Reference, Post-Irony, Hofstadter's Strange Loop, Turing Machine, Automata (less-so), deconstruction (less-so)

Is Tiddlywiki a quine?

Is a cryonics'd human a quine? (wrt future technology)

The definition of parasite used in Nothing in evolution makes sense except in the light of parasites is literally quines.

(Complex multi-step life-cycle in the T5 case, though?. The quine produced a bare-bones quine-compiling "spawner" when it interacts with the wild-type, and then replicates on that. Almost analogous to CS viruses that infect the compiler, and any future compiler compiled by that compiler.)

Process is Art: Is art that demonstrates how to craft the art, a quine on a human compiler?

Some connection to: Compartmentalization (Black Box Testing, unit tests, separation of software components) and the "swappability" of keeping things general and non-embedded, with clearly-marked and limited interface-surfaces (APIs). Generates pressure against having "arbitrary compilers" in practice.

Yay, shortform feeds!

A copy machine copies a 5 minute video over time, going line by line, one line per second. Every pixel it sees, it copies perfectly. Is it an accurate copy?

A teacher teaches a student every thing the teacher knows, building up from basic to advanced knowledge over the course of years. At each weekly lesson, the student learns the planned material perfectly. That exact topic is never revisited again.

By the end of the last class, does the student know and understand everything the teacher knows?

OODAs vs TAPs

TAPs [LW · GW] are just abbreviated or stunted OODA loops, and I think I prefer some aspects of the OODA phrasing.

OODA: Observe -> Orient -> Decide -> Act (loop)

TAP: Trigger -> Action (Plan/Pattern)

TAPs usually map to "Observe->Act."

Orient loosely means setting up a Model/Metrics/Framework/Attitude.

(This can take anywhere from 1/2 a second to years; wildly variable amounts of time depending on how simple and familiar the problem and system is. TAPs implicitly assume this is one of the simple/fast cases.)

One case where OODA's phrasing is better, is "Attitude TAPs":

"Attitude TAPs" are one of the most valuable TAP types, but are a non-obvious feature of the TAPs framework unless this is explicitly mentioned.

Ex: (from here [LW · GW])"when I'm finding my daily writing difficult, and I'm thinking about quitting, I'll notice that and try to figure out what's going wrong (rather than mindlessly checking Twitter to avoid writing)."

Alternatively, this could be phrased as an abbreviated OODA that cements an "Observe ->Orient" without fixing an action. Personally, I find this a bit cleaner.

My nCOV/SARS-CoV-2 research head until I figure out exactly where to post things.

Things to post next: Link and summary of the research into the nCOV spike-protein mutations.

Bubbles in Thingspace

It occurred to me recently that, by analogy with ML, definitions might occasionally be more like "boundaries and scoring-algorithms in thingspace" than clusters per-say (messier! no central example! no guaranteed contiguity!). Given the need to coordinate around definitions, most of them are going to have a simple and somewhat-meaningful center... but for some words, I suspect there are dislocated "bubbles" and oddly-shaped "smears" that use the same word for a completely different concept.

Homophones are one of the clearest examples; totally disconnected bubbles of substance.

Another example is when a word covers all cases except those where a different word applies better; in that case, you can expect to see a "bite" taken out of its space, or even a multidimensional empty bubble, or a doughnut-like gap in the definition. If the hole is centered ("the strongest cases go by a different term" actually seems like a very common phenomenon), it even makes the idea of a "central" definition rather meaningless, unless you're willing to fuse or switch terms.

This is pasted in from a very different context, so it doesn't match how I would format it as a post at all. But it paints a sketch of a system I might want to flesh out into a post sometime (although possibly not on LessWrong): Predictiveness vs Alignment as sometimes-diverging axes of Identity.

There's probably some very interesting things to be said about the points where external predictiveness and internal alignment differ, and I might be a bit odd in that I think I have strong reason to believe that a divergence between them is usually not a bad thing. In fact, at times, I'd call it a legitimately good thing, especially in cases where it feeds into motivation or counterbalances other forces.

Ex: For me in particular, anti-identifying as "sensitive" is load-bearing, and props up my ability to sometimes function as a less-than-excessively-sensitive human being. However, the sensitive aspect is always going to show through in the contexts where there's no action I can take to cover for it (ex: very easily disturbed by loud noise).

There's still some question in my head about what internal alignment even is. It feels to me like "Internal Predictiveness" is a subcomponent of it, and aspiration is another part of it, but neither describes it in full. But it felt like there was a meaningful cluster there, and it was easier to feel-out and measure than it was to describe, so for now I've stuck with that...

Q: What are some identity labels that apply to you? How aligned with or alienated from them do you feel?

I was only able to do this exercise if I separated out "Externally Predictive" (P-, NP, and P+) and "Internal Alignment," (AN-, N/A, AG+, and sometimes 404). These sometimes feel like totally divergent axes, and trying to catch both in a single axis felt deeply dishonest to me. Besides, finding the divergence between them is fun.

(I apologize for the sparse formatting on this; I'll try to find a way to reformat it better later.)

Predictive-First Identities

• Nerd & Geek
  • Predictive (P+), Aligned (AG)
• Rationalist
  • P+, AG+
• Biologist
  • P+, full range from AG+ to bitterly AN- (Alienated)
• Programmer
  • Moderately P+, N/A
• Extroverted Geek
  • P+, Highly AN
• Introvert
  • Slightly P- (used to be P+), AG
• Anxious
  • Slightly P+, Slightly AN

Alignment/Alienation-First Identities

• Healthy Boundaries
  • P+, AG+
• Visual Thinker
  • NP to P+, AG+
• Queer
  • NP (by design!), AG
• Smart
  • P+, 404 (I can't properly internalize it in a healthy way unless I add more nuance to it)
• Not Smart/Good Enough but Getting Better by Doing it
  • NP to Moderately P+, AG+
• Sensitive
  • P+, AN
• Immature
  • NP (used to be P-), AN
• Self-centered
  • P-, AN-

I do think some things are actually quite real and grounded? Everything is shoved through a lens as you perceive it, but not all lenses are incredibly warping.

If you're willing to work pretty close to the lower-levels of perception, and be quite careful while building things up, well and deeply-grounded shit EXISTS.

To give an evocative, and quite literally illustrative, example?

I think learning how to see the world well enough to do realistic painting is an exceptionally unwarping and grounding skill.

Any other method of seeing while drawing, doubles up on your attentional biases and lets you see the warped result*. When you view it, you re-apply your lens to your lens' results, and see the square any warping you were doing.

It's no coincidence that most people who try take one look at their first attempt at realistic drawing, will cringe and go "that's obviously wrong..."

When you can finally produce an illustration that isn't "obviously wrong," it stands as a piece of concrete evidence that you've learned some ability to engage at-will with your visual-perception, in a way that is relatively non-warping.

Or, to math-phrase it badly...

$$\forall (x,y)\text{in}{R}^2,f(x_i,y_j)\ast b^2\approx f(x_i,y_j)\ast b$$

$$\therefore f(x_i,y_j)\ast b\approx f(x_i,y_j)\text{and}b\approx 1$$

* Taking as a totally-unreasonable given, that your "skill at drawing" is good enough to not get in the way.