* 21 entries including: US Constitution (series), fragile AI (series), internet of things (series), genomic probe of protein structures, getting too hot too fast, pandemic boosting green energy, fly connectome, & life at extremes.
* NEWS COMMENTARY FOR MAY 2020: It appears that time is stuck in a loop in 2020. The world remains mired in a COVID-19-driven lockdown, while the USA, under President Donald Trump, lurches towards an election in November.
Resistance against the COVID-19 lockdown has been strong in the USA, with loud objections, armed protests, some deranged acts of violence, and vandalism of government buildings. The lockdown resistance rests on shaky foundations; it's based on minimizing the impact of the virus, maximizing the economic problems of the lockdown, then claiming that the USA ought to simply end the lockdown and move on. That's preposterous; the virus, left unchecked, would bring the economy to its knees anyway. We're stuck with it. Open up movie theaters? What for? People are watching Netflix, and too few will be willing to go to the movies to make theaters much of a paying proposition. Indeed, COVID-19 may kill off movie theaters for good.
As far as the November election goes, the simplest thing to be said is that things are not looking good for Donald Trump. One Matthew Dowd wrote on Twitter:
BEGIN QUOTE:
Exactly what can the incumbent President run on? He can't run on the economy. He can't run on healthcare. He can't run as a moral leader. He can't run on bringing the country together. He can't run on leaders he put in power. He can't run on justice. Exactly what can he run on?
END QUOTE
The general reply was: "Bigotry and hate." Even that doesn't seem to be working well any longer; increasing numbers of Twitter posters who say they voted for Trump in 2016 say there's no way they'll do it in 2020.
Lincoln once said of a general that had suffered a defeat, and was no longer functioning: "He's acting like a duck that's been hit on the head." Trump barely pretends to take the pandemic seriously, encourages resistance, and attempts to shift blame to everyone else -- notably China, which is part of the basis of his re-election strategy. Trump has been talking about China paying reparations; CNN's Fareed Zakaria suggested that would happen the day after Mexico pays for the border wall. As far as the tanked economy goes, Trump is claiming that he will lead an economic recovery better than Joe Biden will. Such a strategy is not likely to appeal to anyone with sense. The only people who think Trump is right in the head, are not right in the head.
* Biden is being afflicted by sexual-assault claims made by an ex-staffer named Tara Reade, but they're not causing him much trouble. Reade has no real evidence to back up her claims, she's changed her story on a regular basis, and nobody else has made such claims about Biden. Yes, he has a reputation for being a "huggy" person, but he's been read the riot act on that, and has made amends.
Reade appears to be backed by sorehead Bernie Sanders fans, who believe that if they can derail Biden, Bernie will get the Democratic nomination instead. Attempting to sort through the lunacy of that idea is not a useful exercise; enough to say that the Reade backers on the Left are few in number, with Reade's story being amplified by Trump backers. Nobody else believes Reade. Biden came out and said that she had a right to be heard, and that what she said was not true.
In response to continued agitation over the matter, Biden said that those who believe Tara Reade should not vote for him. That dismissed the matter: he knows perfectly well that everyone backing Reade wouldn't vote for him in any case. If something new comes up, maybe Reade will start causing Biden real trouble. However, if she'd had anything of substance, she would have used it by now. Reade is going nowhere.
Once Biden is formally nominated, it is likely that Reade will effectively disappear. [ED: She dropped out of sight well before that.] Nothing sticks on him; the Trump campaign plays up his famous gaffes, but Biden apologizes, everyone shrugs and moves on. We're used to them. In the meantime, Trump churns out an endless stream of trash on Twitter -- recently having gone so far as to accuse MSNBC TV host Joe Scarborough of murder.
Scarborough had once been a Member of the House; in 2001, one of his young interns had a heart attack, to fall and hit her head. There was no evidence of foul play in her death, but that didn't slow Trump down in his accusations. The husband of the dead woman wrote a heart-breaking letter to Twitter to ask the company to remove the tweets; they just flagged them instead. We're used to such things from Trump, too, but it's hardly done much to enhance his prestige that he conducts himself so badly while Americans are dying.
Biden is continuing with his virtual campaigning, and doing very well at it. It appears likely there will be at least a partly virtual convention, which sounds like a major technical challenge. [ED: It was fully virtual, and went well.] It has been pointed out that Biden's Twitter following is much smaller than Trump's -- but that's missing the fact that Barack Obama has the biggest Twitter following of all, and he retweets Biden.
By the way, Tara Reade is often pushed by Twitter posters with clearly African names, which leads to the question of: "Why do you care?" I was very puzzled for a time, but it turns out that the Kremlin has a number of troll farms in African countries, which explains a lot of things. The Afro-trolls are particularly fond of agitating black Americans, playing up the killings of blacks at police or vigilante hands.
* As per the comment above on the fadeout of cinemas: they were fading out before the pandemic, and the pandemic has all but done them in. However, drive-in movies have been making a comeback in the era of COVID-19. Back in their prime, drive-ins featured a speaker on a post that could be hung in a car window, but it appears they use short-range FM radio now. Viewers can listen using their car sound systems.
Incidentally, Android smartphones often have FM receivers, though smartphone vendors don't like to point it out, lest they divert users from audio streaming services. One can obtain apps to get to the FM radio on a smartphone; it seems wired headphones are required, since they provide the antenna. It also seems Apple doesn't support FM radio on iPhones, again because streaming services.
In any case, the pandemic has been a boon to virtualization. Some US states are no longer conducting driver's tests to give teens a driver's license, apparently delegating testing to parents. It's an experiment, driven by necessity; it will be interesting to see how it works out. In another virtualization exercise, here in Colorado recipients of SNAP food stamps can use them to buy food online. On a personal note, my nephew Graham just graduated from Baylor University in Texas; since they couldn't have a graduation ceremony, they listed all the graduating students on the stadium Jumbotron screen, and put images of them on Facebook.
In the meantime, Trump has been trying to undermine the movement by a number of states to vote by mail. He claims it encourages cheating, which is nonsense: it's been done here in Colorado -- and some other states -- since ever, without problems. It appears that Trump is trying to step on mail voting as a means of vote suppression. That's also absurd, since to the extent undermining mail voting suppresses votes, it suppresses Republican votes at least as much as Democratic votes. I doubt that he'll accomplish much more than increase the vote against him.
It would seem the real solution would be internet or email voting, but that's predicated on implementing a robust system of national ID -- both for online use and street use, one implies the other. There's a screaming need for that, since it's hard to implement virtual interactions without it. However, there's going to be a lot of resistance to the idea of national ID, and the national data system that would back it up.
One last comment on virtualization: many churches have embraced drive-in religious services, presumably using short-range FM, or virtual services. That led THE ONION, the well-known satirical website, to run an article titled: "Teleconferencing Pastor Requests Any Worshiper Currently Speaking In Tongues Go On Mute". THE ONION tends towards the crass, but that was cute.
COMMENT ON ARTICLE* WINGS & WEAPONS: Raytheon is now in advanced testing of its "AN/ALQ-249 Next-Generation Jammer / Mid-Band" pod, intended to replace the long-standing AN/ALQ-99 jammer pod carried on the Boeing EA-18G Growler electronic countermeasures (ECM) aircraft. The AN/ALQ-99 leaves something to be desired in terms of reliability and self-test; it is also somewhat "draggy", and interferes with the Growler's active electronically-scanned array (AESA) radar.
The AN/ALQ-249 will also have an AESA system, capable of agile and sophisticated electronic attacks against multiple hostile "emitters" simultaneously. It will have greater range, and will be relatively easily upgraded through its "modular open systems architecture (MOSA)". The AN/ALQ-249 uses a ram-air turbine to provide additional power when in operation; so does the AN/ALQ-99, but the spinner for the AN/ALQ-99 is on the nose of the pod, while the AN/ALQ-249's is hidden by doors when not in use, reducing drag.
The modular design will support future "low-band" and "high-band" variants of the pod. A Growler might carry all three variants of the pod, depending on mission.
* As discussed by an article from AVIATIONWEEK.com ("Hypersonic Weapons Reclaim The U.S. Army's Long-Range Offense" by Steve Trimble, 18 October 2019), in the 1990s the US Army acquired the "Advanced Tactical Missile System (ATACMS)" for the "long-range precision fires (LRPF)" mission. ATACMS is based on a surface-launched, solid-fuel missile with a design range limited by a now-defunct treaty to 300 kilometers (185 miles).
The Army is rethinking the LRPF mission, with plans to field an ATACMS replacement -- the "Precision Strike Missile", with a range of up to 800 kilometers (500 miles) by the middle of the next decade -- as well as a cannon-launched projectile with a range of 1,600 kilometers (1,000 miles), and a "Long-Range Hypersonic Weapon (LRHW)" with an unclassified range described only as "thousands of kilometers."
This projected arsenal will allow to the Army to give up its reliance, established in the postwar period, on the US Air Force to support long-range strike. In fact, Army planners assume that improvements in adversary air defenses will effectively neutralize aircraft in defended airspaces. Brigadier General John Rafferty, the Army's LRPF commander, says:
BEGIN QUOTE:
Our access really is challenged. And our aircraft have become incredibly vulnerable I think there is a realization across the joint force that surface-to-surface fires by themselves don't eliminate the A2/AD [anti-access / aerial denial] complex, but they enable the Air Force and the maritime component to penetrate and then disintegrate the A2/AD complex.
END QUOTE
The Army envisions that, in a future conflict, "strategic fires battalions" will launch an opening salvo of LRHWs costing millions of dollars each at hardened bunkers and over-the-horizon radars, along with a barrage of projectiles priced in the hundreds of thousands of dollars each from a future "Strategic Long-Range Cannon." The emphasis would be on targeting launchers and communications centers, to pave the way for Air Force and Navy air power.
The Army launched the LRHW prototype program by selecting Huntsville, Alabama-based Dynetics to build 20 hypersonic glide bodies. The batch will be split between the LRHW; the Navy's "Intermediate-Range Conventional Prompt Strike", and the Air Force's "Hypersonic Conventional Strike Weapon (HCSW)". All three programs are leveraging off the same bi-conic glide body. The Army and Navy weapons will both use a two-stage rocket booster, while the USAF HCSW will use a small, single-stage motor. Lockheed Martin is partnering with Dynetics to support assembly, integration and test of the common hypersonic glide body.
To date, only the Army has committed to fielding operational hypersonic weapons. Until last year, the Army planned to deploy a single LRHW in a fixed-site launcher by 2023 as a limited response to the planned deployments in 2020 of Russia's Avangard and China's DF-17 boost-glide weapons -- according to Robert Strider, deputy director of the rapid capability and critical technologies office (RCCTO). Now the Army plans to field a battery of six LRHWs and three transporter-erector launchers in four years. The schedule is dependent on flight tests and Congressional funding.
* As discussed by an article from JANES.com ("MBDA Deutschland Pitches Air-Launched Enforcer Missile" by Gareth Jennings 27 November 2019), the German arm of European defense giant MBDA is now promoting the "Enforcer" shoulder-fired missile, which was initially test-fired in 2016.
Enforcer is fired from a disposable launch tube. It has a weight of about 9 kilograms (20 pounds), a range from 100 to 2,000 meters (110 to 2,200 yards), and a day-night "fire & forget" guidance, it appears with infrared imaging. It is effective against soft and light armored targets.
MBDA Deutschland is also promoting the "Enforcer Air", an air-launched version appropriate for small drones or light attack aircraft. One MBDA official called it the "little brother" of the MBDA Brimstone air-to-surface missile. It could be drop-launched, or launched from underwing rails.
COMMENT ON ARTICLE* PROBING PROTEINS WITH GENOMICS: As discussed by an article from SCIENCEMAG.org ("Mutant Genes Could Supercharge Efforts To Decipher Protein Structures" by Robert F. Service, 18 June 2019), proteins generally have elaborate structures that are hard to nail down, demanding expensive instrumentation and a lot of patience. Now a research team has figured out a way to use genetic and biochemical techniques to do the job, at far less expense. In addition, unlike traditional methods that have to examine proteins in crystals or solution, the new scheme can also reveal the natural shape of proteins inside cells as they function,
The traditional means of determining protein structure is through x-ray crystallography. It starts with mass-producing the protein molecules, persuading them to form an orderly crystal, bombarding the crystal with x-rays, and then recording the pattern of diffraction of the x-rays. The diffraction pattern is analyzed to determine the position of each atom. Other approaches, including nuclear magnetic resonance spectroscopy and cryoelectron microscopy, also require large amounts of a protein; they are similarly expensive and laborious.
Some work has been done with software that attempts to second-guess the shape of a protein from its sequence of amino acids and probable interactions between atoms. However, so far, the computational approach lags experimental methods. One new trick is to compare the same protein in multiple species to find pairs of amino acids that have evolved together even though they are far apart on the protein's linear sequence. That's a clue that the two amino acids are close together and interact in the folded 2D protein molecule. According to Debora Marks -- a systems biologist at Harvard Medical School in Boston -- says that this approach only works if researchers can identify proteins that are shared by many organisms, but different enough across them to identify multiple pairs of amino acids evolving in parallel.
Now, a team led by Marks and a separate team led by Ben Lehner -- a geneticist at the Barcelona Institute of Science and Technology in Spain -- have independently come up with an idea for finding interacting amino acids within a protein by systematically mutating each amino acid, then tracking how the changes alter the protein's function, such as an ability to bind to another molecule. Instead of inspecting a protein that has evolved across a range of organisms, they in effect evolve the protein itself.
Both groups built from work performed on a bacterial protein fragment named "GB1" performed by a team led by Ren Sun, a systems biologist at the University of California, Los Angeles, In 2014, Sun's team reported creating more than half a million copies of the GB1 gene, each with one or two of its 56 amino acids changed. From there:
Marks's and Lehner's groups realized they could leverage off the binding data of the single and double mutants to determine which amino acids interact most strongly -- and so are likely to sit next to each other in the protein's 3D structure. By tracking dozens of such occurrences and feeding the results into a structure prediction program, the teams computed the shape of GB1's main backbone to within a few angstroms of the resolution of the already known experimental x-ray structure.
The two teams have also demonstrated their technique worked with other small proteins and an RNA with analogous available data. The same approach may be more troublesome for proteins with hundreds or thousands of amino acids, since the number of mutant proteins that must be made increases exponentially as the proteins grow -- with the data analysis burden increasing as well.
However, Marks is not discouraged, since early indications suggest the technique can solve structures using only a fraction of all possible mutants. Lehner's team says the approach could even work using measures of stability for proteins that lack known binding partners, Lehner feels assured that scanning mutations by the thousands may offer new insights into both proteins themselves, and how their structures affect health in living cells.
COMMENT ON ARTICLE* WAY TOO HOT: As discussed by an article from SCIENCEMAG.org ("Lethal Levels Of Heat And Humidity Are Gripping Global 'Hot Spots' Sooner Than Expected" by Warren Cornwall, 8 May 2020), a new study has shown that, as global temperatures rise, there are a growing number of "hot spots" where heat and humidity are reaching levels that will kill. Previously, nobody thought that would happen for decades, but Colin Raymond -- a postdoctoral researcher at NASA's Jet Propulsion Laboratory who led the study -- says: "Previous studies projected that this would happen several decades from now, but this shows it's happening right now."
Lethal heat waves are not all that new. A 2003 heat wave, for example, killed more than 70,000 people in Europe, when outdoor temperatures reached more than 40 degrees Celsius (104 degrees Fahrenheit). Of course, it's not just the heat that kills; at high humidities, sweat doesn't evaporate, and the body can't cool itself. A better indication of the issue is "wet bulb" temperature -- the lowest temperature to which air can be cooled via evaporation.
At wet bulb temperatures above 35C (95F), researchers estimate that even fit people will overheat and possibly die within 6 hours. That equates to a dry heat of 71C (160F). In the heat wave that swept over Europe, wet bulb temperatures hit 28C (82F). Models show that climate change is likely to make such extreme condition more common in places such as southwest Asia, India, and China -- but those models only provide estimates for large blocks of land.
Raymond wondered whether the coarseness of the models might cause them to overlook specific hot spots where geography and weather are already collaborating to create intolerable conditions. He and his team accordingly went through 39 years of hourly data from almost 8,000 weather stations on six continents, dating back to 1979. They found a number of individual spots -- including shorelines along the Persian Gulf and river valleys in India and Pakistan-had crossed the 35C wet bulb threshold, though only for an hour or two at a time. And in 2017, wet bulb conditions topped 30C a thousand times.
The effort uncovered hot spots in other places as well, including Mexican towns near the Gulf of Mexico and the Gulf of California, and the coastal city of San Francisco in Venezuela. Areas in the Caribbean, West Africa, and southern China also had extreme readings. The USA wasn't exempt, either, with the Southeast seeing extreme conditions dozens of times -- mainly near the Gulf Coast in east Texas, Louisiana, Mississippi, Alabama and the Florida Panhandle. The worst spots were New Orleans and Biloxi, Mississippi. Such conditions also extended inland into Arkansas and along the southeastern coastal plain.
Unsurprisingly, incidents tended to cluster on coastlines along confined seas, gulfs, and straits, where seawater evaporates into hot air, resulting in highly humid conditions. In some areas further inland, moisture-laden monsoon winds or wide areas of crop irrigation appear to play the same role.
To prevent from being misled by errant weather stations, Raymond's team compared the data with sea surface temperatures measured by satellites and air temperatures measured by weather balloons near the Persian Gulf, where the temperatures were highest. The study was innovative in factoring in heat and humidity.
Elfatih Eltahir -- a hydrologist and climate scientist at the Massachusetts Institute of Technology who forecasts extreme weather in Asia -- says that the study is disturbing, suggesting that we have been underestimating how fast such conditions will be upon us. He says: "In reality, maybe they are happening faster [than we think]."
COMMENT ON ARTICLE* INTERNET OF THINGS (6): The Internet of Things implies a monstrous security challenge. Break-ins of big data systems happen all the time; break-ins of IoT systems have become appallingly common, in large part because it's been so easy to crack them. There may not be much point in cracking any one IoT device, but they can be used to penetrate more important systems, and to make other kinds of trouble.
David Palmer -- director of technology for Darktrace -- says: "We've seen corporate espionage between suppliers inside a power station. One supplier was using [their] access within the network to look at the performance characteristics of another supplier's equipment." His firm also discovered an attack on fingerprint readers that controlled access to a luxury-goods factory, and malware that spread through a hospital department after infecting a connected fax machine.
Some such incidents have made headlines. In 2016, millions of people in the USA found themselves struggling to reach a number of prominent websites, including those of Twitter, Amazon, Netflix, and Reddit. The culprit was a piece of IoT-focused malware named "Mirai". By exploiting a list of default usernames and passwords, which most users don't bother to change, Mirai had infected hundreds of thousands of connected devices, from smart energy meters to home CCTV cameras and connected baby monitors.
Each infected gadget became part of a "botnet" -- a group of computers in under control of the Black Hats. The botnet then performed a "distributed denial-of-service attack" against Dyn, a company that helps maintain the routing information that allows browsers to reach websites. By flooding Dyn's servers with junk messages generated by the "zombie" devices, the botnet dragged them down, preventing them from responding to legitimate requests.
Since IoT devices often interact with the physical world, their poor security represents a potential threat to life and property. In 2015, a pair of security researchers from Twitter, a social network, and IOactive, a cyber-security firm, staged a demonstration for WIRED, a technology magazine, in which they remotely took control of a car while it was being driven -- turning on the stereo and windshield wipers, turning off the engine, applying the brakes. Cracking a car is troublesome because of a lack of system standards and limited communications capability -- but as cars become more connected, the threat grows.
Security researchers have demonstrated an ability to hack into medical devices, including pacemakers and insulin pumps. Hacking into a pacemaker would be an awkward way of killing someone -- but as we grow more dependent on IoT devices in general, they become more vulnerable to "ransomware", or malware that holds systems hostage. If the victims don't pay up, bad things happen to the system.
It doesn't appear that's happened yet, but it will soon. In 2018, 55 speed cameras in Victoria, Australia, were infected by a piece of ransomware that was actually targeting desktop computers. More recently, Avast Software -- a Czech cyber-security firm -- demonstrated how to install ransomware on a networked coffee machine, making it gush boiling water and constantly spin its grinder until the victim pays up.
Companies are thoroughly aware of the threat. A survey of managers by Bain & Company, a consulting firm, found that concerns about security were the single biggest barrier for companies considering adopting IoT technologies. Consumers are worried as well; a survey of 2,500 of them by Ernst & Young, a management consultancy, found that 71% were concerned about hackers getting access to their smart gadgets. [TO BE CONTINUED]
START | PREV | NEXT | COMMENT ON ARTICLE* AMERICA'S CONSTITUTION (103): The Warren Court was much bolder than the Kennedy Administration, making a number of significant decisions during JFK's term. The first major decision, MAPP V. OHIO, reinforced the Fourth Amendment's protections on search and seizure. On 23 May 1957, police in Cleveland, Ohio, wanted to search the house of one Dolltree "Dolly" Mapp -- a young woman who had got mixed up in the illegal gambling rackets -- in search of a gangster named Virgil Ogletree, who was wanted for questioning in connection with the bombing of the house of a rival racketeer. They didn't have a search warrant, so after talking on the phone to a lawyer, she told them they couldn't come in.
They came back later in force, waving a piece of paper that they claimed was a search warrant. Mapp tried to take it from them, with the result that she was handcuffed for being belligerent. They found Ogletree in a basement apartment and arrested him -- he would be cleared -- and also found betting slips, a pistol, and pornography. Mapp said the pistol and porn had been left behind by a previous tenant. The police tried to bust Mapp on a misdemeanor charge for the betting slips, but she was acquitted.
That failing, the authorities busted her for possession of pornography, which carried a penalty of from one to seven years in lockup. She challenged the conviction, on the basis that the police didn't have a valid search warrant -- indeed, they were never able to prove they did. The Ohio state supreme court affirmed the conviction, so Mapp appealed to SCOTUS.
SCOTUS judged in Mapp's favor, overturning earlier decisions that had concluded the 4th Amendment's prohibition of illegal search and seizure -- and the "exclusionary rule" that declared evidence obtained illegally was inadmissible in court -- only applied to the Federal government, not the states. Instead, Mapp's rights were protected by the 4th Amendment via incorporation through the 14th Amendment.
That would prove a controversial decision, critics saying there was nothing in the 4th Amendment that implied any such impractically severe restriction on the authorities. Besides, what if evidence that was excluded actually exonerated a defendant? Defenders of the MAPP decision say that it gives an incentive for the police to play fair, but there may be less drastic ways of doing that.
The second major decision was ENGEL V. VITALE. The story began when the New York State board of regents drew up a prayer that students could voluntarily recite. In 1958, a group of parents in Hyde Park, New York, led by one Steven Engel, who was Jewish, objected to the prayer, saying it violated the 1st Amendment separation of church and state. The group took the case to state court, suing William Vitale, the local school board president. The state supreme court judged in favor of Vitale, with the case then appealed to SCOTUS.
In 1962, SCOTUS judged in favor of Engel on a 6:1 vote, Justice Black writing for the majority:
BEGIN QUOTE:
We think that by using its public school system to encourage recitation of the Regents' prayer, the State of New York has adopted a practice wholly inconsistent with the Establishment Clause ... It is no part of the business of government to compose official prayers for any group of the American people to recite as a part of a religious program carried on by government.
END QUOTE
ENGEL V. VITALE has been endlessly contested, and also misunderstood, critics often claiming it banned prayer from schools. No, not really: students could pray if they felt like it, or even organize prayer groups, but the school administration had to take a hands-off attitude toward it. [TO BE CONTINUED]
START | PREV | NEXT | COMMENT ON ARTICLE* April was a slow month for space launches, very possibly because of the COVID-19 pandemic:
-- 09 APR 20 / SOYUZ ISS 62S (ISS) -- A Soyuz booster was launched from Baikonur at 0805 UTC (local time + 6) to put the "Soyuz ISS 62S" AKA "MS-16" crewed space capsule into orbit on an International Space Station (ISS) support mission. The crew included ASA astronaut and Soyuz commander Chris Cassidy (3rd space flight), plus RKA cosmonauts Anatoly Ivanishin (3rd space flight) and Ivan Vagner (1st space flight). The capsule docked with the ISS 6 hours after launch. They joined ISS Expedition 62 commander Oleg Skripochka, plus NASA astronauts Jessica Meir and Drew Morgan.
-- 09 APR 20 / NUSANTARA SATU 3 (FAILURE) -- A Chinese Long March B booster was launched from Xichang at 1146 UTC (local time - 8) to put the "Palapa N1" AKA "Nusantara Satu 2" geostationary comsat into orbit for Indonesia's Palapa Satelit Nusantara Sejahtera, a joint venture owned by Indosat Ooredoo and Pasifik Satelit Nusantara. The booster failed and the payload did not make orbit.
-- 22 APR 20 / NOOR -- An Iranian Qased (Messenger) booster put the "Noor (Light)" military surveillance satellite into low Earth orbit. The three-stage Qased satellite launcher used a combination of solid and liquid fuels.
-- 22 APR 20 / STARLINK 6 -- A SpaceX Falcon 9 booster was launched from Cape Canaveral at 1530 UTC (local time + 4) to put 60 SpaceX "Starlink" low-Earth-orbit broadband comsats into orbit. The satellites were built by SpaceX, each having a launch mass of about 225 kilograms (500 pounds). This was the seventh Starlink batch launch.
-- 25 APR 20 / PROGRESS 75P (ISS) -- A Soyuz booster was launched from Baikonur at 0151 UTC (local time + 6) to put a Progress tanker-freighter spacecraft into orbit on an International Space Station (ISS) supply mission. It hooked up to the ISS Zvezda module three hours after launch. It was the 75th Progress mission to the ISS.
COMMENT ON ARTICLE* FRAGILE AI (3): In a lab at the University of California, Berkeley, a robot arm searches through an assortment of objects -- a blue bowl, a spray bottle, a paperback book, and so on. It picks one object up, plays with it a bit, then puts it down and tries another one. The robot arm is linked into a deep-learning system; after several days of non-stop fiddling around, it gets a feel for these objects and what it can do with them.
Researchers assign the robot a goal -- for example, giving it an image of a tray with a few objects on it, and then ordering it to set up a tray in the same way. The machine improvises, and can adapt to work with objects it hasn't seen before. With practice, it can optimize its efforts. Chelsea Finn -- who worked at the Berkeley lab and is now continuing that research at Stanford University in California -- says: "Compared to other machine-learning techniques, the generality of what it can accomplish continues to impress me." Finn adds that this sort of "hands-on" learning gives an AI a much richer understanding of objects and the world in general: "Your understanding of the world would be much shallower than if you could actually interact with them."
The problem is that it's slow. Learning from simulated environments is much faster; the AlphaGo Go-playing AI played tens of millions of games against itself to acquire its dominating proficiency. Acquiring that level of experience with an interactive system would demand years of dedicated work by a single robot -- and in that time, the hardware might degrade, undermining the reliability of the data. As a result, there's an inclination to keep on using simulations. Simulators are continuously improving all the time, and researchers are getting better at applying lessons learned in virtual worlds over to the real.
Nonetheless, trying to make simulations perfectly realistic demands ever-increasing resources. Finn believes that the richness of real-world experience compensates for the relatively low speed of the trials. Her tool-using robot took a few days to learn a relatively simple task, but did it without much baby-sitting: "You just run the robot and just kind of check in with it every once in a while." She envisions a time when there's a network of robots around the world, left to their own devices to learn all the time. Juergen Schmidhuber says: "A baby doesn't learn by downloading data from Facebook."
A small child can also recognize new examples from just a few data samples. If given a picture book with a cartoon of a giraffe, she can easily recognize a giraffe at the zoo. This is because she has acquired a wide knowledge base, learning to recognize animals, and sees a giraffe as a distinctive variation on other animals. In AI, that concept is called "transfer learning", the ability to transfer knowledge acquired from previous rounds of training to new tasks. If we have a DNN that's trained to recognize animals in general, it should not have too much trouble learning how to recognize giraffes.
Taken to an extreme, transfer learning implies a DNN that can be trained to a specific case with a few examples, possibly only one. This is called "few-shot" or "one-shot" learning. If a small child has learned to recognize different animals, she may only need to see one image of a giraffe to know what a giraffe looks like. For an AI example, suppose we have a facial-recognition system that can identify people in a criminal database. It has already been trained on millions of faces, not necessarily those in the database, and so has a good chance of recognizing a particular face from one sample -- and then searching the database for that face. Still, such DNNs might be at a loss when confronted with something too far from their experience.
The fragility of AI makes it hard to know just what will work and what won't. Really successful AI systems, such as DeepMind's AlphaZero, have an extremely narrow domain of expertise. AlphaZero's algorithm can be trained to play both Go and chess, but not both at once. Anything it learns about Go is useless for playing chess. Finn points out that humans are not so inflexible: "If you think about it from the perspective of a human, this is kind of ridiculous."
There's a common belief among AI researchers that AI systems work much better if they have some reasoning ability, not just the ability to match patterns. AlphaZero, for example, didn't just collect examples of games; it also had a search algorithm to help it prune down its possible game moves. Google's Francois Chollet suggests that, as a next step, AIs should need to be able to devise their own algorithms. Humans, confronted with problems, figure out procedures to deal with them, so why not machines?
Chollet believes that combining pattern-matching with reasoning abilities would make AIs better at dealing with inputs outside of their comfort zone, he argues. Computer scientists have for years studied "program synthesis", in which a computer generates code automatically. Chollet thinks that deep learning combined with program synthesis could lead to systems with DNNs that are much closer to the abstract mental models that humans use.
AI researchers say they are making progress in fixing deep learning's flaws, but admit they're groping around in the dark. Berkeley's Dawn Song says: "If something doesn't work, it's difficult to figure out why. The whole field is still very empirical. You just have to try things." [END OF SERIES]
START | PREV | COMMENT ON ARTICLE* PANDEMIC & GREEN ENERGY: As discussed by an article from SCIENCEMAG.org ("Renewable Power Surges As Pandemic Scrambles Global Energy Outlook, New Report Finds" by Warren Cornwall, 30 April 2020), the COVID-19 pandemic has been a massive shock to the global system, with far-reaching consequences. One has been a drop in energy demand and related carbon emissions -- which, according to a report by the International Energy Agency (IEA), could well have wide implications for energy production in the world after the pandemic.
The drastic decline in energy use has not hit all energy producers equally. Fossil fuels, particularly coal and oil, have been hit hard -- coal having been in decline to begin with, and oil suffering from a fratricidal price war even before the pandemic began. Use of renewables, however, continues to rise as prices drop and installations keep coming online. Fatih Birol, executive director of the Paris-based IEA: "The energy industry that emerges from this crisis will be significantly different from the one that came before."
The numbers spell out the impact of the pandemic on the energy world:
Of course, yearly estimates can be off, but the first quarter of 2020 demonstrated the magnitude of the decline, with global energy demand falling by 3.8% -- much of that in March, when the pandemic really began to bite. Countries with tough lockdowns have seen a 25% drop in week-to-week energy demand as factories are closed and people stay home. Demand fell 18% in nations with partial lockdowns. Demand for electricity, a subset of total energy, has fallen as much as 20% in locked down countries, and daily patterns of electricity consumption resemble those usually seen on a typical Sunday.
Coal suffered painful losses in the first quarter, with an almost 8% drop compared with the first quarter of 2019. The decline was due to the lockdown in China, a major coal user; competition from natural gas and renewable energy; and a mild winter. Demand for oil fell by 5%, as car traffic was cut in half while air travel declined by 60% by the end of March. Coal and oil supply chains have been hammered. Renewable energy use, in contrast, grew 1.5% in the first quarter of 2020.
The IEA report suggested that the pandemic might also interfere with renewable energy projects, through lockdowns or supply disruption. What happens when the pandemic recedes depends on energy policies established by the world's governments. Birol encouraged them to put clean energy technologies "at the heart of their plans for economic recovery." Officials in South Korea, Germany, and the UK have indicated environmental concerns will inform recovery efforts.
Daniel Kammen -- an energy policy expert at the University of California, Berkeley, and a co-author of a proposal for a "green stimulus" initiative in the USA -- hopes there will be a change in mindset, thanks to the pandemic: "One version of the coronavirus crisis is it all eases, and we go back to what we were doing before. The other version of it is people say: 'Wow, I hadn't realized how bad things were.'"
COMMENT ON ARTICLE* INTERNET OF THINGS (5): The Internet of Things is a technological revolution, the dispersal and atomization of computing and sensing technology. Take, for example, the "Plastic Armpit (PA)", a collaborative effort between the University of Manchester in the UK, Advanced RISC Machines (ARM) of the UK, British-Dutch consumer giant Unilever, and Pragmatic -- the last being a UK firm focused on wearable electronics. The PA project is intended to develop a cheap chip capable of chemical analysis of gases and liquid. Such a device could have many uses; it got its cheeky name from the idea that it might be woven into clothing, to tell the wearer to take a bath. More broadly, it could be built into food packaging, to tell supermarkets and shoppers when a product goes bad.
The market potential for such a device is staggering. An estimated 259 million PCs were sold around the world in 2018; the number of smartphones at present is more than 2.5 billion. ARM -- which makes cheap, low-power processors for everything from smartphones to TVs -- runs its business on the assumption that there will be a trillion computers of all sorts in the world in 2035.
The goal of the PA project is to produce a cheap, robust, bendable, mass-producible computer with sensors and a wireless link, for a small fraction of a penny apiece. A prototype on display at ARM headquarters in Cambridge looks like a stiff piece of tape, pattered with printed circuit traces.
The processor uses plastic circuitry printed with flexible organic semiconductors. It has only about a thousand logic gates in all. There are eight sensors on the chip; each one isn't very useful in itself, the processor has to sum them and sort out the results. It's not easy to make such a feeble processor smart enough to do that. It uses a straightforward statistical analysis scheme -- which has to be tweaked for different applications. A PA chip for monitoring strawberries won't work for chicken; a different pattern has to be printed to change the programming.
Pragmatic's engineers are implementing software tools to permit easy construction of a chip for a particular application. The idea is to define a new algorithm in Python, a widely-used programming language, and then translate that algorithm into a circuit diagram that can be fed to a machine that produces the chips. America's Defense Advanced Research Projects Agency (DARPA) is paying close attention, having similar ideas with the DARPA Electronics Research Initiative.
The PA prototypes are battery-powered, but production devices may not need a battery. The PA chip has an antenna to allow it to communicate with a smartphone or other device; the antenna can also provide power to the PA chip. Contactless cards work in much the same way. Applications that need continuous operation may be able to scavenge radio energy from the ambient environment. This is a challenge, because the density of that ambient energy is very low. It's not so hard to harvest energy for processing, but it takes more to obtain energy for wireless communications.
Self-powering chips would be particularly useful in applications where it's troublesome to replace batteries -- for example, devices that monitor structures like bridges or tunnels. The sheer number of IoT devices also makes battery power problematic: if there were a trillion chips, each powered by a lithium button cell, that would demand three years' production of lithium. As an ARM official points out, with dry British understatement, a trillion is "quite a big number, when you think about it". [TO BE CONTINUED]
START | PREV | NEXT | COMMENT ON ARTICLE* AMERICA'S CONSTITUTION (102): In the fall of 1962, the Cold War took a frightening turn. In 1958, as an interim measure, Eisenhower had arranged for the siting of Jupiter intermediate-range ballistic missiles (IRBM) in Italy and Turkey. They were meant as a stopgap until longer-range strategic missiles were available -- but the missiles in Turkey made the Kremlin very nervous, since they could hit many of the Soviet Union's major cities with only a few minutes' warning.
Premier Khrushchev, having decided that JFK was weak, decided to sneak intermediate-range missiles into Cuba. That would balance the threat of the Jupiters, and also in principle help protect Castro's revolution. However, the US was maintaining surveillance of maritime traffic in and out of Cuba, and the missile shipments were soon observed. The US military brass was inclined to take action against Cuba, but Kennedy didn't want war. Instead, on 22 October 1962, he announced to the nation that he had declared a "quarantine" of Cuba -- "blockade" would have seemed too warlike -- and had demanded withdrawal of the missiles.
Khrushchev didn't back down right away, but on 27 October a U-2 spyplane was shot down over Cuba, the pilot being killed. Pentagon brass wanted to bomb the missile sites, but Kennedy had already received a message from Khrushchev, saying the Soviets would remove the missiles, if the US promised not to invade Cuba and withdrew the Jupiter missiles from Turkey. Kennedy went public to announce that the US would not invade Cuba, but only privately agreed to remove the Jupiters from Turkey. They would have been withdrawn anyway; they were obsolete once ICBMs and SLBMs came on line.
In any case, Khrushchev removed the missiles from Cuba in haste and humiliation. Relations between the two super-powers then settled down the time being; by that time, both nations were flying spy satellites over each other's territory, the USA with its CORONA satellites, the USSR with its Zenit satellites. The satellite imagery helped reassure each side that the other was not preparing to attack.
Calmer relations helped pave the way towards the international "Partial Test Ban Treaty (PTBT)" of 1963, which grew out of concerns over radioactive fallout released by nuclear tests. The fallout problem had come to light during the Eisenhower Administration, with Eisenhower pushing for a nuclear testing ban; Kennedy picked up where that effort left off, with the goal of a comprehensive test ban. That couldn't happen at the time, with the full name of the accord -- "Treaty Banning Nuclear Weapon Tests in the Atmosphere, in Outer Space and Under Water" -- neatly defining its scope. The UK, USA, and USSR were the original signatories.
The treaty restricted nuclear tests to underground detonations to limit fallout, making it effectively an environmental measure. It also banned nuclear tests in space. There had been a number of nuclear tests in near space from 1958 to 1962, which had unintended consequences, most prominently spreading temporary radiation belts around the Earth that indiscriminately fried satellites. With both superpowers becoming increasingly dependent on Earth satellites, and no compelling reason to continue the space detonations, they agreed to stop setting off nukes in orbit.
* Domestically, JFK pursued a liberal agenda, under the label of the "New Frontier". He proposed more Federal funding for education, medical care for the elderly, economic assistance to depressed rural areas, and an end to racial discrimination. Although JFK made high-profile appointments of blacks in his administration and promoted affirmative-action policies in business, he had no prospect of accomplishing much through legislation, and no significant civil-rights legislation was passed on his watch. It was a time of high tension over desegregation, with the Kennedy Administration left behind the learning curve -- and in the end, he accomplished not much more than to get both sides annoyed at his administration.
Indeed, although Kennedy has been stereotyped as an ultra-liberal in memory, that was hardly the truth. He was generally pragmatic, starting out with a pledge to balance the Federal budget, and working up towards efforts late in his administration to cut tax rates. His administration did run up a moderate budget deficit, but the economy boomed while inflation remained tame. The administration took on the steel industry for price raises, the president bullying them, going so far as to have steel executives investigated by the FBI. [TO BE CONTINUED]
START | PREV | NEXT | COMMENT ON ARTICLE* GIMMICKS & GADGETS: As discussed by an article from ECONOMIST.com ("Speaker See, Speaker Do", 9 May 2019), "smart speakers" -- like Amazon Echo, Google Home and Apple HomePod -- are taking over the world, with users asking them for the weather, or to buy something online, or to play music. However, smart speakers are limited in that they are controlled by voice inputs. It might be useful for them to be aware of other actions by their users.
Chris Harrison and Gierad Laput of Carnegie Mellon University, in Pittsburgh PA, are working to extend the perception of smart speakers -- the two researchers calling their concept "SurfaceSight". They start by fitting an Amazon Echo smart speaker with a lidar -- light radar -- that scans the area around the device, with the Echo smart enough to recognize household objects and user hand gestures. At present, the lidar only scans close to the speaker's base, that restriction being established to assure user privacy.
The lidar-equipped speaker is smart enough to be taught to recognize a wide range of objects -- saucepans, cereal boxes, screwdrivers, bunches of carrots, smartphones. Recognitions can be used to cue actions. One experimental app recognizes utensils and ingredients laid out on a preparation surface, to direct and monitor cooking a particular recipe. Another app recognizes the user's smartphone, and links via bluetooth to the music collection on the phone.
Gesture recognition would also be useful; for example, a user might skip music tracks by swiping a finger over the area the lidar is scanning. It could be similarly used to control a presentation, while also keeping track of what the audience is doing. Of course, other appliances could make similar use of the technology.
Harrison and Laput are not alone in making surfaces active. Swan Solutions of Houston, Texas, sells "Knocki" -- an accelerometer that can be fixed to a surface to detect a user rapping on the surface, with the sequence of raps being used to specify different functions to be executed. A British firm named Audio Analytics is even figuring out how to sense different sounds, for example the hiss of opening a can of Coke or popping a cork from a bottle. One wonders how far this concept can be taken, however; there's a fine line between being attentive to a user and being a nuisance.
* As discussed by an article from SCIENCEMAG.org ("This New Device Generates Light From The Darkness Of Space" by Robert F. Service, 12 September 2019), solar panels are great, but they have a major limitation in that they don't generate power at night.
Researchers have now developed a roof-mounted thermoelectric power generation system that works at night. Thermoelectrics are nothing new; in the past, they were primarily seen as a means of scavenging electricity from, say, a hot engine exhaust. In the rooftop system, however, a panel based on a ceramic material is heated on the bottom by warm air from the rooftop, with the top of the panel exposed to the cooler darkness. That creates a 2 degrees Celsius (3.6 degree Fahrenheit) temperature difference, which generates electricity.
It's not very efficient, providing only 25 milliwatts of power per square meter, about enough to drive a light-emitting diode. The researchers believe that the scheme might be useful in powering carbon-free lights and sensors in remote areas. It's an interesting idea in principle, but there is the question of how cost-effective it would be.
* As discussed by a video from REUTERS.com, Ford and McDonald's have collaborated on a project to make car parts -- for example, a headlight housing -- from "coffee chaff", the skin that comes off from roasting the beans. The part is still plastic, the chaff being used as filler, replacing talc. The end result is as sturdy, lighter in weight, and heat-resistant.
In related news, CNN.com reports that South Africa startup company Munch Bowls has developed a disposable bowl made of wheat, instead of troublesome plastic. It can hold hot soup for five hours, and has a shelf life of 15 months. When done with the bowl, it can be eaten. Munch Bowls is selling their product internationally.
In still somewhat related news, researchers have come up with a scheme for making artificial rhino horns out of horsehair, glued together with a matrix of regenerated silk. The rationale is to end the trade in rhino horns, which are supposed to have remarkable medicinal properties. They don't, so the fakes work as well as the real thing. It's an elegant idea: just tell the world about the fake horns, scammers will start producing them in quantity, and nobody will know the difference. The end result will be either to normalize the fakes, or to counterfeit the market to death.
COMMENT ON ARTICLE* FRAGILE AI (2): Digital neural networks are powerful, since they can pick out patterns from many different features of an input when attempting to classify it. An AI trained to recognize aircraft might find that features such as patches of color, texture, or background are just as strong predictors as the things that we would consider salient, such as wings. However, that also means that a very small change in the input can tip it over into what the AI considers an apparently different state.
One solution is to throw just more data at the AI; in particular, to repeatedly expose the AI to cases that give it trouble, and correct its errors. In this form of "adversarial training", as one network learns to identify objects, a second tries to change the first network's inputs so that it makes mistakes. In this way, adversarial examples become part of a DNN's training data.
Hendrycks and his colleagues have suggested quantifying a DNN's robustness against making errors by testing how it performs against a large range of adversarial examples. However, they also point out that training a network to resist one kind of attack might weaken it against others.
Another approach to improving the performance of DNNs, being worked on by a team of researchers under Pushmeet Kohli at Google DeepMind in London, is to inoculate DNNs against making mistakes. Many adversarial attacks work by making tiny tweaks to the component parts of an input, such as subtly altering the color of pixels in an image, until a DNN stumbles into a misclassification. Kohli's team is working on making DNNs less sensitive to small changes in inputs.
However, at present, DNNs remain brittle. The basic problem, according to Yoshua Bengio, is that DNNs don't have any means of knowing what features are important in recognizing a particular object and which are not. Again, humans can break the image of a lion down into its components; DNNs can't. Bengio says: "We know from prior experience which features are the salient ones. And that comes from a deep understanding of the structure of the world."
Obtaining that "deep understanding" is a very difficult problem. One approach to do so is to combine DNNs with "symbolic AI", which was the dominant paradigm in AI before machine learning. In symbolic AI, instead of soaking up a vast store of patterns, machines actually reasoned using hard-coded rules about how the world worked, such as that it contains discrete objects that are related to each other in various ways.
Symbolic AI is very laborious, with classic experiments including attempts to instruct, for example, a machine to plug different-sized blocks into different-sized holes. Machine learning is far easier to deal with, but it comes along with fundamental limitations. Psychologist Gary Marcus of New York University believes hybrid AI models are the way to get out of this blind alley: "Deep learning is so useful in the short term that people have lost sight of the long term." Marcus, a persistent critic of deep-learning technology, has founded a startup named "Robust AI" in Palo Alto, California, which aims to hybridize deep learning with rule-based AI techniques to develop robots that can operate safely alongside people.
There is also a challenge in making sure a DNN is trained with plenty of realistic data. For example, most computer-vision systems fail to recognize that a can of beer is cylindrical, since they were trained on data sets of 2D images. That's why Nguyen and colleagues found it so easy to fool DNNs by presenting familiar objects from different perspectives. Learning in a 3D environment, real or simulated, can help.
Simply feeding data to a DNN is also limited, Bengio saying: "Learning about causality needs to be done by agents that do things in the world, that can experiment and explore." Another deep-learning pioneer, Juergen Schmidhuber at the Dalle Molle Institute for Artificial Intelligence Research in Manno, Switzerland, similarly believes that a next, bigger wave of machine learning is on the way, through "machines that manipulate the world, and create their own data through their own actions."
AIs that use trial and error to learn how to master games -- playing games against themselves, and gradually working their way up -- are a step in that direction. However, games are, as a rule, well-defined systems; interaction with the open-ended real world is a much bigger challenge. [TO BE CONTINUED]
START | PREV | NEXT | COMMENT ON ARTICLE* FLY CONNECTOME: As discussed by an article from ECONOMIST.com ("The Biggest, Most Detailed Map Yet Made Of Brain Cells", 23 January 2020), the first person to probe into the details of the nervous system was a Spanish scientist named Santiago Ramon y Cajal. At the beginning of the 20th century, he probed into the nervous system, documenting his work with beautiful drawings of the treelike cells of the brain and spinal cord. For his efforts, he was awarded the Nobel Prize in 1906.
In the 21st century, modern researchers carry on Cajal's work, leveraging off advanced microscopy, robotics, and machine learning. The focus is now to create "connectomes": Three-dimensional maps of all the neurons in entire brains, and how those neurons link together. Now a research team has assembled a map of about a quarter of a fruit fly's cerebral capacity.
That map -- of the fly's "hemibrain", a set of about around 25,000 neurons in the center of the organ -- took more than a decade to put together. The work to obtain the fly's connectome was pushed by Gerry Rubin, a biologist who runs the Janelia Research Campus in Virginia, a part of the Howard Hughes Medical Institute that is dedicated to neuroscience. He had previously worked on mapping the fruit fly's genome as a proof of principle for the Human Genome Project.
The hemibrain connectome effort was only the first phase of the campus's "FlyEM" project, which seeks to map the fruit fly's entire brain, which contains around 100,000 neurons. That is trivial in comparison to the 85 billion neurons in the human brain, or even the 70 million neurons in a mouse brain. However, as with the fly's role in the Human Genome Project, FlyEM will be a proof of principle.
Neuron are long, slender cells, with their body or "axon" branching into clusters of "dendrites" at the end, which link to other neurons through junctions called "synapses". There are an average of about 200 synaptic connections per neuron, or about 200 million synapses in all. It is the organization and interconnection of the fly's brain that supports its behaviors. The only full connectome mapped out so far is that of the nematode worm C. elegans -- which has either 302 or 385 neurons in its nervous system, depending on whether it is hermaphrodite or male, there being no females as such. The worm has about 7,000 synapses.
Tracking down the connectome of C. elegans involved laborious procedures that were not different from those of Cajal. The researchers sliced their worms into thin sections using diamond knives; stained the slices to reveal the cells in them up more clearly; and then obtained electron-microscope images of the result. The images were inspected by eyeball to obtain the result.
For any larger connectomes, such an approach is impractical, so Rubin and his research team sought means of automating the process. For example, one trick they devised was to "sandblast" a sample of brain tissue with a beam of gallium ions, stripping away a few nanometers of the tissue. An electron microscope obtained an image of the sample; with the sandblasting and imaging persisting until the sample was fully analyzed.
The electron microscopes were built especially for FlyEM. They sit on air-filled pads to minimize vibrations that might blur the images, and the room containing them rests on its own concrete slab, to isolate it from the remainder of the laboratory. In addition, although electron microscopes are typically designed to run for a few hours at a time at most, these instruments were made to run continuously for months.
The specialized electron microscopes used by the FlyEM project to probe the hemibrain of the fly generated millions of high-resolution images, with computing power used to stitch them together into a 3D map. That left the big problem of identifying and labeling all the connections in the hemibrain. That was obviously impossible to do by hand, so Stephen Plaza -- the FlyEM program manager -- turned to Google for help.
Computer vision systems are now routinely used to scan through streams of CCTV or satellite images to spot and report items of interest; artificial intelligence (AI) has greatly improved their performance over the last few years. At the request of the Janelia Research Center, Google trained one of its AI algorithms to recognize neurons and synapses within the FlyEM images, and also trace axons and dendrites.
To get started, the research team trained the AI on images that had already been marked up by human experts. As the crunched through more images, human proofreaders checked its output and informed it of errors. In time, the manual workload declined and the speed with which images were correctly annotated shot up. It would have taken centuries for Plaza and his team of 50 proofreaders to deal with the annotation; with the AI's help, they cut it down to a few years.
The FlyEM data was released to the public. Anyone with an internet connection can look up lists of neurons that are connected to each other and see 3D diagrams of the neural map. At Janelia, other research groups are already mining that data to glean insights. Vivek Jayaraman's team, for example, has been investigating how a fruit fly's brain helps the insect first to understand its orientation in space, and then use that information to help it navigate. Previously, Jayaraman worked with theoretical models; now the hemibrain map shows him exactly what's going on.
Having completed the hemibrain, FlyEM's research team plan to finish the rest of the fruit fly connectome in about two years. That is an end in itself, but the project is also about building tools to tackle bigger connectome projects: the team plans to move on to mouse brains next. Rubin estimates that assembling a mouse connectome would cost around $500 million USD -- more than ten times the ultimate total cost of FlyEM. After that, there's the challenge of the human brain. That seems impossible, but Rubin says that, early in the Human Genome Project, that seemed impossible as well. He says: "And now, we have projects where we're going to do 10,000 human-genome sequences."
COMMENT ON ARTICLE* INTERNET OF THINGS (4): Siemens has big ambitions in its smart building efforts, company officials envisioning the firm's systems in thousands of such structures. There are possibilities beyond monitoring; when personnel are monitored and inventory is kept up-to-date by beacons on all equipment, a virtual "digital twin", a computer simulation of the structure, can tell personnel where everything is, and how to get it. That would be literally life-saving in a hospital.
Of course, along with tracking personnel, customers could be tracked as well. Smart cameras are now available that can track customers in a store's aisles a and derive some information about them -- for example, their gender. Smart camera systems can also be used to measure waiting times at airports; a Florida mega-church uses the system to monitor attendance. Shoppers' smartphones can also be monitored to obtain customer data. Industry analysts project that global demand for such "in-store analytics" is growing by 23% a year and will be worth $3.2 billion USD by 2023.
Customers can be tracked outside of buildings as well. Many insurers have offered discounts to drivers willing to install a black box that collects data from their car on acceleration, cornering, braking and the like, and relays it back for analysis. The black box is increasingly unnecessary, since modern cars are heavily wired with sensors, and have wireless links to report data. Insurers can used the data to offer lower insurance rates to cautious drivers, and even provide them with helpful hints. Another option is to dynamically vary insurance rates, depending on how much a customer drives.
Going further, customers of Ping An -- a Chinese insurer, the world's biggest -- can register accounts with facial-recognition tech; the tech also can obtain a customer's body-fat percentage, which factors into life-insurance premiums. Similarly, John Hancock Financial of the US plans to link health insurance to data from customer smartphones, or activity trackers like Fitbits, that can determine how much a customer exercises.
Obviously, there are limits to how much customers will tolerate such intrusiveness. Morgan Stanley has done surveys asking people if they would be willing to share their data, in exchange for a price cut. Asian respondents were generally willing to do so, while Westerners were not so willing. The Germans proved to be the least willing of all.
However, businesses such as life insurance have a very strong motive to collect data on their customers; those who can identify high-risk customers will have a competitive advantage against those who don't. In the absence of legal constraints on data collection, such companies will be under pressure to collect as much data as they can.
* Farms are becoming increasingly digitized as well. Keenan Systems of Ireland makes computerized feed-mixing wagons for dairy cows. The computer is programmed with the nutritional requirements for the herd, with sensors on the wagon determining exactly what a farmer puts into the feed mix. The data are transmitted over the mobile-phone network to nutritionists tasked with monitoring the feed; if the feed mix isn't right, the nutritionists send a text message to the farmer with corrective recommendations.
Cainthus, another Irish company, is one of a number of firms working on computer vision to boost farmyard productivity. Their system uses cameras to track cows in barns and fields, with machine learning analyzing the images. David Hunt, the firm's boss, says the tech is smart enough to track individual animals, and to determine if a cow isn't feeding when it should be, or acting in a way that suggests it's sick. For now, he says, the company is working mainly on Friesian and Holstein cows, whose distinctive markings "mean they're basically walking QR codes", though he plans to extend the system to other breeds in time.
Another approach is to put the sensors inside the cows themselves. An Austrian firm named smaXtec has developed a sensor that can be swallowed. It lodges inside the reticulum, one of a cow's four stomachs, and stays there for the rest of the animal's life -- monitoring body temperature, movement and stomach acidity, and uploading the results when the cow is near a wireless detector.
According to Stefan Rosenkranz, smaXtec's co-founder, the system can detect when animals are in heat, and spot the early signs of calving up to 15 hours before it happens. It can spot diseases several days before they become obvious to human observers, allowing early, more effective, and cheaper treatment. Rosenkranz says that more sophisticated sensors are in the works, and that sales are doubling. With 278 million dairy cows in the world, there are plenty of customers. [TO BE CONTINUED]
START | PREV | NEXT | COMMENT ON ARTICLE* AMERICA'S CONSTITUTION (101): The election of 1960 pitted Richard Nixon, Eisenhower's vice president, against Senator John F. Kennedy (JFK) of Massachusetts, from one of the wealthiest families in the United States. The election was surprisingly close, given that JFK entirely telegenic -- indeed, the first presidential candidate who could be described as such -- while Nixon was not.
JFK's primary focus was on the Cold War. In June 1961, he met with Premier Khrushchev in Vienna, who trod all over him; Kennedy, instead of walking out, attempted to push on, only to convince Khrushchev that he was weak. To the extent there was a focus in the discussion, it was on East Berlin. The presence of divided Berlin in East Germany had always been an irritation to the Soviets, and was becoming much more of one as East Germans defected in large numbers to West Berlin.
Shortly after JFK came back home, the Soviets announced that they would sign a treaty with East Berlin, which would abrogate all third-party occupation rights in either sector of the city. That was unacceptable to Kennedy; he prepared for war, announcing a major defense buildup in July, and stating that the US would regard an attack on West Berlin as an attack on the USA. The announcement met with widespread public approval. Khrushchev chose to erect a wall dividing East and West Berlin, with Kennedy saying: "Better a wall than a war."
The Cold War was presenting challenges closer to home. In 1959, a Cuban revolutionary movement led by Fidel Castro had overthrown the government of dictator Fulgencio Batista, with Castro installing a Communist government. The CIA came up with a scheme in which anti-Castro Cuban expatriates would invade Cuba, and overthrow Castro in turn. Kennedy inherited the plot from the Eisenhower Administration and let it go ahead, though doubts had accumulated as to the wisdom of the scheme.
A brigade of 1,500 Cuban expatriates was landed at the Bay of Pigs on 17 April 1961. The underlying assumption of the CIA, it appears, was that US military power would be brought to bear if the invaders ran into trouble. JFK didn't go along with it, and the exiles were crushed. It was a major humiliation for Kennedy.
In consequence of the Bay of Pigs fiasco, the Kennedy Administration established a covert effort, codenamed MONGOOSE, to subvert the Castro regime, with its scope including possible assassination of Castro. It was under the general direction of Attorney General Robert F. Kennedy, the president's brother, making it one of the more unusual activities ever found in the portfolio of an attorney general. MONGOOSE was a complete failure -- it even involved talks with mobsters, who cheerfully strung the government along, and did nothing -- and an embarrassment when it came to light years later.
On 12 April 1961, only days before the Bay of Pigs invasion, the Soviet Union put a man, Yuri Gagarin, into orbit around the Earth. His space capsule, named "Vostok", was actually a variant of a spy satellite, named "Zenit", that would take photos and parachute-land with its payload. In any case, JFK felt challenged on a number of fronts at the time; on 25 July 1961, in a major speech to both houses of Congress, the president outlined a program for confronting the Communist threat, detailing a range of efforts and asking for more funding. The administration had already put forward a scheme by which American civilian volunteers would go to undeveloped countries to provide assistance, with the "Peace Corps" formally established in September.
The speech also proposed that the USA put a man on the Moon within a decade -- the Cold War put the Space Race on an accelerated footing. The result was the Apollo program. It wasn't to be done as a scientific exercise, but as a political demonstration: a way of demonstrating the power of American democracy to the Soviets, without pointing a gun at them.
More quietly, the Kennedy Administration pushed on in support of anti-Communist governments in Laos and South Vietnam, pumping in military aid and advisors. Laos was a backwater; it wasn't as important as South Vietnam, and wasn't as visible to the world. South Vietnamese strongman Ngo Dinh Diem was problematic, in that his government was heavy-handed, inept, corrupt, and lacking in popular support. The fight against Viet Cong -- Vietnamese Communist -- insurgents wasn't going well, but the Kennedy Administration wasn't sure what to do about it. [TO BE CONTINUED]
START | PREV | NEXT | COMMENT ON ARTICLE* SCIENCE NOTES: The Peninsula Open Space Trust (POST) -- a San Francisco-area wildlife conservation group -- went viral with a surprising "true-life nature adventure" video. POST has been using cameras to monitor animal traffic through culverts and under highway bridges to determine how much such "safe passages" help wildlife.
In the viral video, a coyote appears at the mouth of a culvert. It doesn't enter, instead prancing about a bit, as if telling someone else: "Well come on, hurry up." Then a badger enters the scene, and the two go through the culvert. POST staffers were very surprised by this team-up, but according to a 2016 article from the US Fish & Wildlife Service, it's not news:
BEGIN QUOTE;
Coyotes and badgers are known to hunt together and can even be more successful hunting prairie dogs and ground-squirrels when they work in tandem. Studies have shown that this unusual relationship is beneficial for both species. The coyote can chase down prey if it runs and the badger can dig after prey if it heads underground into its burrow systems. Each partner in this unlikely duo brings a skill the other one lacks. Together they are both faster and better diggers than the burrowing rodents they hunt.
END QUOTE
Interspecies symbiosis is nothing unusual in the animal kingdom. One wonders how far the cooperation goes: do they share kills? It is tempting to believe that the coyote, a notoriously clever animal, is the brains of the operation.
* As discussed by an article from SCIENCEMAG.org ("The Genes That Make Squid Eyes Also Make Your Legs" by Elizabeth Pennisi, 15 July 2019), one of the fundamental components of the genomes of all organisms are "developmental genes", which provide high-level direction for the construction of organisms. They don't do anything in themselves, they just tell other genes what to do: "Make an eye." "Make a foot." -- and so on. The same developmental genes can be common across organisms, but they may perform different functions. Consider, for example, the cephalopods -- the squid, octopus, and their siblings. Cephalopods are known to have "camera-type" eyes, much like our own, even though they are evolutionarily very distant from us.
Evolutionary developmental biologists have now shown just how distant the connection is by finding that the developmental genes that direct the initial formation of legs in us and other vertebrates also direct the formation of the lens of a squid's eye. The squid lens forms as extra-long membranes associated with specialized eye cells overlap to form a tight ball. Our lenses are actually degraded cells themselves, packed with a clear protein. To learn how the squid lenses form, these researchers carefully tracked where, when, and which genes turn on and off as embryos of Doryteuthis pealeii -- a squid commonly served as a fried appetizer -- develop.
The researchers were startled to find that a gene regulatory network previously known for generating limbs was at work in generating a squid eye. The researchers are now working to figure out exactly what each of these genes is doing to make the lens form. The squid eye is a product of parallel evolution, disconnected from the evolution of the human -- except at its deep genetic roots.
* As discussed by an article from SCIENCEMAG.org ("How A Volcanic Eruption Helped Create Modern Scotland by Sid Perkins, 5 December 2019), during seven disastrous years in the 1690s, Scotland suffered from crop failures and famine, with up to 15% of the population dying as a result. The consequence of the "Scottish ills" was the formation of the United Kingdom with England.
Why the troubles? There have been suspicions that they were linked to a volcanic eruption someplace that threw aerosols into the upper atmosphere, resulting in cooling that can last up to several years, helping to trigger droughts and crop failures. Paleoclimatologists like to hunt for clues of such events in tree rings, which track variations in climate -- the poorer the growing conditions, the less tree growth and the narrower the ring -- but until recently, tree-ring data from northern Scotland, which was the worst-hit, was sketchy.
In 2017, however, researchers assembled a complete record of climate in that area from 1200 to 2010. They used data from still-living trees and logs that had fallen into lakes, where they were preserved for centuries. The research team -- led by Rosanne D'Arrigo, a paleoclimatologist at Columbia University's Lamont Doherty Earth Observatory in Palisades, New York -- second-coldest decade of the past 800 years ran from 1695 to 1704. Summertime temperatures during this period were about 1.56 degrees Celsius lower than summertime averages from 1961 to 1990. The troublesome cool climate links to two major volcanic eruptions in the tropics: one in 1693 and an even bigger one in 1695.
The difficult climate wasn't Scotland's only problem. Others were backwards agricultural practices; a government push to encourage grain exports, which left low reserves when crops failed; and, most famously, a disastrous attempt to set up a colony in Panama from 1698 reduced Scotland to ruin. The result is not disputed: in 1707, the Scots Parliament decided to end Scottish independence, joining England to form the United Kingdom.
COMMENT ON ARTICLE* FRAGILE AI (1): As discussed by an article from NATURE.com ("Why Deep-Learning AIs Are So Easy To Fool" by Douglas Heaven, 9 October 2019), artificial intelligence (AI) has been growing in power by leaps and bounds, in particular through "deep neural networks (DNN)", which can be "trained" through examples to recognize patterns, and direct actions accordingly. Such "machine learning" systems are not just lab toys any more, either; they run everything from automated telephone systems, to user recommendations on the streaming service Netflix.
There's a huge problem with DNNs, however -- in that they work most of the time, but fall down badly on corner cases. Researchers have, for example, demonstrated that an AI system could misinterpret a STOP sign for a 45 MPH sign, just by adding a set of stickers to the sign. Similarly, researchers have frustrated face-recognition systems by sticking a printed pattern onto a hat or glasses; and have tricked speech-recognition systems into picking up phantom phrases by inserting patterns of white noise in the audio. Such unpredictability makes the use of DNNs to, say, drive cars or diagnose medical symptoms problematic. An AI system used to protect computer networks against hackers might well be subverted by hackers who know how to exploit its weaknesses, and so get the AI system working for them.
Dan Hendrycks -- a doctorate student in computer science at the University of California, Berkeley -- says the unpredictable behavior of DNNs is not a bug that has a straightforward fix. Any small changes in inputs outside of their training sets, even some that a human would be hard-pressed to notice, can stymie them. Hendrycks sees DNNs as "fundamentally brittle": brilliant at what they do until, taken into unfamiliar territory, they break in unpredictable ways. Francois Chollet, an AI engineer at Google in Mountain View, California, says: "There are no fixes for the fundamental brittleness of deep neural networks."
Researchers tend to believe the problem with DNNs is that they are too narrow: experts at one particular task, with no capability in any others. The belief is that an AI with broader capability would have the "good sense" not to make ridiculous mistakes.
DNNs are only loosely based on the architecture of the brain. They are software systems made up of large numbers of digital neurons, stacked in many layers. Each neuron is connected to others in layers above and below it. Raw input coming into the bottom layers, for example pixels in an image, trigger some of those neurons, which then pass on a signal to neurons in the layer above according to simple mathematical rules. Training a DNN network involves exposing it to a massive collection of examples, each time tweaking the way in which the neurons are connected so that, eventually, the top layer gives the desired answer -- such as always interpreting a picture of a lion as a lion, even if the DNN hasn't seen that particular picture before.
In 2013, following the initial rush of enthusiasm for deep learning, Google researcher Christian Szegedy and his colleagues released a paper showed that a DNN trained to recognize a lion could fail completely if given an image of a lion with only a few pixels changed -- recognizing it as a library, for example. In 2014, Jeff Clune and Anh Nguyen -- then computer scientists at the University of Wyoming -- and Jason Yosinski at Cornell University in New York, showed that DNNs could be tricked into seeing things that aren't there, such as spotting a penguin in a set of wavy lines.
The problem is that a DNN sees a particular pattern as, say, a lion, but doesn't really recognize the parts of a lion. Human recognition of a lion is hierarchical, involving recognition of the general appearance of a lion, as well as its subsidiary components: fur, mane, eyes, nose, teeth, whiskers, legs, paws, claws, tail. If any of those subsidiary elements are ridiculously wrong, a human spots it immediately -- and also easily recognized extraneous items on a lion, for example a golden collar, and categorizes them as such. A DNN, while factoring in all available details, achieves no such hierarchy of understanding.
Yoshua Bengio -- of the University of Montreal in Canada, a pioneer of deep learning -- says: "Anybody who has played with machine learning knows these systems make stupid mistakes once in a while. What was a surprise was the type of mistake. That was pretty striking. It's a type of mistake we would not have imagined would happen."
It hasn't been getting any better, either. In 2018 Nguyen, now at Auburn University in Alabama, showed that simply rotating objects in an image was sufficient to stymie off some of the best image classifiers around6. In 2019, Hendrycks and his colleagues reported that even perfectly natural images can sometimes trick image classifiers into, say, identifying a mushroom as a pretzel or a dragonfly as a manhole cover.
It's not just images, either; all DNN recognition systems can be fooled. Speech can be misunderstood, while AIs that play games can be sabotaged. In 2017, computer scientist Sandy Huang -- a doctoral student at the University of California, Berkeley -- and her colleagues investigated DNNs that had been trained to beat Atari video games through a scheme called "reinforcement learning". In this approach, an AI is given a goal and, in response to a range of inputs, learns through trial and error what to do to get to that goal. It's been used to implement superhuman game-playing systems, such as AlphaZero and the poker bot Pluribus. However, Huang's team was able to make their AIs lose games by adding one or two random pixels to the screen.
In 2019, AI doctoral student Adam Gleave at Berkeley and his colleagues showed that it is possible to introduce an agent to an AI's environment that operates on an "adversarial policy' designed to confuse an AI. For example, an AI footballer trained to kick a ball past an AI goalkeeper in a simulated environment can't score when the goalkeeper starts to behave in unexpected ways, such as falling to the ground.
Such weaknesses in a DNN raise the prospect of it being hacked. In 2018, a Google team showed that it was possible to use adversarial examples not only to force a DNN to make specific mistakes, but also to reprogram it entirely -- in effect, repurposing an AI trained on one task to do another. There's nothing much in a DNN that's application-specific, except for its training. Changing the training can change the application. Clune says: "In theory, you can turn a chatbot into whatever program you want. This is where the mind starts to boggle."
Clune imagines a situation in the near future in which hackers could hijack neural nets in the cloud to run their own spambot-dodging algorithms. Computer scientist Dawn Song at the University of California, Berkeley, sees DNNs as sitting ducks: "There are so many different ways that you can attack a system. And defense is very, very difficult." [TO BE CONTINUED]
NEXT | COMMENT ON ARTICLE* LIFE AT THE EDGES: As discussed by an article from NATURE.com ("These Microbial Communities Have Learned To Live At Earth's Most Extreme Reaches" by Monique Brouillette, 12 March 2020), the ocean deeps are a something of a desert, where life has little to sustain itself. However, it clings on there nonetheless. An analysis of rock samples from Atlantis Bank, part of a seafloor mountain in the Indian Ocean where deep crustal rock is exposed close to the surface, has turned up microbes adapted to life within nutrient-poor, hairline fractures in the Earth.
A team under marine microbiologist Virginia Edgcomb -- of Woods Hole Oceanographic Institution in Massachusetts -- found species of bacteria, fungi and archaea that live in the rocks, where they feed on carbon from fragments of amino acids and other organic molecules carried in deep ocean currents. Such microorganisms were once considered unusual forms of life, but research over the past few decades indicates that microbes are common in extreme environments, all over the Earth: in deep sediments under the oceans, the frigid deserts of Antarctica and even the stratosphere.
Such harsh living conditions have led to the evolution of diverse means of staying alive. Some of these microbes can breathe metals, even radioactive ones such as uranium; some capture nutrients from trace gases in the air; others, like those found buried deep in the sludge of the ocean floor, metabolize so incredibly slowly that they might survive for hundreds or thousands of years old, eating and reproducing infrequently.
The first hints that life existed deep within Earth's crust emerged surfaced in the 1920s, when oil prospectors noticed that groundwater around their oil fields was laced with hydrogen sulfide and bicarbonate, which are both made by bacteria. In the 1980s, microbiologists began counting the microbes in cores brought back from the Deep Sea Drilling Project -- a major effort to explore the sea floor -- and were startled by the quantities.
In the early 2000s the JOIDES Resolution -- a drill ship equipped with a floating lab -- departed San Diego, California, on a mission to explore life in the deep biosphere. The vessel carried a team of scientists led by Bo Jorgensen, a geomicrobiologist from Aarhus University in Denmark, and Steven D'Hondt, an oceanographer from the University of Rhode Island. It went to the eastern Pacific Ocean, where the team sampled rock cores as deep as 5,300 meters off the coast of Peru, capturing sediment up to 35 million years old.
The team confirmed that the sediments contained plenty of microbes, able to survive on about 1% of the carbon that a normal surface-living microorganism has access to. Early experiments on the JOIDES showed they had glacially slow metabolisms -- leading some researchers to wonder if they were really alive, or were just slowly starving to death. Jorgensen disagrees, pointing to later research that shows the microbes have active protein and DNA repair mechanisms. He says: "We always expect that bacteria are growing fast, this is what you see in the laboratory, but I found that most of them are growing extremely slowly. What we used to think was extreme is the normal."
Most of these fringe dwellers can't be studied in a lab, since they typically don't grow in cultures. Even the ones that do, act differently under artificial conditions from how they would in the wild. That's made them difficult to study -- but now that's changing, thanks to the introduction of metagenomic techniques that allow scientists to track gene expression in whole communities simultaneously. Researchers have now identified genes that are involved in low-energy protein and DNA repair processes, as well as energy-efficient metabolic strategies. They have even found genes that allow bacteria to survive off trace gases, such as carbon monoxide and hydrogen.
Rick Colwell -- a microbial ecologist at Oregon State University in Corvallis [ED: my alma mater] -- says that findings by Edgcomb's team "extend what we are learning about how microbes live in fractured rocks that make up much of Earth's subsurface. We are gaining more evidence that the things they subsist on, like hydrogen as a source of energy, create a different tempo for life."
* As discussed by an article from SCIENCEMAG.org ("Some Spots On Earth Are Too Hostile For Life" by Elizabeth Pennisi, 29 October 2019), although microorganisms are often found in extreme environments, there are still a few places where life can't get a hold at all. Hot briny lakes in Africa's Rift Valley and the cold, dry soil of Antarctica's Shackleton Glacier Valley are entirely devoid of life, too hostile for the toughest organisms. In the Rift Valley lakes, in Ethiopia, volcanic gases venting from below acidify the water, which is also rich in salts from brines created by the evaporation of earlier bodies of water. Add the heating effect of the volcanic activity, and the lakes represent an environment more extreme than any found in Yellowstone National Park or even in the Atacama Desert.
Purificacion Lopez-Garc?a -- a biologist at with the French national research agency CNRS and the University of Paris-Sud -- and her team used microscopy, cell sorting, genomics, and other techniques to look for signs of life in the water. In some ponds, microbes known as archaea thrived; but some in isolated canyons and lakes, where salt concentrations top 50% and acidity was high, they could find no trace of life.
They did notice balls of minerals that others have found elsewhere and interpreted as "microfossils" -- but the researchers doubt they have biological origins. According to Lopez-Garcia, the high salinity, plus the acidity and temperature, are a deadly combination, particularly because the salts there are magnesium-based. Too much magnesium causes cell membranes to dissolve.
The Shackleton Glacier Valley almost literally a different pole from the Rift Valley lakes. In some parts of the ice-carved landscape, the soil has been exposed for hundreds of thousands of years to frigid, bone-dry winds. Nicholas Dragone -- a graduate student at the University of Colorado in Boulder -- took soil samples from those sites and from low-lying spots that were ice-covered until recently. Two months of attempting to get something to grow in the samples got nothing; DNA tests got nothing; and adding glucose with labeled carbon, in hopes of getting labeled CO2 from metabolism, got nothing. The soil was thoroughly dead.
Multiple types of bacteria and fungi have been found in lowland soils, but some of the high-elevation, older soils were devoid of life. Dragone suspects these high-elevation soils are inhospitable because they lack any liquid water.
COMMENT ON ARTICLE* INTERNET OF THINGS (3): Google and Amazon.com are best positioned to impose order on smart-home technology, thanks to their smart-speaker product offerings. According to Ben Wood, until fairly recently, the assumption was that the smartphone would be the central controller for a smart home. Not so: "Pulling out your phone, unlocking it, tapping an app, then using it to turn the lights on, is much more complicated and annoying than simply walking across the room and pushing a button". It's even easier with voice commands.
Amazon's Alexa and Google Home, the two firms' smart-speaker products, already have greater market penetration than rival smart-home hubs. An estimated 78 million smart speakers were sold in 2018, more than twice as many as in 2017, with Amazon and Google accounting for roughly a third each. The remainder were mostly sold by the Chinese tech companies Alibaba, Xiaomi and Baidu. Surveys suggest that about a quarter of American smart-speaker owners use them to control at least one other device.
Smart speakers, designed to control a home network, and wireless connectivity has made the smart home practical. Integration and control is no longer so much of a problem, and there's no need to wire up a home to connect all the nodes in the network. In addition, a smart home can be set up incrementally, a node at a time, as a user sees fit. Consumer-goods firms are increasingly eager to ensure that their devices can work with Google and Amazon's smart speakers. Many tech firms have set up certification programs, as well as smartphone-style app stores aimed at third-party developers who want to integrate Google Home and Alexa with their own products.
The tech companies have two goals in this. First, by controlling the smart home, and much else besides, the smart speaker will become a dominant technology, in much the same way as a smartphone. Put simply, Google and Amazon want to duplicate the success of the iPhone. Second, that smart home will be a gold mine of data -- for example what users watch on TV, what's in their smart fridge, and even the pattern of light usage.
There are already tales of disputes between smart-device makers and the tech giants over who has access to those data, and how much must be collected. So far, however, most users aren't really aware of the privacy issues involved. Some who are, refuse to have smart speakers in their homes.
* As we head into the age of the smart home, we also are entering the age of the smart office. When German industrial giant Siemens rebuilt its offices in the Swiss town of Zug, they were built green from the ground up. Water from nearby Lake Zug is piped in and fed through pumps to heat or cool the offices. None of the materials used in the building came from more than 800 kilometers (500 miles) away. Rain that falls on its grass-covered roof is collected to flush the toilets.
The buildings were also digitally enabled from the ground up, the buildings being designed partly as a showcase for the firm's new "Smart Infrastructure" division. Some of the smarts in the building are there for the workers: an app named "Comfy" -- made by an American firm named Building Robotics that Siemens bought in 2018 -- allows workers to adjust temperature and light levels in their offices with their phones. Over time, the system will learn the preferences of individual workers, and automatically warm or cool their offices. The app can also be used to find unoccupied desks; check the cafeteria's menu; book meeting rooms; and request maintenance, for example replacing a broken chair.
Other features are intended for managers. The building is wired with hundreds of sensors made by Enlightened, another US company, which Siemens also bought in 2018. The sensors are integrated with the building's light fixtures, which supply power; and come with a low-resolution infrared camera, a Bluetooth networking beacon, plus sensors to measure energy consumption, air temperature and light levels. Individual sensors can hook up to form a wireless network.
Christoph Leitgeb, the building's designer, says the sensors have a wide range of uses. They can keep track of daylight levels, turning up the lighting on gloomy days, and cutting back on sunny ones. The result can be a 38% saving in energy consumption. Building Robotics claims that better lighting can boost employees' productivity by 23%. The infrared cameras can be used to track employees. That data can be converted into a heat map of the building, showing popular areas and less-traveled ones, helping managers make the best use of space. Occupancy data can be fed to the heating systems, allowing energy savings when the building is sparsely populated.
For now, data gathered by sensors in the Siemens building are anonymous. The cameras see heat blooms, meaning they can record only numbers and general circulation within a building. Leitgeb says that more personal tracking is possible, via the sensors' Bluetooth beacons, which could track smartphones or building passes. So far Siemens is not making use of that capability, but is engaged in discussions with workers about it. [TO BE CONTINUED]
START | PREV | NEXT | COMMENT ON ARTICLE* ANOTHER MONTH: COVID-19 is, to no surprise, leading to a lot of public hysteria, in particular chasing after quack cures. The one that took center stage, after being boosted by Trump, was hydroxychloroquine -- a drug used to treat malaria, and also to suppress auto-immune reactions. The main push for hydroxychloroquine was from an eccentric French researcher named Didier Raoult, who published a study claiming remarkable improvements in COVID-19 patients from use of hydroxychloroquine.
The study was preposterous, covering only 20 patients; worse, it started out with 26, with 6 removed for awkward reasons, for example dying. Similar and also sketchy studies showed no improvement in patient outcomes from use of hydroxychloroquine. On top of that, Raoult is a climate-change denier who also denounces randomized double-blind medical trials. He runs a research institute that appears to be devoted to churning out papers for him to sign, while he cultivates Rightist French politicians. His bogus study was leaked to Fox News, to then make its way to Donald Trump.
It is not entirely clear that using an immune-system suppressant is a good way to treat someone with a raging viral infection. In any case, proper trials of hydroxychloroquine are underway. An antiviral drug named remdesivir appears much more promising, Dr. Anthony Fauci having put in a good word for it -- but it hasn't been properly qualified by trials, either.
Not so incidentally, Dr. Fauci was asked who should play him if SATURDAY NIGHT LIVE did a skit featuring him. He replied: "Brad Pitt, of course." And of course, SNL got Brad Pitt to do the skit, with Pitt getting in a few good shots:
BEGIN QUOTE:
I'm going to be there, putting out the facts for whoever's listening. And when I hear things like: "The virus can be cured if everyone takes the Tide pod challenge!" -- I'll be there to say: "Please don't."
END QUOTE
At the end of the skit, Pitt took off his wig and said: "To the real Dr. Fauci, thank you for your calm and your clarity in this unnerving time."
* The COVID-19 pandemic is producing literally a world of personal stories, varying in some ways from place to place, in some ways staying the same. Colorado has pushed for everyone to mask up in public; it doesn't offer much protection for oneself, but it helps to keep from contaminating others if infected.
I had some old bandanas in a box, they work as well as any other informal mask. I went to the King Soopers supermarket with a bandana first Tuesday of April, wondering how many people would be covered; it turned out to be the majority, if not the overwhelming majority. Some people are not quick to get a clue.
I was chatting with a Californian on Twitter who said he had been at a Safeway supermarket, with people social-distanced in a line, and a fellow started ranting about how foolish people were to go along with the pandemic hoax, that it was all fake news -- and then dared people to tell him if they knew anyone who had died. Two hands went up. He went quiet.
Anyway, the checkout clerks are masked now, and have plastic shields on their stations. I have to bag my own groceries, but I normally do anyway. The sales clerk -- a sweetheart, many of the customers are fond of her -- told me: "I'm sorry I can't help." I replied: "I wouldn't let you."
The store shelves are slowly starting to recover. Toilet paper is still scarce, but at least the store has breakfast cereal. I could understand the run on TP, but I was puzzled as to why there was a run on breakfast cereal. Anyway, I was running low on Sugar Crisp ... they call it Golden Crisp these days to duck the sugar bullet, nobody is fooled. The shortages persist. Meat looks like it will be a real problem -- that almost certainly being good news for makers of plant-based meats. It's an ill wind that blows no good.
Incidentally, in 1973 late-night TV host Johnny Carson heard a story that there might be a shortage of toilet paper, and told his TV audience: "There's a shortage of toilet paper." BOOM! It disappeared off the store shelves all over the USA.
And then, there's The Howl: at 8 PM, people go out and howl at the night sky. It's big down in Denver, it's caught on here in Loveland, I think it's going super-viral. I have a box of police-type whistles, and use one of them, too.
Alas, in the last ten days of April, I came down with a nasty case of hay fever, coughing all the time. There's never a good time to have hay fever, but with COVID-19 on the loose, walking around in public with a cough is likely to evoke fear and hostility.
* Thanks to three readers for donations this last month. They are much appreciated.
COMMENT ON ARTICLE