With AlphaGo’s victory, Carl Shulman won his $100 bet with me (announced before the match here). For hindsight, here is a bit more evidence that AlphaGo’s win isn’t that shocking — perhaps even right on schedule — and therefore shouldn’t cause you to update much on overall AI progress:
Comment from mjn:
Fwiw, the point where the Go curve massively changes slope is when Monte-Carlo Tree Search (MCTS) began to be used in its modern form. I think that’s been an underreported part of AlphaGo’s success: deep networks get the lion’s share of the press, but AlphaGo is a hybrid deep-learning / MCTS system, and MCTS is arguably the most important of the algorithmic breakthroughs that led to computer Go being able to reach expert human level strength.
- Also: Ken Jennings welcomes Lee Sedol to the Human Loser Club. And: Do the Go prodigies of Asia have a future? (H/t Tyler Cowen.) These articles basically write themselves.
Also from Tyler: It was only a matter of time before Facebook began to hire reporters. And: “Will all of economic growth be absorbed into life extension?“:
Some technologies save lives—new vaccines, new surgical techniques, safer highways. Others threaten lives—pollution, nuclear accidents, global warming, and the rapid global transmission of disease. How is growth theory altered when technologies involve life and death instead of just higher consumption? This paper shows that taking life into account has first-order consequences. Under standard preferences, the value of life may rise faster than consumption, leading society to value safety over consumption growth. As a result, the optimal rate of consumption growth may be substantially lower than what is feasible, in some cases falling all the way to zero.
And: “Things organized neatly“. Finally, an oldy but a goody: Tyler’s alternative to the signaling theory of education. Doesn’t really account for the sheepskin effect, but is consistent with the fact that no one just uses 7th-grade standardized test scores. Caplan would say this is because completion of a degree is a hard-to-fake signal of grit/conformity/conscientiousness in a way that test scores cannot be. The debate continues here, here, here, and here.
“Here’s why the next SpaceX launch isn’t just about the booster landing“:
A company called Bigelow Aerospace wants to build space stations for the government and hotels for private customers that will inflate like balloons once they reach outer space. Bigelow’s inflatables have the potential to revolutionize spaceflight by providing lighter, and much larger, places to live in space….
NASA intends to find out and has signed a $17.8 million contract with Bigelow to do so. As early as April 8 a deflated module will launch inside the trunk of a SpaceX Dragon spacecraft. The space agency has agreed to attach a test module to the International Space Station, inflate it, and over the course of two years determine if such a contraption can work in space. Crew won’t live in it—inflatables remain too experimental to risk life and limb. But if the module holds up, NASA will invest more money into the technology.
…[Bigelow] turned to NASA and developed the Bigelow Expandable Activity Module, or BEAM. It offers Bigelow a chance to further refine its systems, and it gives NASA a chance to kick the tires on a promising but untested technology. NASA will observe the module to see how it handles radiation from space, measure its thermal properties, and test other environmental conditions, such as noise.
If all goes well, NASA will eventually will let its astronauts go inside. “When an astronaut steps inside a Bigelow habitat for the first time, that will be a big moment for the company,” Gold said.
SIze and weight are the two biggest hurdles to putting lots of stuff into space. It took NASA dozens of costly space shuttle launches to assemble the International Space Station piece by piece. Lacking a rigid structure, inflatables can be folded inside the limited diameter of a rocket fairing. Once in space they can be expanded to create a massive amount of volume. There is also a considerable mass savings.
As for radiation, Gold said Bigelow’s inflatables should be as good, or better than the space station in terms of limiting radiation. Unlike the station’s metallic shell, which scatters radiation from solar flares, the non-metallic skin of the expandable module should reduce this scattering effect.
Then there is debris. “The major concern I hear is, if it’s a balloon, will it pop?” Gold said. “Quite the opposite.” The expandable’s kevlar-like weave should be at least as protective as the station’s aluminum hull when it comes to orbital debris, Gold said. Because of this fabric-like material, Gold also said Beam is likely to prove a quieter location than the notoriously noisy station interior.
Even while Bigelow is working on Beam, it has its eyes set on bigger things, what it calls its B330 module. It is so named because it would offer 330 cubic meters of interior space. By comparison the station has about 425 cubic meters of habitable volume.
Bigelow says this module might support its space hotel ambitions, but also could be used for all manner of spaceflight endeavors. NASA might want to use one as a space station near the Moon, and it could also be modified to support operations on the surface of the Moon. Last July the space agency signed a “NextStep” agreement with Bigelow to study utilization of the B330 module for deep-space exploration. A successful Beam would go a long way toward validating the company’s larger ambitions.
Cross cnoidal swell:
- Autism is a very wide spectrum. Parents with subclinical but autistism-like traits are more likely to have autistic children.
- John Ioannidis on hi-jacking of evidence-based medicine. He bemoans the lack of public funding for clinical trials to complement/counter private funding bias. Also cautions against the value of Cochrane Collaboration reviews!
- Deep-sea creatures are hard to study because they generally die quickly at lower pressures. The Abyss Box keeps a small volume of water at high pressure to keep them alive.
- Creating an artificial bacteria by stripping out genes one at a time to find the minimum genome. [Article.] (H/t Sean Carroll.)
- I generally find Robin Hanson’s outside view reasoning about AI timelines more compelling than alternative takes. Here is another good analogy which has pessimistic implications for de novo AI vs. emulated AI.
- A much more sensible, albeit overly libertarian, take on bullshit jobs.
- I’m bummed that I only just now discovered the defunct Selected Papers Network from this (still live) blog post from Tim Gowers.
- Bespoke magnets. (HN Comments.)
- Pessamistic news from Google on self-driving car timelines.
- Slightly more data comes in on the diphoton excess. I’m feeling slightly less sure about my 10% prediction…
- Coming CRISPR applications. (H/t Gwern.) Also from Gwern: “Mutation and Human Exceptionalism: Our Future Genetic Load“.
“A hand-drawn graph of the absolute value of the complex gamma function, from Tables of Higher Functions by Jahnke and Emde” (1909):
- Don’t forget about SciRate!
Project West Ford:
At the height of the Cold War, all international communications were either sent through undersea cables or bounced off the natural ionosphere. The United States Military was concerned that the Soviets might cut those cables, forcing the unpredictable ionosphere to be the only means of communication with overseas forces. So, a ring of 480,000,000 copper dipole antennas (1.78 cm long needles, 25.4μm  / 17.8μm  in diameter) was placed in orbit to facilitate global radio communication. The length was chosen because it was half the wavelength of the 8 GHz signal used in the study. The dipoles collectively provided passive support to Project Westford’s parabolic dish (located in the town of Westford) to communicate with distant sites. In 1958, at MIT’s Lincoln Labs, Walter E. Morrow started Project Needles.
A failed first attempt launched on October 21, 1961; the needles failed to disperse. The project was eventually successful with the May 9, 1963 launch, with radio transmissions carried by the man-made ring. However, the technology was ultimately shelved, partially due to the development of the modern communications satellite and partially due to protests from other scientists. The needles were placed in medium Earth orbit between 3,500 and 3,800 kilometres (2,200–2,400 mi) high at 96 and 87 degree inclinations and contributed to Earth’s orbital debris. British radio astronomers, together with optical astronomers and the Royal Astronomical Society, protested this action. The Soviet newspaper Pravda also joined the protests under the headline “U.S.A. Dirties Space”. The issue was raised in the United Nations where then US Ambassador to the UN Adlai Stevenson defended the project. Stevenson studied the published journal articles on Project West Ford. Using what he learned on the subject and citing the articles he had read, he successfully allayed the fears exhibited by the vast majority of UN ambassadors from other countries. He and the articles explained that sunlight pressure would cause the dipoles to only remain in orbit for a short period of approximately three years. The international protest ultimately resulted in a consultation provision included in the 1967 Outer Space Treaty. Fifty years later in 2013, some of the dipoles that did not deploy correctly still remain in clumps which make up a small amount of the orbital debris tracked by NASA’s Orbital Debris Program Office.
As of 2013, 46 clumps of needles are still in orbit, and occasionally re-enter.
- Wikipedia articles near you.
- New numerical results concerning the prime k-tuples conjecture, for which twin primes (cf. Yitang Zhang) are a special case. See Terrance Tao for detailed commentary. (H/t Peter Woit).
Praise for “y’all” as English’s natural second-person plural pronoun. My favorite wrinkle in this is this is that “all y’all” can be used when the pronoun refers to people who are not being addressed, either because they are not physically present, or because they are simply out of ear shot. In principle this is super useful for distinguishing between (say) the two people you’re speaking with from a larger group they are members of. “Are y’all going to all y’all’s conference?” Sounds silly if you’re not used to hearing it, but very natural and logical if you are.
Unfortunately, usage isn’t really consistent. People often use “all y’all” just for emphasis (i.e., a second-person very-plural pronoun), such as when addressing a crowd.
Don’t try this at home.
A US farmer risks his life to stop a wildfire that broke out in a hayfield in Weld County, Colorado. Eric Howard uses his tractor and a set of discs to plough up a firebreak just a few feet from the scorching blaze. After a thunderstorm brought lightning to a rural area near Brighton, Colorado the fire broke out and quickly consumed 30 acres of an 80 acre field. Mr Howard’s actions were designed to prevent the fire’s spread as it continued its path across the land.
Neil Turok goes very short on string theory, which I would not have expected him to express so publicly.
I think what people are sort of expressing is that we haven’t had a big revolution in physics. String theory was hoped for to be that revolution in the 1980s but it hasn’t really panned out in the sense that it hasn’t given a single prediction. Instead it’s given us a huge collection of theories where, if you like, there’s no overarching theory to tell which particular version of string theory is the one that describes the world. It’s almost self-destructed, I would say because it turned out to be not just one theory but this vast collection of theories which could all give different descriptions of the world.
So I think that sort of theoretical catastrophe, as I view it — meaning the logical pursuit of quantum mechanics and relativity over a hundred years was tremendously successful at some level but finding its own successor theory, it hasn’t been successful. I think that is also laying the ground for some sort of revolutionary change in the sense that we basically will have to go back to the founding principles. It looks like the founding principles of modern physics — quantum theory and relativity — have played out and they have not given us the answers we need. And so we have to go back and question those founding principles and find whatever it is, whatever new principle will replace them. So matching these great puzzles posed by the observations are equally great puzzles in our fundamental theories. And so that is just a wonderful thing to contemplate in itself. I mean, partly people become very pessimistic and say, oh my god, I’ve devoted 50 years of my life to studying this incredibly technical and difficult theory and now I find it’s blown up in my face, it’s not giving any predictions at all…and so some people talk about the multiverse where the universe would be wild and chaotic on large scales and almost anything you could imagine would actually exist somewhere in the universe. I mean, this is literally a scenario which became very popular among a category of physicists, that there is a multiverse out there. Yet the evidence is exactly the opposite. That, as we look around us, things could not be simpler. There’s no evidence for chaos on large scales in the universe. It’s totally the opposite. It’s pristine, elegance, minimalism is all we see.
So, I think this is a very, very exciting time to be doing theory. The challenge is enormous. The clues are enormous. We’re waiting and we’re preparing and we’re encouraging people to take radical leaps.
(H/t Peter Woit.)
- Wikipedia Zero: Life — and porn — find a way.
The Honda Civic LX features some pretty advanced lane-assist. Trumped as self-driving, but note this clarification from ben1040:
This looks like the “Sensing” feature that Honda has implemented on some of their other vehicles. I just bought a 2016 Accord that does the same thing — there’s a camera mounted on the windshield, another camera in the front grille, and a radar sensor on the front bumper.
Calling it “self-driving” is kind of a misnomer and I think the article kind of blows it out of proportion.
It will track the car in front of you and keep a safe following distance, keeping either the maximum cruise control speed you’ve set, or whatever speed the vehicle ahead of you is driving, whichever is slower. It will accelerate or brake accordingly. It will also attempt to stay in the lane by using the onboard cameras for tracking the lane markings.
The lane keeping assist is not nearly as autonomous as the article makes it out to be. It does not like to work on sharper curves on the freeway, for one — the system will disengage and tell you to steer manually. It still wants you to keep your hands on the wheel. It must be looking for very very subtle movements on the wheel, because the system will yell at you if you take your hands off the wheel for longer than 10-15 seconds.
All in all it’s a pretty cool feature for longer road trips (keeping in your lane can just get kind of tiring, even with cruise control) but it’s not the sort of autonomous driving that the article here paints it out to be.
Uber’s tech allows for a more efficient utilization of driver’s time (and car mileage) than traditional taxis:
On average, the capacity utilization rate is 30 percent higher for UberX drivers than taxi drivers when measured by time, and 50 percent higher when measured by miles, although taxi data are not available to calculate both measures for the same set of cities….Four factors likely contribute to the higher utilization rate of UberX drivers: 1) Uber’s more efficient driver-passenger matching technology; 2) Uber’s larger scale, which supports faster matches; 3) inefficient taxi regulations; and 4) Uber’s flexible labor supply model and surge pricing, which more closely match supply with demand throughout the day.
- Minecraft as an AI testing ground.
The most EA article ever written:
The admission of a dying patient to the ICU when organ donation may be possible is of considerable community benefit, yielding an average of over seven times the QALYs per ICU bed-day compared with the average benefit for ICU patients expected to survive. When it is possible to offer end-of-life care in the ICU, it should not be denied on the basis of concerns about lack of benefit or inappropriate use of resources.”
(H/t Hauke Hillebrandt.) Note that many (most?) terminally ill patients are not eligible to donate organs, such as most metastatic cancer patients.
- Sean Carroll covers a recent paper on the LIGO detection and primordial black holes as the dark matter.
- Andrew Gelmen on the dispute over the reproducibility study in psychology. See many more good links on this from Scott Alexander.
- Immune desensitization for increasing kidney transplant compatibility.
- Back in November, the asteroid sample-return probe Hayabusa 2 got a great shot of the Earth and the Moon in the same image.
- c/10 glider discovered in Conway’s Game of Life. Amazing that they are still discovering small fundamental objects. I guess 2^100 is a big number.
A modern version of the needed (as estimated in 1976) and actual fusion power funding:
(I commented on this in my last post.) There must be another side to this story, but I haven’t heard it. On it’s face, it implies that all of our energy efficiency, carbon reduction, and energy independence efforts — all of it for decades — has been a huge waste of time. It hurts to think about.
As the zygote develops into an embryo, cells follow a mostly-one-way process of specialization, from totipotent stem cells to each specialized type. Apparently, true (“naïve”) pluripotent stem cells have been derived from human embryos for the first time. Previous stem cells were either “embryonic”, “cord blood”, “adult” cell (progressively further along in the specialization chain) or were normal skins cells tricked into behaving, imperfectly, like pluripotent stem cells (i.e., not naïve). Other facts that were new to me: The cells of the placenta correspond genetically to the fetus, not the mother (although it meshes with maternal cells, so that the physical organ contains a layer of maternal cells that intermingle with fetal cells). The division between placental and non-placental cells is, rather sensibly, the first step of specialization in the developing fetus. The chain of increasing specializatio can be broken up like this:
- Totipotent (a.k.a. omnipotent) stem cells can differentiate into embryonic and extraembryonic cell types. Such cells can construct a complete, viable organism. These cells are produced from the fusion of an egg and sperm cell. Cells produced by the first few divisions of the fertilized egg are also totipotent.
- Pluripotent stem cells are the descendants of totipotent cells and can differentiate into nearly all cells, i.e. cells derived from any of the three germ layers.
- Multipotent stem cells can differentiate into a number of cell types, but only those of a closely related family of cells.
- Oligopotent stem cells can differentiate into only a few cell types, such as lymphoid or myeloid stem cells.
- Unipotent cells can produce only one cell type, their own, but have the property of self-renewal, which distinguishes them from non-stem cells (e.g. progenitor cells, muscle stem cells).
- A mountain range on Pluto has been dubbed “Cthulhu Regio” by the New Horizons analysis team. Nerds.
- BioArxiv gets pushed at recent biology conference, and appears to be making inroads. Interestingly, biologists have not yet formed a consensus that posting to a preprint server establishes priority, although that seems inevitable.
- George Church’s Veritas Genetics is now offering $1k whole genome sequencing.
- Does Moore’s law slowdown make developing superconducting computers sensible? (H/t Robin Hanson.)
The joint European-Russian mission ExoMars lifted off.
The main objective of Schiaparelli is to demonstrate its landing system. (The European Space Agency’s last attempt to land on Mars — the Beagle 2 spacecraft, which accompanied the Mars Express orbiter in 2003 — failed.)
- The absurd explosion in the cost of constructing Nuclear power plans in the US wasn’t inevitable. Many other countries kept costs reasonable.
LaTeX in commentsInclude [latexpage] to render LaTeX in comments. (More.)
- Legendre transform (4)
- Branches and matrix-product states (6)
- Jess Riedel This prompted me to take an old email where I listed some models of branching... – Oct 12, 3:16 AM
- interstice Thanks. Do you know of any systems, toy or real, where we can identify a... – Oct 11, 10:06 PM
- Jess Riedel Great question. It is not obvious that we should define branches from the tensor structure... – Oct 10, 8:00 PM
- interstice Right, so to argue that the branches evolve independently, one must first define what branches... – Oct 10, 4:19 PM
- Jess Riedel You're right that the criteria for defining branches I give in the paper doesn't make... – Oct 05, 2:33 PM
- interstice I have a stupid(?) question. The criteria you give in the paper doesn't involve time-evolution... – Oct 05, 12:36 PM
- Comments on Weingarten's preferred branch (1)
- Links for August-September 2017 (2)
foreXiv by C. Jess Riedel is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.