[ / / / / / / / / / / / / / ] [ dir / random / 93 / biohzrd / hkacade / hkpnd / tct / utd / uy / yebalnia ]

/ratanon/ - Rationalists Anonymous

Remember when /ratanon/ was good?

Name
Email
Subject
REC
STOP
Comment *
File
Password (Randomized for file and post deletion; you may also set your own.)
Archive
* = required field[▶Show post options & limits]
Confused? See the FAQ.
Embed
(replaces files and can be used instead)
Oekaki
Show oekaki applet
(replaces files and can be used instead)
Options

Allowed file types:jpg, jpeg, gif, png, webp,webm, mp4, mov, pdf
Max filesize is16 MB.
Max image dimensions are15000 x15000.
You may upload5 per post.


File: 24023823032a209⋯.jpg (4.67 KB,259x194,259:194,1490202583735.jpg)

File: 95d11af5851e780⋯.png (73.06 KB,240x250,24:25,1492428917067.png)

 No.9520

Whether a single-factor model is appropriate

The positive manifold exists, but the existence of the positive manifold is merely necessary rather than sufficient for the existence of a “general factor” with all-positive loadings. From a psychological standpoint, the fact that many of the existing methods used in the literature can’t differentiate between single-factor causal models vs. causal models in which a single causal factor doesn’t exist is deeply unsatisfying regardless of the statistical validity of the single-factor representation!

There also seem to be a priori reasons to believe that the “true” causal model of intelligence is multidimensional rather than unidimensional, with perhaps the most compelling one being that human personality seems highly multidimensional (breaking down in a complicated non-hierarchical fashion). Even more generally, complex mental traits don’t seem to be well-represented by single factors in any sense aside from the tautological positive-manifold-implies-general-factor sense.

This is certainly not to say that e.g. a mutualism model [1] of intelligence must necessarily be correct, but I’m disappointed that more complex statistical techniques, such as Bayesian modeling and model misspecification testing, seem to not have seen much application in the psychological sciences.

I’m also very interested in how education and pedagogy might be best adapted for children 4+ standard deviations above the mean—topics which are woefully understudied and underappreciated! If we really do find ourselves nearly upon the precipice of a societal revolution brought about by iterated embryo selection, then figuring out what to do with these children—how to educate and socialize them—is plausibly one of the most important things we can do. See, e.g., Vulnerabilities of highly gifted children [2] and The Outsiders [3]. It’s plausible to me that a variety of developmental difficulties stem from having high overall cognitive ability but very unbalanced subtest performance, especially in the case of high verbal intelligence paired with lower working memory. Naturally such differences become obscured if too much focus is given to g alone… Even if a single-factor causal model is appropriate for the majority of the population, Spearman’s law of diminishing returns (i.e., the increased cognitive differentiation at the high end) implies that a more nuanced and multifactorial study of intelligence is sorely needed to make real headway here.

Despite all of the above, it does seem that a lot of human variation can be represented on a single dimension. Some people are functional and successful; others less so. I would have a hard time disputing that g (as presently conceptualized and measured) does a decent—not perfect, but decent—job of capturing this variation in a statistical sense. But from a psychological perspective, if we (1) eschew causality for predictive power, (2) don’t really make an effort to distinguish the true underlying model, and (3) take into account correlates of g that are plausibly disconnected from cognitive ability, then (c.f. papers like Intelligence tests with higher g-loadings show higher correlations with body symmetry: Evidence for a general fitness factor mediated by developmental stability [4]), it seems that what we’re getting at is far too general to really be called a general factor of intelligence! (One can contend that I’m just splitting hairs here and that predictive power is predictive power, but wouldn’t accepting the above imply that it makes more sense, at least on the margin, to examine multidimensional models of intelligence? If the true causal structure of intelligence is multidimensional, then we should get better results with iterated embryo selection for the causal variants associated with the different sub-intelligences rather than for g, right?)

(It’s telling that John L. Horn seems to advocate [5] for a multidimensional theory of intelligence! If you haven’t seen the linked paper, it’s worth reading for a historical perspective on psychometric theory.)

____________________________
Disclaimer: this post and the subject matter and contents thereof - text, media, or otherwise - do not necessarily reflect the views of the 8kun administration.

 No.9521

Capturing causality in predictive models

It occurs to me that epistasis (insofar as it accounts for phenotypic variance in a given trait) might be statistically well-represented in a predictive model in a way that fails to capture causality, which has consequences for the effectiveness of (iterated) embryo selection. This is potentially true both of additive models (e.g., LASSO) and more complex nonlinear models (GxE interactions may also exacerbate this problem). I suppose the question here is how one would expect the estimates of effect sizes to change if this were the case. If additive models still identify the right causal variants and merely underestimate the effect sizes, the problems aren’t that consequential; however it’s also plausible to me that these complications could end up to identification of false positives/false negatives or incorrect estimates of the signs of effect sizes while still giving good predictive power!

I don’t have a good sense of what one ought to expect here (one wants to ideally have a sense of not just how much epistasis is present in, say, animals, but also the form of that epistasis). One possibly relevant paper is Epistasis and the release of genetic variation during long-term selection [6], described in this review [7] as follows:

Some of the issues that arise in QTL mapping in the presence of gene interactions are illustrated by a recent study by Carlborg et al. into the genetic basis of the response to selection on body size in chickens (FIG. 3). After approximately 40 generations of selection, males in the low-weight line weighed six times less than males in the high-weight line. It would seem likely that there would be a strong genetic signal behind such a difference. However, after examining many marker loci for their individual effects, only one QTL (named Growth9) seemed to have an effect, and the signal for that was weak. However, by looking for epistatic relationships among the markers the authors were able to identify five additional genomic regions with significant effects on growth, each of which only showed their effects in the high-growth background, that is, the line with the Growth9 QTL (FIG. 3). Together, this loose network of epistatic genes accounted for 45% of the difference among the selected lines, an overall effect of 3.3 phenotypic standard deviations. The individual effect of Growth9 was completely accounted for by its epistatic interactions with the other QTLs.

Another review, Epistasis and quantitative traits: using model organisms to study gene–gene interactions [8], says:

The role of epistasis in the genetic architecture of quantitative traits is controversial, despite the biological plausibility that nonlinear molecular interactions underpin the genotype–phenotype map. This controversy arises because most genetic variation for quantitative traits is additive. However, additive variance is consistent with pervasive epistasis. In this Review, I discuss experimental designs to detect the contribution of epistasis to quantitative trait phenotypes in model organisms. These studies indicate that epistasis is common, and that additivity can be an emergent property of underlying genetic interaction networks. Epistasis causes hidden quantitative genetic variation in natural populations and could be responsible for the small additive effects, missing heritability and the lack of replication that are typically observed for human complex traits.

See also Dissection of the Genetic Architecture of Body Weight in Chicken Reveals the Impact of Epistasis on Domestication Traits [9]. In general, the importance of distinguishing between “true” genetic additivity and additivity as an emergent property of epistasis seems underappreciated. I hope to look into the literature more as time permits, as I think I’ve barely scratched the surface.

Disclaimer: this post and the subject matter and contents thereof - text, media, or otherwise - do not necessarily reflect the views of the 8kun administration.

 No.9522

One tangentially related point is that I’m very skeptical of claims that environmental influences on g “wash away” in the long run mainly because I have the pessimistic view that almost all modern environments are terrible for humans, especially for (but not limited to!) those born with high innate ability. I mean this in a variety of different senses: that the modern world and modern education verge on being traumatically difficult to live in for the majority of people, that approximately ~99.999% of schools have embarrassingly poor pedagogical practices, that the vast and overwhelming majority of parents have no idea how to raise children (let alone inculcate intellectual tendencies in them), and so on and so forth … If the cultural standard of parenting and education is the equivalent of ineptly trying to put out a bonfire by blowing on it, then of course no effect will be found! (For an example of what I mean, just consider the embarrassingly small number of math textbooks that one can actually consider to be “written well” in an expository sense…)

All considered, I’m increasingly loath to make any confident extrapolations from studies of heritability, interactions, environmental effects, …, outside of the specific context in which they were performed.

(I know that one can in principle [10] use regularized regression to account for nonlinearity nearly optimally given certain reasonably plausible assumptions, although on priors alone I’m skeptical of practical efficacy until the method is further validated with biological data.)

The evolutionary argument for genetic additivity

I remember being struck by the argument

However, quantitative differences between individuals within a species may be largely due to independent linear effects of specific genetic variants. As noted, linear effects are the most readily evolvable in response to selection, whereas nonlinear gadgets are more likely to be fragile to small changes. (Evolutionary adaptations requiring significant changes to nonlinear gadgets are improbable and therefore require exponentially more time than simple adjustment of frequencies of alleles of linear effect.)

which seemed very elegant to me.

Upon further reflection, it seems that this might prima facie be in tension with the model of the genetics of intelligence where the minor alleles of the causal variants reduce intelligence. If we assume that natural selection has already been going on for so long that the major alleles are predominantly those which increase intelligence, that implies the existence of a long period of time over which one could imagine accumulation of random, fitness-neutral genetic drift, which plausibly results in epistatic and nonlinear mechanisms. It’s also not entirely clear that selective pressures are always biased toward the evolution of independent effects in a modular (rather than hierarchical or centrally controlled) framework. I quote from The frailty of adaptive hypotheses for the origins of organismal complexity [11]:

The hypothesis that expansions in the complexity of genomic architecture are largely driven by nonadaptive evolutionary forces is capable of explaining a wide range of previously disconnected observations (13,40) (Table 2). … The emergence of modular gene structure by the nonadaptive processes of duplication, degenerative mutation, and genetic drift is fully compatible with the known magnitudes of these forces in multicellular species (13). … [B]ecause of the mutational cost of allelic complexity, the likelihood of completion of semineutral modularization processes becomes negligible once 1/N_g becomes smaller than the excess mutational burden (66). Thus, contrary to popular belief, natural selection may not only be an insufficient mechanism for the origin of genetic modularity, but population-genetic environments that maximize the efficiency of natural selection may actually promote the opposite situation, alleles under unified transcriptional control.

See also An Evolutionary Perspective on Epistasis and the Missing Heritability [12]:

The relative importance between additive and non-additive genetic variance has been widely argued in quantitative genetics. By approaching this question from an evolutionary perspective we show that, while additive variance can be maintained under selection at a low level for some patterns of epistasis, the majority of the genetic variance that will persist is actually non-additive. … We demonstrate that the perception of independent additive effects comprising the majority of the genetic architecture of complex traits is biased upwards and that the search for causal variants in complex traits under selection is potentially underpowered by parameterising for additive effects alone.

Disclaimer: this post and the subject matter and contents thereof - text, media, or otherwise - do not necessarily reflect the views of the 8kun administration.

 No.9523

One can certainly argue, as some do, that additive models seem to work well for animals [13], but as I note above, the effectiveness of additive predictive models is merely necessary rather than sufficient for demonstrating additivity at a causal level.Plant breeding is not comparable to animal breeding

An acquaintance of mine, Lila Rieber [14], pointed out that polyploidy in plants means that it’s probably inappropriate to use the effectiveness of plant breeding as evidence for the potential of human genetic selection. This seems true to me to the extent that trait selection in plants arises from the selection of extremely polyploid genomes. See Artificial polyploidy as a tool in plant breeding [15], Some Applications and Misapplications of Induced Polyploidy to Plant Breeding [16], etc.

In fact, I see now that she wrote [17] in response to an interview Steve Hsu gave in 2013:

Comparisons with plants are misleading. The remarkable selective breeding of crops has been achieved through polyploidy. For example, strawberries are octoploid. Unlike plants, animals don't tolerate polyploidy. The closest humans have to polyploidy is trisomy, and one of the only autosomal trisomies that's nonlethal is Down syndrome, which obviously is severely disruptive and leads to reduced lifespans.

General lack of statistical sophistication in studies of g

Disclaimer: this post and the subject matter and contents thereof - text, media, or otherwise - do not necessarily reflect the views of the 8kun administration.

 No.9524

My overall impression is that every step of the process

Initial model building → Psychometric instrument design and validation → Statistical analysis of data to identify dimensions of intelligence → Inferences about distributions over populations or heritabilities

seems to practically be done wrong more often than done right, and that even if it seems prima facie as though many lines of evidence support some psychometric conclusion about intelligence, careful and critical examination of the methodology is needed to see how many of the lines of reasoning really hold up in practice. If I could just do whatever I want, I suppose I would spend several months researching this material in great depth and then write a long review paper about all of it, but in practice I unfortunately don’t have that luxury.

Given the potentially very high near-term importance of iterated embryo selection, I think there’s a real need to work this material out in a comprehensible way and figure out (1) where the uncertainty lies and (2) how that uncertainty affects our expectations of the effectiveness of embryo selection. I could imagine it being the case that stronger versions of the arguments above present compelling theoretical reasons why new models, statistical techniques, etc., might have to be developed for effective embryo selection past, say, the +4 standard deviations level; I could also imagine it being the case that to the extent that the objections are valid, they don’t imply barriers more significant than a larger sample size or computational time requirement (which will naturally topple with time without the need for theoretical developments).

It’s certainly plausible that after studying this material for a longer time, I’ll end up re-converging back to the views expressed in On the genetic architecture of intelligence and other quantitative traits [24]!

Endnotes

[1]: http://hvandermaas.socsci.uva.nl/Homepage_Han_van_der_Maas/Publications_files/papers/psyrev4.pdf

[2]: http://www.davidsongifted.org/Search-Database/entry/A10065

[3]: http://www.worlddreambank.org/O/OUTSIDRS.HTM

[4]: https://www.sciencedirect.com/science/article/pii/S0160289604000777

[5]: http://www.iapsych.com/articles/Horn2006b.pdf

[6]: http://www.nature.com/ng/journal/v38/n4/full/ng1761.html

[7]: https://www.nature.com/nrg/journal/v9/n11/full/nrg2452.html

[8]: https://www.nature.com/nrg/journal/v15/n1/full/nrg3627.html

[9]: http://www.genetics.org/content/179/3/1591.short

[10]: https://arxiv.org/abs/1408.6583

[11]: http://www.pnas.org/content/104/suppl_1/8597

[12]: http://journals.plos.org/plosgenetics/article?id=10.1371/journal.pgen.1003295

[13]: https://www.ncbi.nlm.nih.gov/pubmed/19109259

[14]: https://www.google.com/search?q=lila+rieber&oq=lila+rieber&aqs=chrome..69i57j69i60j69i61j69i60l3.1032j0j7&sourceid=chrome&ie=UTF-8

[15]: https://www.cabdirect.org/cabdirect/abstract/19581600186

[16]: https://link.springer.com/chapter/10.1007/978-1-4613-3069-1_23

[17]: https://intelligence.org/2013/08/31/stephen-hsu-on-cognitive-genomics/#comment-2493279854

[18]: http://www.sciencedirect.com/science/article/pii/S0160289616303233

[19]: http://linkinghub.elsevier.com/retrieve/pii/S0160289609001470

[20]: http://linkinghub.elsevier.com/retrieve/pii/S104160801000035X

[21]: http://www.sciencedirect.com/science/article/pii/S016028960400056X

[22]: http://linkinghub.elsevier.com/retrieve/pii/S0191886909002475

[23]: https://link.springer.com/article/10.1007/s11336-006-1447-6

[24]: http://arxiv.org/abs/1408.3421

Disclaimer: this post and the subject matter and contents thereof - text, media, or otherwise - do not necessarily reflect the views of the 8kun administration.

 No.9525

>>9523

Someone please edit this post (if possible) to add two newlines before "Plant breeding".

Disclaimer: this post and the subject matter and contents thereof - text, media, or otherwise - do not necessarily reflect the views of the 8kun administration.

 No.9526

>>9520

I know dick about shit, so I'm unqualified to comment on this, but I'd like to thank you for taking the time to write this, as it raises the /ratanon/ waterline.

Disclaimer: this post and the subject matter and contents thereof - text, media, or otherwise - do not necessarily reflect the views of the 8kun administration.

 No.9527

>>9526

I think /ratanon/ isn't well-suited for long-form posts. That OP wants to but can't edit his is evidence for that. It would be better if OP posted it on LW or https://gist.github.com (it's superior to pastebin because it allows markup) or registered a blog, then linked it here with a short summary. Moreover, it would mean you can discuss it with "normal" rats without linking to 8ch.

Disclaimer: this post and the subject matter and contents thereof - text, media, or otherwise - do not necessarily reflect the views of the 8kun administration.

 No.9528

>>9527

I refuse. Normal rats are repulsive.

Disclaimer: this post and the subject matter and contents thereof - text, media, or otherwise - do not necessarily reflect the views of the 8kun administration.

 No.9529

>>9527

Yeah, the need to split a long post, the inability to edit posts, and the lack of hyperlink footnotes (which are sadly also not present most other places :/) make me dubious that /ratanon/ is actually the optimal place to make a long post. However, /ratanon/ is slightly better than those other methods in at least one dimension; it's better than LessWrong because you can include arbitrary files with your post, better than gist/pastebin because you don't want to unnecessarily decouple information, and better than a registering a blog because your domain name or blogging service won't lapse. So it's not as if /ratanon/ is *strictly* worse than any other option, in the sense of being a subset of useful features.

>>9520

Sorry I still have nothing to say, please enjoy pic related.

Disclaimer: this post and the subject matter and contents thereof - text, media, or otherwise - do not necessarily reflect the views of the 8kun administration.

 No.9530

>>9529

Nice quads. I agree about domains, but do you expect 8chan to outlast Blogger or WordPress.com?

Disclaimer: this post and the subject matter and contents thereof - text, media, or otherwise - do not necessarily reflect the views of the 8kun administration.

 No.9531

>>9530

I expect to carry a leatherbound tome of the complete totality of /ratanon/ inscribed in vellum in the post-apocalyptic future, and be worshiped as a god for it.

Disclaimer: this post and the subject matter and contents thereof - text, media, or otherwise - do not necessarily reflect the views of the 8kun administration.

 No.9532

>>9530

Frankly, I don't even think 8chan will outlast *this week*.

Disclaimer: this post and the subject matter and contents thereof - text, media, or otherwise - do not necessarily reflect the views of the 8kun administration.

 No.9533

>>9531

Include me in the inscription.

Disclaimer: this post and the subject matter and contents thereof - text, media, or otherwise - do not necessarily reflect the views of the 8kun administration.

 No.9534

Best post on this board so far.

Disclaimer: this post and the subject matter and contents thereof - text, media, or otherwise - do not necessarily reflect the views of the 8kun administration.

 No.9535

>>9528

what's so repulsive about them? you don't like the gendershit? or you don't like the "smol birb fremb" aesthetic?

Disclaimer: this post and the subject matter and contents thereof - text, media, or otherwise - do not necessarily reflect the views of the 8kun administration.

 No.9536

>>9535

I don't like that they're navel-gazers who don't actually DO anything. Since I like DOING stuff, like writing the stuff I posted, I don't like them.

Disclaimer: this post and the subject matter and contents thereof - text, media, or otherwise - do not necessarily reflect the views of the 8kun administration.

 No.9537

File: b9255aa55e48818⋯.png (21.53 KB,255x135,17:9,IQ 70.png)

File: 8fe0a8ad0450d99⋯.png (33.42 KB,255x180,17:12,iq 70 2.png)

Disclaimer: this post and the subject matter and contents thereof - text, media, or otherwise - do not necessarily reflect the views of the 8kun administration.



[Return][Go to top][Catalog][Nerve Center][Random][Post a Reply]
Delete Post [ ]
[]
[ / / / / / / / / / / / / / ] [ dir / random / 93 / biohzrd / hkacade / hkpnd / tct / utd / uy / yebalnia ]