Links for January 2017

  • A Four Planet System in Orbit, Directly Imaged“.
  • The main arched cover of the new 100-year sarcophagus for the Chernobyl site was slid into place in November, the largest object ever moved.
  • Time-resolved 2-million-year-old supernova activity discovered in Earth’s microfossil record
  • How do you make reasonable decisions under large empirical and normative uncertainty when literally billions of dollars are at stake? Holden Koronophsky grapples with this extreme version of a key EA dilemma. Relatedly: this Wired profile of John Arnold, billionaire trader who pioneered some evidence-based philanthropic strategies celebrated by EAs.
  • New strong evidence for the hypothesis that schizophrenia is ameliorated by nicotine. This suggests that there were serious negative effects of unfocused efforts to stop smoking with, e.g., blanket bans at psych hospitals, as emphasized by Scott Alexander. Also by way of Scott: Agora is an experimental web forum threading system where comments can not only have multiple branching reply comments (like reddit) but also be in reply to multiple parent comments.
  • Alex Tabarrok on the importance of the FDA commissioner.
  • Why it Takes so Long to Connect to a WiFi Access Point?” (Excellent HN discussion.)
  • More on the recent success of solid C02 sequestration in Iceland.
  • OK, this is extremely sappy, but I’m a sucker for it. “Wanderers” is a short film by Erik Wernquist featuring illustrations of future exploration within the solar system with narration by Carl Sagan:

    Wanderers – a short film by Erik Wernquist from Erik Wernquist on Vimeo.

  • Eliezer Yudkowsky and Bryan Caplan bet on the end of the world.
  • PRL is now charging authors $800 and that doesn’t include open access (which is $2900)? Surprisingly, I haven’t heard much discussion of this.
  • Galileo satellites experiencing multiple clock failures. It’s unclear if this will hold up the launch of the rest of the constellation until the problem is isolated.
  • Unviewed videos on YouTube as a window into random parts of humanity. Excellent execution of this idea. (“It’s literal reality TV.”)
  • 538 illustrates, with some great plots, how effective coaches and players are at pressuring nearby officials to give them the calls they want:

    …refs make more defensive pass interference calls on the offensive team’s sideline but more offensive holding calls on the defensive team’s sideline. What’s more, these differences aren’t uniform across the field — the effect only shows up on plays run, roughly, between the 32-yard lines, the same space where coaches and players are allowed to stand during play.

    I do worry about publication bias, since they only gave charts for a few classes of penalties. It’s possible they simply selected those that behaved like they expected. (H/t Will Riedel.)

  • Here is an excellent Google Earth 3D trajectory of the recent successful SpaceX launch, complete with real-time plots of altitude and velocity, synced up with the broadcast video:

    This gives you a great feel for the trajectory actually taken by launches to low-Earth orbit. It includes the path of the 1st stage booster, which landed on the recovery drone ship. Incidentally, SpaceX is scheduled in midlate-February to re-launch a previously flown booster for the first time.
  • The number of new physics PhD’s per year in the US grew exponentially from the end of the 19th century right up until about 1970, when it suddenly flatlined.

    it should be clear by now that with more than half the kids in America already going to college, academic expansion is finished forever.

    The average American professor in a research university turns out about 15 Ph.D students in the course of a career. In a stable, steady-state world of science, only one of those 15 can go on to become another professor in a research university.

    I think this gives you license to ignore advice on getting a permanent job from anyone who landed one before about 1975…

  • HN Discussion: “Why do traders in investment banks feel their jobs are immune from AI, etc?
  • Sarah Constantin replies to critics.
  • Update on the state of MOOCs from the Economist.
  • Gate Foundation joins cartel of justice by prohibiting it’s fundees from publishing in journals if they aren’t completely open access. (Six month embargoes not acceptable.) In particular, Nature, Science, Cell, and PNAS not approved. All journals will eventually move to OA, but it is depressing that it will literally take multiple decades since OA become obviously good and started to be adopted before it actually becomes universal.
  • Banishment as a criminal punishment is prohibited in most US states, and at the Federal level, but still exists in a few state constitutions and was used in Washington, DC a few years ago as a condition of bail.
  • Example of confused discussion that unfolds when people don’t acknowledge the possibility education is largely about signaling.
  • The view of the Solar system from Alpha Centauri.
  • NuScale submits proposal for modular 50-MW nuclear reactors to the Nuclear Regulatory Commission.
  • A possible future escalation on ransomware: “quiet encryption“. Malware secretly encrypt the all the data on the infected user’s device, but invisibly adds a wrapper around it so that any time the user requests data, the device obtains the key from the attacker’s external server and returns the data like nothing is wrong. After an extensive “incubation period”, the attackers disable the external server, and now the user is unable to access their data. This prevents using frequent backups as a defense against ransomware, since even the backups of data created during the incubation period will be encrypted. What’s not clear to me is whether this is really easier than just silently disabling the backup process during the incubation period.
  • Interview about life in North Korea from two defectors.

    (Contains oral description of violence.) The population is 25 million people, and the regime has now persisted for half a century. A famine between 1994 and 1998 killed at least 2% of the population, and perhaps as much as 10%.
  • Nature News: “Five big mysteries about CRISPR’s origins“.
  • Sabine Hossenfelder on new, non-terrible data for dispersion effects in gamma-ray bursts, with speculative explanations from quantum gravity.
  • Fancy architecture for Antarctic research stations.
  • Hearing originally developed in early animals as very fine sensation in jaw bones that probably was used to detect vibrations in the Earth when the jaw rested against the ground, only later specializing for air vibrations.

    The evolution of mammalian auditory ossicles is one of the most well-documented and important evolutionary events, demonstrating both numerous transitional forms as well as an excellent example of exaptation, the re-purposing of existing structures during evolution.

    In reptiles, the eardrum is connected to the inner ear via a single bone, the columella, while the upper and lower jaws contain several bones not found in mammals. Over the course of the evolution of mammals, one lower and one upper jaw bone (the articular and quadrate) lost their purpose in the jaw joint and were put to new use in the middle ear, connecting to the existing stapes bone and forming a chain of three bones (collectively called the ossicles) which transmit sounds more efficiently and allow more acute hearing.

  • Two courses on identifying bullshit. I wonder if these are useful.
  • Will automated pricing systems be able to collude to fix prices in a more stable manner, with less of a paper trail, than humans? How much of the efficiency in capitalism is due to the difficulty of coordinating by computationally limited humans?

    If you take this set of claims seriously…then the most fundamental problem that the Internet poses is not one of network advantage, increasing returns to scale and so on advantaging big players (since, with a non-supine anti-trust authority, these could in principle be addressed). It’s the problem of how radically cheaper communication makes new forms of implicit and explicit collusion possible at scale, squeezing consumers.

    H/t Tyler Cowen. Also from him: Declassified CIA maps.

  • Collision of two stars at the end of binary system in-spiral is predicted for 2022. Will go from invisible to visible with the naked eye.
  • If you’re not already watching the Primitive Technology YouTube channel, check it out. The wattle-and-daub hut is spectacular:
  • Federal investigation of Tesla autopilot death very positive, praises Tesla. Over the air update gives a good natural experiment: enabling autosteer abilities decreases accident rate by 40%. (Decent HN discussion.) On the other hand, Toyota’s Gill Pratt argues that people underestimate the huge jump in difficulty from Level 4 to Level 5 autonomous vehicles (using the SAE automated vehicle classification). However, note that the consumer already gets enormous benefits at Level 4, even if it’s only available in certain circumstances like on the highway with good weather. Furthermore, a lack of Level 5 can be further ameliorated by driving call centers. That is: If self-driving cars can reach a sufficient threshold of reliability, and if the car can autonomously recognize the edge cases that confuse it, then the last 0.1% of cases can be handled with teleoperation:

    “It’s going to be massively important,” says Karl Iagnemma, co-founder and CEO of self-driving startup nuTonomy, which is developing a remote control system. Even cars that can handle just about anything will have the occasional failure, even if that’s being hit by another vehicle. And in that case, you want a human around to decide what to do. It’s like an elevator, Iagnemma says: You don’t need a human operator, but you’ve still got a button to call for help when you need it.

    Google’s self-driving car outfit, Waymo, has studied the idea, a spokesperson says. Uber declined to comment on teleoperation, but in 2015 the company filed a patent for a system that would let an autonomous vehicle follow a human-driven car, or get help from a remote operator. Stealthy self-driving car startup Zoox has a patent for a “teleoperation system and method for trajectory modification of autonomous vehicles;” Toyota has one for “remote operation of autonomous vehicle in unexpected environment.”

    Now, Nissan’s cubicle-based drivers aren’t emergency backups. If the car hits black ice, it’s in charge of staying on the road. There’s no feasible way to get the human into the loop in time to act. But they can help out when the car encounters conditions it’s unsure how to handle. If a Nissan happened upon the construction scene from Sierhuis’ photo, it would stop and ping its control center. A human operator would look around using the car’s cameras and other sensors and issue new instructions—direct control would pose latency issues. Like: When it’s safe, cross the double yellow and get back to the right side after 20 yards. Or a new instruction set could ensure packages and disabled passengers get dropped off in exactly the right spot, and help assess potentially dangerous situations on the road. But most of all, the teleoperator is there to make sure the car’s doesn’t just shut down when it’s too dumb to know what’s going on.

    Relatedly, here is VC Frank Chen discussing 16 open questions driving the development of self-driving cars. At the end is a summary of predictions for when these will appear. And Nvidia and Audi claim Level 4 Autonomy by 2020. Lots of answers between 2020 and 2023, but I think correcting for over optimism makes me still feel OK about my bet with Paul.

  • Documentary on the world’s largest building by volume:
  • Probably unlikely, but it may be possible for serious “mind crimes” to come before any form of AI. Human cells, originating from injected stem cells, have successfully been grown in pig fetuses. This is motivated by dreams of growing human organs in pigs, but opens up the possibility (for now, remote) that human stem cells could migrate to the embryonic animal brain. Worth noting that this is a topic on which it’s possible to worry about wildly unlikely scenarios that run far ahead of our capabilities, with terrible popular coverage.
  • The Columbia-class submarine is a long-term project:

    The Columbia-class submarine…is a future United States Navy nuclear submarine class designed to replace the Trident missile-armed Ohio-class ballistic missile submarines. The first submarine is scheduled to begin construction in 2021 and enter service in 2031 (some 50 years after its immediate predecessor, the Ohio class, entered service). From there, the submarine class will serve through 2085.

    A fleet of 12 was recently approved by the Pentagon.

  • Children on leashes.
  • Alexa: Amazon’s Operating System“. (Good HN discussion.)
  • When does it make sense to ship things from China to Europe by rail rather than sea? Rough numbers from jakozaur:

    Rail is 4 times more expensive than by sea. Air is at least 6 times more expensive than rail, but can be even 18 times more expensive if payload is heavy.

    Cost of shipping 40′ HC container:

    • by sea: $1,600
    • by train: $6,800
    • by air, light, so you pay for volume: 4.6 USD / kg, 40 ft got volume 67.7 m3 * 143 kg / m3 ~= 9 tons, $41,400
    • by air, heavy, so you pay for real weight: 4.6 USD / kg, 40tf can be 27tons, $124,200
  • The Lucas device is a mechanical CPR machine.

    It’s unclear whether such devices are a net improvement. Interestingly, it’s also unclear whether or not people experiencing a cardiac arrest outside of a hospital should be immediately transported.

    But for Dr Jonathan Benger (University of the West of England, Bristol, UK), the strategy of taking these sick patients to the hospital made sense when the hospital was the only place that had defibrillators.

    “However, this is no longer the case, and hospitals have nothing to offer almost all such patients beyond the care that is provided by a well-trained and equipped ambulance service,” writes Benger. “Preparing patients for transport, moving them, and driving them to the hospital leads to pauses in CPR and suboptimal chest compressions, even with the most skilled and committed staff.”

  • Michael Nielsen on scientometrics

    In this essay I argue that heavy reliance on a small number of metrics is bad for science. Of course, many people have previously criticised metrics such as citation count or the h-index. Such criticisms tend to fall into one of two categories. In the first category are criticisms of the properties of particular metrics, for example, that they undervalue pioneer work, or that they unfairly disadvantage particular fields. In the second category are criticisms of the entire notion of quantitatively measuring science. My argument differs from both these types of arguments. I accept that metrics in some form are inevitable – after all, as I said above, every granting or hiring committee is effectively using a metric every time they make a decision. My argument instead is essentially an argument against homogeneity in the evaluation of science: it’s not the use of metrics I’m objecting to, per se, rather it’s the idea that a relatively small number of metrics may become broadly influential. I shall argue that it’s much better if the system is very diverse, with all sorts of different ways being used to evaluate science. Crucially, my argument is independent of the details of what metrics are being broadly adopted: no matter how well-designed a particular metric may be, we shall see that it would be better to use a more heterogeneous system.

    See also CognitiveMedium.com for his more recent thoughts on other topics.

Bookmark the permalink.

Leave a Reply

Include [latexpage] in your comment to render LaTeX equations with $'s. (More.)

Your email address will not be published. Required fields are marked with a *.