bambamramfan:

nostalgebraist:

bambamramfan:

As you may know, Brandon Sanderson (author of the Mistborn series, among other great work)writes about hard magic, the kind with rules, and soft magic, the kind that’s unpredictable and mysterious. He prefers to write the former and argues that problems should only be solved by magic when the magic makes sense, so that the author doesn’t take the easy way out.

I think this phrasing reveals a lot about what rationalist authors think of romantic fantasy. And it’s at once both very intuitive, and if you think about it, utterly incomprehensible.

What the fuck does “take the easy way out” mean???

Sanderson means “I have set up a hard system so that the heroes have to think to figure out a solution that comes from those rules, instead of fiating a deus ex machina that took me no effort to puzzle out.” And that sure sounds like hard work by him, but how is the opposite easy?

If someone writes an unsatisfactory story where DeM magical solutions are used, and people dislike the ending and the entire set of characters and themes because of it, was that easy. I don’t think failing a math test counts as easy. If you wrote a bad story, that should be assessable from the story itself, and not only from how much effort the author used.

And if people do like the end? Is that a worse story than the hard systems? “What the hell she could have used her magic shoes to go home at any time? This whole ruby slippers and there’s no place like home is MAIL FRAUD.” Like, no, it’s still pretty good. There was real effort there by the hero (the emotional journey to reveal what you truly want) even if the system was metaphorical as hell.

I think the quote actually makes a lot of sense on a slightly different reading.  As I understand it, “taking the easy way out” is a description of an event in the writing process, not a property of the finished work.  It’s not that deus ex machina solutions never “work” in the finished product; instead, Sanderson and Moseman are pointing to a dynamic in the writing process which tends to produce the kind of deus ex machina solutions that don’t work.

Specifically, I think the point makes the most sense if we take “solving problems by magic” to mean solving the writer’s problems by magic: running into some plotting snafu with no obvious resolution that works in terms of character and theme and all that, and forcibly undoing the snafu with in-universe magic.  Sometimes, such a solution will “works in terms of character and theme and all that,” but the point is that it will be appealing whether or not it does.  Some di ex machina are there because a deus ex machina was really the right thing for the story at that moment, but many are there just because the writer wanted to resolve a plotting snafu.  The latter type can work artistically, but only by happy accident.

You can’t look at a story and say definitively whether the writer “took the easy way out” with some decision, but you can look at a story and say “this deus ex machina doesn’t work artistically.”  And if you ask why the author chose to put such a thing in their story, often the answer will be that they introduced it to solve a plotting problem without doing the hard work necessary to invent a solution that coheres artistically with the rest of the story.  That is, they took the easy way out.

So under your explanation, the model of writing is “I have brought the characters to a certain point. I know at the end of the story they need to be somewhere else. So I have a problem to solve, of how to get them from A to B.” And the low-effort way to solve this (a wizard appears and does it!) is “taking the easy way out.” Whereas a more praiseworthy author would have a consistent system in place or have figured out some in universe trick.

So serious question: does anyone do this? Not just the easy way out, but even that entire conception of writing. Like in The Northern Caves, did you know where you wanted to end up, but just didn’t know how you’d get there? Because TNC doesn’t read in the slightest like that.

This is my personal experience with writing, and of all the writers I have observed. The theme of the work works with the characters to take the narrative in certain directions, and that’s just where everything ends up. Or maybe you knew the end from the get go, so everything the character does has been like an arrow leading up to this point. But there’s never been a… puzzle to solve, of how to get this character to cross the final gap, whereby my choices are “magic!” or “hard work of coming up with a logical system.”

Like I really don’t think Frank Baum got 90% of the way through Oz and said “shit, how is Dorothy getting home.” It’s not a matter of how he answered that question, but I don’t think the question even came up. But, maybe other authors will say “yes, that is a gap we must frequently solve.”

In which case, a more interesting differentiation is not between authors who use hard versus easy methods of solving these “plotting problems” but authors who write such that a plot problem even comes up, versus those who don’t.

Yes, my personal experience with plotting is like what I describe (I am an example of someone who “writes such that plot problems come up”).  Plotting TNC was a giant headache at times, as was plotting Floornight.

I don’t have any ready examples at hand, but I feel like writers talking about “plotting problems” is a fairly common thing – like, on writing advice sites, or blogs of writers who talk a lot about their process.

When I am designing a plot, there are always pieces that are more certain and less certain.  I can know exactly what a scene 2/3 of the way through is going to look like, but not how we’ll get there; I can have one character’s story planned in a lot of detail while a second character’s is very fuzzy, even though the two characters are supposed to constantly interact; all sorts of other permutations are possible.

The basic reason that “plotting problems” come up is that this uncertainty doesn’t follow a causal-like pattern, even though narratives are supposed to (usually) obey the rules of causality.  In real life, uncertainty increases as you think about times further and further in the future; in a plot under construction, the past can be hazier than the future, the cause hazier than the effect.  And you can have networks of causation that make sense, so long as certain fuzzy pieces spit out certain effects – event X makes sense thematically and character-wise in all these different ways for everyone/everything involved but it requires character Z to do a particular thing beforehand, and I always sorta assumed character Z was going to do that and never thought about it explicitly, but now I need to find a reason for Z to do that, so that this event that naturally flows from every other piece can still happen … 

I guess I am confused by the alternative experience you describe because it sounds so easy – are you saying that for many (most?) writers, there is never such thing as a “plotting problem”?  That a plot which makes both thematic and causal sense just comes to them naturally, in one fell swoop?

—

David Friedman (via atidd)

How do agricultural subsidies play into this?

(via nostalgebraist)

“Those who raise corn should not be taxed to encourage those who desire to raise beets.The power to tax was never vested in a Government for the purpose of building up one class at the expense of other classes.”

Subsidies protect American farmers - from other American farmers (sugar is one of the most heavily protected products at around 55%; grains are iirc. ~20% and most other products have comparatively little subsidy, ~0-5%). And from American taxpayers who otherwise might have more money to spend on agricultural products. And from producing the things that would be genuinely most in demand. And so on.

(via oktavia-von-gwwcendorff)

What I mean is, supposing that agricultural subsidies stay basically the same while the tariff is implemented, wouldn’t they absorb some of the effects of lower demand for wheat production on individual farmers, where the same is not true for similar effects on individual auto workers?

Like a theory of the second best thing

(via oktavia-von-gwwcendorff-deactiv)

you-have-just-experienced-things:

you-have-just-experienced-things:

nostalgebraist:

A number of people recommended this post to me, and it is indeed good and worth reading.  I say that only partly because it provides evidence that aligns with the preconceptions I already had :P

Specifically, here is what I wrote in this post:

I was thinking about this stuff after I was arguing about deep learning the other day and claimed that the success of CNNs on visual tasks was a special case rather than a generalizable AI triumph, because CNNs were based on the architecture of an unusually feed-forward and well-understood part of the brain – so we’d just copied an unusually copy-able part of nature and gotten natural behavior out of the result, an approach that won’t scale

The gist of Sarah’s post is that in image recognition and speech recognition, deep learning has produced a “discontinuous” advance relative to existing improvement trends (i.e., roughly, the trends we get from using better hardware and more data but not better algorithms) – but in other domains this has not happened.  This is what I would expect if deep learning’s real benefits come mostly from imitating the way the brain does sensory processing, something we understand relatively well compared to “how the brain does X” for other X.

In particular, it’s not clear that AlphaGo has benefitted from any “discontinuous improvement due to deep learning,” above and beyond what one would expect from the amount of hardware it uses (etc.)  If it hasn’t, then a lot of people have been misled by AlphaGo’s successes, coming as they do at a time when deep learning successes in sensory tasks are also being celebrated.

Sarah says that deep learning AI for computer games seems to be learning how to perform well but not learning concepts in the way we do:

The learned agent [playing Pong] performs much better than the hard-coded agent, but moves more jerkily and “randomly” and doesn’t know the law of reflection.  Similarly, the reports of AlphaGo producing “unusual” Go moves are consistent with an agent that can do pattern-recognition over a broader space than humans can, but which doesn’t find the “laws” or “regularities” that humans do.

Perhaps, contrary to the stereotype that contrasts “mechanical” with “outside-the-box” thinking, reinforcement learners can “think outside the box” but can’t find the box?

This is reminiscent of something I said here:

My broad, intuitive sense of these things is that human learning looks a lot like this gradient descent machine learning for relatively “low-level” or “sensorimotor” tasks, but not for abstract concepts.  That is, when I’m playing a game like one of those Atari games, I will indeed improve very slowly over many many tries as I simply pick up the “motor skills” associated with the game, even if I understand the mechanics perfectly; in Breakout, say, I’d instantly see that I’m supposed to get my paddle under the ball when it comes down, but I would only gradually learn to make that happen.

The learning of higher-level “game mechanics,” however, is much more sudden: if there’s a mechanic that doesn’t require dexterity to exploit, I’ll instantly start exploiting it a whole lot the moment I notice it, even within a single round of a game.  (I’m thinking about things like “realizing you can open treasure chests by pressing a certain button in front of them”; after opening my first chest, I don’t need to follow some gradual gradient-descent trajectory to immediately start seeking out and opening all other chests.  Likewise, the abstract mechanics of Breakout are almost instantly clear to me, and my quick learning of the mechanical structure is merely obscured by the fact that I have to learn new motor skills to exploit it.)

It is a bit frustrating to me that current AI research is not very transparent about how much “realizing you can open treasure chests”-type learning is going on.  If we have vast hardware and data resources, and we only care about performance at the end of training, we can afford to train a slow learner that can’t make generalizations like that, but (say) eventually picks up every particular case of the general rule.  I’ve tried to look into the topic of AI research on concept formation, and there is a lot out there about it, but a lot of it is old (like, 1990s or older) and it doesn’t seem to the focus of intensive current research.

---

It’s possible to put a very pessimistic spin on the success of deep learning, given the historically abysmal performance of AI relative to expectations and hopes.  The pessimistic story would go as follows.  With CNNs, we really did find “the right way” to perform a task that human (and some animal) brains can perform.  We did this by designing algorithms to imitate key features of the actual brain architecture, and we were able to do that because the relevant architecture is unusually easy to study and understand – in large part because it is relatively well described by a set of successive “stages” with relatively little feedback.

In the general case, however, feedback is a major difference between human engineering designs and biological system “design.”  Biological systems tend to be full of feedback (not just in the architecture of the nervous system – also in e.g. biochemical pathways).  Human engineers do make use of feedback, but generally it is much easier for humans to think about a process if it looks like a sequence of composed functions: “A inputs to B, which inputs to C and D, which both input to E, etc.”  We find it very helpful to be able to think about what one “part” does in (near-)isolation, where in a very interconnected system this may not even be a well-defined notion.

Historically, human-engineered AI has rarely been able to match human/biological performance.  With CNNs, we have a special case in which the design of the biological system is unusually close to something humans might engineer; hence we could reverse engineer it and get atypically good AI performance out of the result.

But (I think; citation needed!) the parts of the brain responsible for “higher” intelligence functions like concept formation are much more full of feedback and much harder to reverse engineer.  And current AI is not any good at them.  If there are ways to do these things without emulating biology, many decades of AI research has not found them; but (citation needed again) we are no closer to knowing how to emulate biology here than we were decades ago.

It is a bit frustrating to me that current AI research is not very transparent about how much “realizing you can open treasure chests”-type learning is going on.

This seems like a problem with the current paradigm of throwing really big ANNs at things. Many-layered neural networks don’t look as attractive when it is important to understand why the model made a certain decision (e.g., when making algorithmic trades).

Currently, unless I am deeply mistaken about the state of the art, answering the “does the model know it can open treasure chests” question is not easy. (Then again, it was only easy in the previous paradigm of SVM, random forests, etc, because those models really aren’t that sophisticated in comparison.) This makes me somewhat skeptical of generalizations like this:

Perhaps, contrary to the stereotype that contrasts “mechanical” with “outside-the-box” thinking, reinforcement learners can “think outside the box” but can’t find the box?

At face value, this is probably true. But it’s not clear to me (at this specific point in the research process) that we would be able to recognize that “conceptual learning” is happening in a given ANN even if it was actually happening. I am hoping that we can just solve this by simply getting better at understanding why big many-layered networks behave in certain ways.

(I was trained in the previous paradigm. The best thing I can say about the people I worked with is that they were extreme sticklers about understanding exactly how every model worked, always. Nowadays, my coworkers use a GPU cluster + massive neural nets to extract tables from PDF’s without caring about such things.)

@nostalgebraist​:  

@you-have-just-experienced-things: what you say seems true in the general case, but simple video games are a particular case where it’s easy to make the judgment, at least intuitively – so it bothers me that people /are/ throwing ANNs at these games but /not/ treating them as a test bed for this question.

(cont.) both sarah and i had to look at supplemental videos of gameplay and try to answer these questions for ourselves; where are the papers that put the questions front and center?  researchers can interpret gameplay videos just like anyone else, after all

I generally agree with this. And I strongly endorse the “Where are the papers?” question, especially because it seems like it can be answered by understanding the trends/incentives of the deep learning folks doing the research.

Re the Pong video: I don’t agree with Sarah’s interpretation–it doesn’t look like the right paddle ever stops moving, which suggests the model was created with only two output states (Up/Down) rather than Up/Down/Stop. This could explain the jiggling as the ball gets closer to the paddle.

This is kind of why I am wary of making these interpretations based on the observed output, even in these simple cases. There is a lot of nuance to how these networks are trained that is not necessarily obvious from looking at gameplay videos. What I really want is the kinds of analysis you can do on a much simpler model, where (once you understood the math) there is no ambiguity about why a classifier was making a certain prediction. I acknowledge that this is way too high of a standard, though.

(via you-have-just-experienced-things)

you-have-just-experienced-things:

nostalgebraist:

evolution-is-just-a-theorem:

ilzolende:

katiewompus:

I might as well have asked her ‘Can you tell me how to reticulate splines using a hexagonal decode system so that I can build a GUI in Visual Basic and track an IP Address.’

Here’s an idea. When they hit eleven, give them a plaintext file with ten-thousand WPA2 keys and tell them that the real one is in there somewhere. See how quickly they discover Python or Bash then.

Quality lines in this article.

I feel Called Out, except I actually have done things like upgrade memory. But I still would need to ask for help or use Atom if I wanted to, say, search for a string in a folder of text files…

First:

• Find all instances of string: grep -r “$string” path/to/folder
  
• Find all instances of string in text files only: find . -name “*.txt” | grep -r “$string” (many variants of this exist depending on exactly what you want)

Second (warning for tech elitism and me being really rather mean):

Keep reading

@evolution-is-just-a-theorem​ you sure know how to write an entertaining takedown <3

The original blog post is so bad, and so comformant with the worst of Annoying Tech Guy stereotypes, that I almost can’t believe it’s sincere – but that’s my bias towards the nicer side of Poe’s law speaking, and the guy has a web presence (on Github etc.) that’s consistent with his self-presentation in the post.

There are so many weird little things to grumble about – like, those examples of people who “don’t know how to use a computer”?  A lot of those sound similar to the phenomenon where one person spends hours hunting for a bug and then someone else reviews their code and notices it quickly, because they weren’t focused on the same details.  Yeah, the guy whose ethernet cable was unplugged looks “dumb” in retrospect, but it’s not like ethernet cables are constantly getting unplugged by gremlins in everyday life; everyone comes in to a problem with some base assumptions and “no one messed with the hardware because why would they” is pretty good as base assumptions go.

(There’s another point which is still vague in my mind, but runs something like this: computer technology is important enough that we should educate people about it, but not necessarily by forcing them to confront the inner workings of that technology when they’re trying to use it for other things.  It might be good to know how the plumbing of my house works, but I’m glad I don’t usually have to think about the plumbing at all when I just want some tap water.  And the knowledge that makes someone a good IT computer-fixer != the knowledge needed to understand the big-important-scary stuff happening with computers in the world, necessarily.  Does my plumber understand water scarcity issues any better than I do?)

The original article is so bad that it’s pointless to respond to it directly. I think I understand what the OP is trying to argue against, he just does it in the worst possible way. Here is an attempted steelman/reinterpretation:

1) There is a pretty flawed narrative out there about digital natives and how technology comes naturally to them. This is sometimes taken to absurd extremes. (My aunt even told me once that I should be worried about my job security as a programmer, because she read an article about how all these “digital natives” are so inherently good with computers that they will make me obsolete.) I don’t think this narrative is particularly important, but it is observably wrong.

2) When you work in IT support, you see the worst of humanity. I only did this for 18 months in college, but for those 18 months it was a miserable experience. IT support is a service job, and most people treat service workers horribly–but something about dealing with technology is uniquely frustrating for many people, and they are more than willing to unleash that frustration on whoever is helping them. It’s difficult to describe how horrible it can feel to be on the receiving end of this unless you’ve experienced it. It’s regrettable that the OP used this essay to complain about his job, because that emotion really needs to be kept separate from the argument.

3) People like treating things as magic, and technology especially so. Learned helplessness abounds. This is also one of the reasons why IT is so soul-crushing; @evolution-is-just-a-theorem seems to agree here. (Incidentally, I feel pretty put off by their critique of the original article, for reasons I can’t articulate. Might just be general uneasiness about snark.) Many people would benefit from investing just a little bit of time into basic IT skills like “how to google for an error message.” There is, of course, a point of diminishing returns to learning how to be good at debugging computer issues, but most people are nowhere near that point yet.

4) Anecdotally, I observe a strong correlation between programming knowledge and being concerned about big-picture technology issues, specifically government surveillance. I don’t have a good explanation for this, and I don’t know if I’m simply looking at a biased sample. Nor am I confident that the causation doesn’t go the other way. But it seems like understanding how things work on a slightly deeper level does lead people to think more deeply about the big-picture stuff. (I would definitely trust my plumber’s opinion about water scarcity issues above my own.)

I agree with most of 1-4 above, when phrased this way. And I think if you strip away all the terrible stuff from the original article, this stuff is there beneath the surface. The OP is just obnoxious.

(via you-have-just-experienced-things)