May 2018 News
May 2018 Gwern.net newsletter with 3 essays, links on genetic
engineering/
This is the May 2018 edition of the Gwern.net newsletter; previous, April 2018 (archives). This is a collation of links and summary of major changes, overlapping with my Changelog; brought to you by my donors on Patreon.
Writings
ZMA sleep self-experiment (inconclusive but suggestive of benefits)
Bacopa quasi-experiment (no correlates)
Media
Links
Genetics:
Engineering:
Tryon’s maze-bright rats (1972; selective breeding can affect heritable psychological traits in mammals like maze-running in just a few generations and produce total separation & non-overlapping distributions after ~7 generations. The mechanization of testing is also quite clever to allow large data & avoid any experimenter bias.)
“Early Canid Domestication: The Farm-Fox Experiment: Foxes bred for tamability in a 40-year experiment exhibit remarkable transformations that suggest an interplay between behavioral genetics and development”, 1999 (The Russian domesticated foxes; response to selection was such that they had to require much tougher scoring after just 6 generations to avoid ceiling effects.)
“DNA Testing Offers New Hope for Infants With Genetic Disease” (Reviewing some of the recent work on cost-benefits of much more extensive DNA testing.)
Everything Is Heritable:
“Comparison of Genotypic and Phenotypic Correlations: Cheverud’s Conjecture in Humans”, et al 2018 (Human phenotypic correlations are genetic correlations—everything is heritable & confounded. See also 1988, 1996, et al 2008, & 2011.)
“Study of 300,486 individuals identifies 148 independent genetic loci influencing general cognitive function”, et al 2018 (4.3% PGS; followup of et al 2017. One interesting part: lots of reaction time hits. We continue to await the third SSGAC paper, somehow. Appropriately, et al 2018 was released almost exactly on the 5th anniversary of et al 2013…)
“Common genetic variants contribute to risk of rare severe neurodevelopmental disorders”, et al 2018 (common variant burden partially moderates severity of ‘monogenic’ Mendelian disorders, liability-threshold style)
Recent Evolution:
“Global genetic differentiation of complex traits shaped by natural selection in humans”, et al 2018 (Racial differences in height/
waist-to-hip-ratio/ schizophrenia due to recent human evolution, with signs of differences on education/ IQ PGSes as well.) “The transition to modernity and chronic disease: mismatch and natural selection”, et al 2018
“Signatures of Long-Term Balancing Selection in Human Genomes”, et al 2018 (Seems like balancing selection is relatively minor outside co-evolutionary stuff like infections.)
“Exploring the genetic correlations of antisocial behavior and life history traits”, et al 2018; “Genome-Wide Association Studies of a Broad Spectrum of Antisocial Behavior”, et al 2017 (see also “Genome-wide association study of antisocial personality disorder”, et al 2016)
AI:
Scaling papers: A few datapoints in peak computation available for large-scale AI research (It’s a lot, but the hardware is still extremely inexpensive compared to many Big Science efforts—never mind white elephants like nuclear physics—and we should expect much more compute to be coming online in the near-future; neither the deep learning nor genomics revolution is over… Fortunately for non-corporate researchers, it’s still easy to make good contributions with small resources. It’s worth noting that the graph is highly incomplete, RL-focused, and omits a number of large-scale prior efforts: some evolutionary results, and very large CNNs like et al 2015/
et al 2017/ et al 2017/ et al 2017/ et al 2018/ et al 2018/ et al 2019 or GPipe scaling to 1663-layer/ 83.4b-parameter Transformers .)The Uber fatality: “Preliminary NTSB Report: Highway HWY18MH010” (Uber deliberately disabled the AI emergency braking because the AI was too erratic & unreliable to trust on real roads, deliberately disabled the standard Volvo emergency braking, and implemented no kind of alert system for the human driver. So the car detected the pedestrian well in advance, decided to emergency break, but was not able to break, and the human driver was not even alerted before killing the pedestrian! There should be manslaughter charges filed over this extraordinary negligence.)
Statistics/
“Tacit Knowledge, Trust, And The Q Of Sapphire”, 2001 (Tacit knowledge in replicating physics experiments measuring the Q factor: you need to grease the thread it hangs from, but the experiment works only with specific people’s skin grease!)
“Producing Wrong Data Without Doing Anything Obviously Wrong!”, et al 2009 (Programmers may not be interested in measurement error & the Replication Crisis but they’re interested in them: benchmark & optimization research have reproducibility & measurement error problems because they focus on small effects which can be swamped by variability from apparently trivial & irrelevant changes in the computer system as a whole.)
Politics/
“Homogeneous: The Political Affiliations of Elite Liberal Arts College Faculty”, 2018
“On The Number Of Siblings And p-th Cousins In A Large Population Sample”, 2017 (2nd cousin matches near-guaranteed in biobanks with even small fractions of populations; easier discussion: “How lucky was the genetic investigation in the Golden State Killer case?”; see also et al 2017.)
“How the Chinese Government Fabricates Social Media Posts for Strategic Distraction, not Engaged Argument”, et al 2017
“Basque Group ETA Disbands, After Terrorist Campaign Spanning Generations” (Terrorism is not effective at achieving goals.)
Psychology/
“An updated lead-crime roundup for 2018”; “Should Psychiatry Test For Lead More?”
“Revisiting the Marshmallow Test: A Conceptual Replication Investigating Links Between Early Delay of Gratification and Later Outcomes”, et al 2018 (effect size shrinks considerably in replication and the effect appears moderated largely through the expected individual traits like IQ; maybe the real marshmallow test was waiting for replications & better controls of the marshmallow test?)
“Results of the Self-Selection of Diets by Young Children”, 1939 (An odd diet experiment: what if you allow infants being weaned onto solid food complete freedom of choice of a list of random foodstuffs ranging from peas to sea salt to sour milk to bone marrow to kidneys to haddock, feeding them that way for 6 years? Turns out they come up with weird meals but wind up perfectly healthy.)
“Strength and Physique Systematic Review and Meta-Analysis Master List” (A list of weight-lifting/
exercise meta-analyses; a good starting point for getting a quick idea of what the literature says.)
Technology:
“Measuring Consumer Sensitivity to Audio Advertising: A Field Experiment on Pandora Internet Radio”, et al 2018 (Media. Do ads work, are they harmful, and can even extremely large-scale but short-term experiments show their harms? Yes, yes, and no. Results parallel my ad A/
B test ; w/bonus correlation ≠ causation analysis.) “Google Clips”: a lifelogging camera with automatic selection of interesting video snippets using on-device NN trained by 50m human pairwise comparisons & 100s-way CNN categorization.
Economics:
“500 Life-Saving Interventions and Their Cost-Effectiveness”, et al 1995 (Doing good is easy to not do well, and there are astronomical differences between charities: estimates of intervention cost-benefits range over 11 orders of magnitude.)
“How a Start-Up That Wouldn’t Break the Rules Was Forced to Fail”
Philosophy:
Misc:
Books
Nonfiction:
The Cat in Ancient Egypt, 1997 (some nice full-color images; Malek’s accounts are fairly speculative, and could have used a basic discussion of the conventions of ancient Egyptian art, which he often appeals to in interpreting the cat images and explaining why something is/
isn’t important, but never explains.)
Film/TV
Live-action:
Music
Misc:
“Leonardo” (Il Giardino degli Specchi; Oltremare {2018}) [post-rock]