Gram_Stone

• Gram_StoneMar 1, 2017, 5:02 AM

  0 points

  in reply to: tristanm’s comment on: Right­ing a Wrong Question

  Can you break that down to the extent that I broke down my confusion above? I’m having a hard time seeing deep similarities between these problems.

• Gram_StoneMar 1, 2017, 4:46 AM

  11 points

  on: Ra­tion­al­ity Quotes Jan­uary—March 2017

  This is rather as if you imagine a puddle waking up one morning and thinking, ‘This is an interesting world I find myself in, an interesting hole I find myself in, fits me rather neatly, doesn’t it? In fact, it fits me staggeringly well, must have been made to have me in it!’ This is such a powerful idea that as the sun rises in the sky and the air heats up and as, gradually, the puddle gets smaller and smaller, it’s still frantically hanging on to the notion that everything’s going to be all right, because this World was meant to have him in it, was built to have him in it; so the moment he disappears catches him rather by surprise. I think this may be something we need to be on the watch out for.

  Douglas Adams, The Salmon of Doubt

• Gram_StoneFeb 28, 2017, 6:00 PM

  0 points

  on: Right­ing a Wrong Question

  “Why was I born as myself rather than someone else?” versus “Why do I think I was born as myself rather than someone else?”

  This never got solved in the comments.

  I was sitting in microeconomics class in twelfth grade when I asked myself, “Why am I me? Why am I not Kelsey or David or who-have-you?” Then I remembered that there are no souls, that ‘I’ was a product of my brain, and thus that the existence of my mind necessitates the existence of my body (or something that serves a similar function). Seeing the contradiction, I concluded that I had reasoned, incoherently, as if ‘I’ were an ontologically fundamental mental entity with some probability of finding itself living some particular lives. That’s unsurprising, because as a great deal of cognitive science and Eliezer’s free will solution has demonstrated, humans often intuitively evaluate ‘possibility’ and plausibility by evaluating how easy it is to conceive of something, as a proxy. “I can conceive of ‘being someone else,’ thus there must be some probability that I ‘could have been someone else’, so what is the distribution, what is its origin, and what probability does it assign to me being me?”

• Gram_StoneFeb 24, 2017, 12:40 PM

  6 points

  in reply to: Stuart_Armstrong’s comment on: Near­est un­blocked strat­egy ver­sus learn­ing patches

  The quote is from this article, section 4.1. There might be other descriptions elsewhere, Lenat himself cites some documents released by the organization hosting the wargame. You might want to check out the other articles in the ‘Nature of Heuristics’ series too. I think there are free pdfs for all of them on Google Scholar.

• Gram_StoneFeb 23, 2017, 3:02 PM

  15 points

  on: Near­est un­blocked strat­egy ver­sus learn­ing patches

  Recently in the LW Facebook group, I shared a real-world example of an AI being patched and finding a nearby unblocked strategy several times. Maybe you can use it one day. This example is about Douglas Lenat’s Eurisko and the strategies it generated in a naval wargame. In this case, the ‘patch’ was a rules change. For some context, R7 is the name of one of Eurisko’s heuristics:

  A second use of R7 in the naval design task, one which also inspired a rules change, was in regard to the fuel tenders for the fleet. The constraints specified a minimum fractional tonnage which had to be held back, away from battle, in ships serving as fuel tenders. R7 caused us to consider using warships for that purpose, and indeed that proved a useful decision: whenever some front-line ships were moderately (but not totally) damaged, they traded places with the tenders in the rear lines. This maneuver was explicitly permitted in the rules, but no one had ever employed it except in desperation near the end of a nearly-stalemated battle, when little besides tenders were left intact. Due to the unintuitive and undesirable power of this design, the tournament directors altered the rules so that in 1982 and succeeding years the act of ‘trading places’ is not so instantaneous. The rules modifications introduced more new synergies (loopholes) than they eliminated, and one of those involved having a ship which, when damaged, fired on (and sunk) itself so as not to reduce the overall fleet agility.

• Gram_StoneFeb 21, 2017, 10:57 PM

  0 points

  in reply to: skeptical_lurker’s comment on: Open Thread, Feb. 20 - Feb 26, 2017

  Why do you mourn when you can contemplate politics no more? What makes you think about it so much in the first place? That just seems like something you wouldn’t want to ignore.

• Gram_StoneFeb 21, 2017, 5:24 PM

  0 points

  in reply to: lifelonglearner’s comment on: Lev­ers, Emo­tions, and Lazy Eval­u­a­tors:

  I admit I don’t quite understand what MINERVA-DM is...I glanced at the paper briefly and it appears to be a...theoretical framework for making decisions which is shown to exhibit similar biases to human thought? (With cells and rows and ones?)

  I can’t describe it too much better than that. The framework is meant to be descriptive as opposed to normative.

  A complete description of MINERVA-DM would involve some simple math, but I can try to describe it in words. The rows of numbers you saw are vectors. We take a vector that represents an observation, called a probe, along with all vectors in episodic memory, which are called traces, and by evaluating the similarity of the probe to each trace and averaging these similarities, we obtain a number that represents a global familiarity signal. By assuming that people use this familiarity signal as the basis of their likelihood judgments, we can simulate some of the results found in the field of likelihood judgment.

  I suspect that with a bit of work, one could even use MINERVA-DM to simulate retrospective and prospective judgments of task duration, and thus, planning fallacy.

• Gram_StoneFeb 21, 2017, 1:17 AM

  1 point

  in reply to: lifelonglearner’s comment on: Lev­ers, Emo­tions, and Lazy Eval­u­a­tors:

  I am assuming the first point is about this post and the second two are about the planning primer?

  The first two are about this article and the third is about the planning fallacy primer. I mentioned hypothesis generation because you talked about ‘pair debugging’ and asking people to state the obvious solutions to a problem as ways to increase the number of hypotheses that are generated, and it pattern matched to what I’d read about hypothesis generation.

  There are definitely several papers on memory bias affecting decisions, although I’m unsure if we’re talking about the same thing here. What I want to say is something like “improperly recalling how long things took in the past is a problem that can bias predictions we make” and this phenomena has been studied several times.

  I’m definitely talking about this as opposed to the other thing. MINERVA-DM is a good example of this class of hypothesis in the realm of likelihood judgment. Hilbert (2012) is an information-theoretic approach to memory bias in likelihood judgment.

  I’m just saying that it looks like there’s a lot of fruit to be picked in memory theory and not many people are talking about it.

• Gram_StoneFeb 21, 2017, 12:23 AM

  1 point

  on: Lev­ers, Emo­tions, and Lazy Eval­u­a­tors:

  There is a problem where I say “Your hypothesis is backed by the evidence,” when your entirely verbal theory is probably amenable to many interpretations and it’s not clear how many virtue points you should get. But, I wanted to share some things from the literature that support your points about using feelings as information and avoiding miserliness.

  First, there is something that’s actually just called ‘feelings-as-information theory’, and has to do with how we, surprise, use feelings as sources of information. ‘Feelings’ is meant to be a more general term than ‘emotions.’ Some examples of feelings that happen to be classified as non-emotion feelings in this model are cognitive feelings, like surprise, or ease-of-processing/​fluency experiences; moods, which are longer-term than emotions and usually involve no causal attribution; and bodily sensations, like contraction of the zygomaticus major muscles. In particular, processing fluency is used intuitively and ubiquitously as a source of information, and that’s the hot topic in that small part of cognitive science right now. I have an entire book on that one feeling. I did write about this a little bit on LW, like in Availability Heuristic Considered Ambiguous, which argues that Kahneman and Tversky’s availability heuristic can be fruitfully interpreted as a statement about the use of retrieval fluency as a source of information; and Attempts to Debias Hindsight Backfire!, which is about experiments that manipulate fluency experiences to affect people’s retroactive likelihood judgments. The idea of ‘feelings as information’ looks central to the Art.

  There is also a small literature on hypothesis generation. See the section ‘Hypothesis Generation and Hypothesis Evaluation’ of this paper for a good review of everything we know about hypothesis generation. Hardly inspiring, I know. The evidence indicates that humans generate relatively few hypotheses, or we may also write, humans have impoverished hypothesis sets. Also in this paper, I saw studies that compare hypothesis generation between individuals and groups of various sizes. You’re right that groups typically generate more hypotheses than individuals. They also tried comparing ‘natural’ and ‘synthetic’ groups, natural groups are what you think; the hypothesis sets of synthetic groups are formed from the union of many individual, non-group hypothesis sets. It turns out that synthetic groups do a little better. Social interaction somehow reduces the number of alternatives that a group considers relative to what the sum of their considerations would be if they were not a group.

  Also, about your planning fallacy primer, I think the memory bias account has a lot more going for it than a random individual might infer from the brevity of its discussion.

• Gram_StoneJan 28, 2017, 10:53 PM

  0 points

  in reply to: RomeoStevens’s comment on: 2017: An Ac­tual Plan to Ac­tu­ally Improve

  I’ll try to write a short post on it at some point.

  Please do!

• Gram_StoneJan 26, 2017, 10:23 PM

  6 points

  on: 80,000 Hours: EA and Highly Poli­ti­cal Causes

  (Tentatively upvoted.)

  I find that a good way to make statements criticizing individuals or organizations less provocative is to frame your criticism as a confusion. This simultaneously allows you to demonstrate that you’ve thought about their reasoning for more than five minutes and tends to make any further discussion less adversial.

  The abstract reasoning about why prison reform is a bipartisan cause makes sense to me: prisons cost lots of money (bad conservative metric) and they’re disproportionately inhabited by minorities (bad liberal metric), but if your descriptions of their recommended organizations are charitable, then I too am confused right now.

• Gram_StoneJan 24, 2017, 2:46 AM

  0 points

  on: Open thread, Jan. 23 - Jan. 29, 2017

  Does anyone have an electronic copy of the Oxford Handbook of Metamemory that they’re willing to share?

• Gram_StoneJan 22, 2017, 4:17 AM

  6 points

  on: Pro­jects-in-Progress Thread

  I think it’s possible to exercise Hufflepuff virtue in the act of encouraging more Ravenclaw virtue, right? That is, getting an arbitrary ball rolling is a Hufflepuff thing to do, even if you roll the ball in a Ravenclaw direction? That’s an important distinction to me.

  A mid-term goal of mine is to replicate Dougherty et al.‘s MINERVA-DM in MIT/​GNU Scheme (it was originally written in Pascal; no, I haven’t requested the authors’ source code, and I don’t intend to). I also intend to test at least one of its untested predictions using Amazon Mechanical Turk, barring any future knowledge that makes me think that I won’t be able to obtain reliable results (which has only become less plausible as I’ve learned more; e.g. Turkers are more representative of the U.S. population than the undergraduate population that researchers routinely sample from in behavioral experiments; there’s also a few enthusiasts who have done some work on AMT-specific methodological considerations).

  MINERVA-DM is a formal model of human likelihood judgments that successfully predicts the experimental findings on conservatism), the availability heuristic, the representativeness heuristic, the base rate fallacy, the conjunction fallacy, the illusory truth effect, the simulation heuristic, and the hindsight bias. MINERVA-DM can also be described as a modified version of Bayes’ Theorem. I’m not too far yet, having just started learning Scheme/​programming-in-general, but I have managed to hobble together a one-line program that outputs an n-vector with elements drawn randomly with replacement from the set {-1, 0, 1}, so I guess I’ve technically started writing the program.

  It’s worth saying that I’m not very confident that MINERVA-DM won’t be overturned by a better model, and that’s not the point.

  I need some sort of example, and MINERVA-DM has good properties as an example, because its math is exceedingly simple (i.e., capital-sigma notation, arithmetic mean, basic probability theory (see Bolstad’s Introduction to Bayesian Statistics, Ch.3), etc. There are probably plenty of improvements that we need to and could make as a community, but my own concern is that it’s never been winter-night-clear to me why at least some of us aren’t trying to perform (Keyword Alert!) heuristics and biases/​judgment and decision making (JDM)/​behavioral decision theory research on LW or on whatever conversational focus we may be using in the near- to mid-term future. There is no organization in the community for this; CFAR is the closest thing to this, and AFAICT, they are not doing basic research into H&B/​JDM/​BDT. People around here seem to me more likely than most to agree that you’re more likely to make progress on applications if you have a deep understanding of the problem that you’re trying to solve.

  I think it is intuitive that you simply cannot productively do academic work solely in the blogosphere, and when you’re explaining a counterintuitive point, a point that is not universal background knowledge, you should recurse far enough to prevent misunderstandings in advance. I no longer find it intuitive that you can’t do a substantial amount of work on the blogosphere. For one, a good deal of academic work, especially the kind we’re collectively interested in, doesn’t require any special resources. Reviews, syntheses, analyses, critiques, and computational studies can all be done from a keyboard. As for experiments, we don’t need to buy a particle accelerator for psych research, you guys; this is where Mechanical Turk comes in. E.g. see these two blog posts wherein a scientist replicates one of Tversky and Kahneman’s base rate fallacy results with N = 66 for US$3.30, and replicates one of Tversky and Kahneman’s conjunction fallacy results with N = 50 for US$2.50. (Here’s a list with more examples.)

  Arguing that there’s important academic work that doesn’t require anything but a computer (reviews, syntheses, analyses, computational studies), and demonstrating that you can test experimental predictions with your lunch money seems like a good start on preempting the ‘you can’t do real science outside of academia’ criticism. (It’s not like there isn’t a precedent for this sort of thing around here anyway.) It also prevents people from calling you a hypocrite for proposing that the community steer in a certain direction without your doing any of the pedaling. I probably would’ve kept quiet for a lot longer if I didn’t think it were important to the community to respond to calls like this article, especially considering that we may be moving to a new platform soon.

• Gram_StoneJan 21, 2017, 3:31 AM

  6 points

  in reply to: Qiaochu_Yuan’s comment on: Willpower du­al­ity (a very short es­say)

  (for example, focusing seems to be related to nonverbal parts, but it sort of breaks the S1/​S2 dichotomy by being nonverbal but slow.)

  Noncentral nitpick that is meant to be helpful: Focusing is a counterexample to the lay dual process theory that people sometimes use around here, but not the up-to-date, cognitive-scientific one.

  Briefly, the key distinction (and it seems, the distinction that implies the fewest assumptions) is the amount of demand that a given process places on working memory.

  nonverbal

  Although language is often involved in Type 2 processing, this is likely a mere correlate of the processes by which we store and manipulate information in working memory, and not the defining characteristic per se. To elaborate, we are widely believed to store and manipulate auditory information in working memory by means of a ‘phonological store’ and an ‘articulatory loop’, and to store and manipulate visual information by means of a ‘visuospatial sketchpad’, so we may also consider the storage and processing in working memory of non-linguistic information in auditory or visuospatial form, such as musical tones, or mathematical symbols, or the possible transformations of a Rubik’s cube, for example. The linguistic quality of much of the information that we store and manipulate in working memory is probably noncentral to a general account of the nature of Type 2 processes. Conversely, the production and comprehension of language must often be an associative or procedural process, rather than a deliberative one; otherwise you might still be parsing the first sentence of this comment. That’s all technically original research and I Am Not A Cognitive Scientist, but I think it should be pretty obvious even from a layperson’s perspective.

  slow

  There’s nothing stopping Type 2 from being relatively fast, either; it’s just another correlate that doesn’t always hold. Trivial example: Have you ever awoken and not been able to make mental sense of what you’re seeing for a few seconds? It might take you longer to do that than to perform one transformation of a Rubik’s cube while fully awake, even though the former is automatic and the latter deliberate. In general, people sometimes seem to act as if there has never been a judgment that was simultaneously deliberate and fast, because they have come to describe all fast judgments as automatic. Such judgments are plausible in my experience.

  See also: Evans (2013).

• Gram_StoneJan 19, 2017, 10:38 PM

  0 points

  in reply to: Vaniver’s comment on: Thoughts on “Oper­a­tion Make Less Wrong the sin­gle con­ver­sa­tional lo­cus”, Month 1

  comments to comments

  Do you know if you’ll be able to maintain their familial relationships as well?

• Gram_StoneJan 16, 2017, 1:51 AM

  0 points

  in reply to: fubarobfusco’s comment on: X Is Not About Y: Tech­nolog­i­cal Im­prove­ments and Cog­ni­tive-Phys­i­cal Demands

  Not sure if this should make me feel better or worse.

• Gram_StoneJan 15, 2017, 9:10 PM

  0 points

  in reply to: TiffanyAching’s comment on: X Is Not About Y: Tech­nolog­i­cal Im­prove­ments and Cog­ni­tive-Phys­i­cal Demands

  Yeah, post hoc rationalization or deception makes more sense than what I said.

X Is Not About Y: Tech­nolog­i­cal Im­prove­ments and Cog­ni­tive-Phys­i­cal Demands

• Gram_StoneJan 14, 2017, 11:41 PM

  0 points

  in reply to: Jiro’s comment on: Plan­ning the Enemy’s Retreat

  Stipulation is obviously sometimes a cheat. I would be surprised if it was always one.

• Gram_StoneJan 14, 2017, 6:23 AM

  2 points

  in reply to: TiffanyAching’s comment on: Plan­ning the Enemy’s Retreat

  (Upvoted.) Just wanted to say, “Welcome to LessWrong.”