Month: December 2022

Posted on

Interesting articles, December 2022

A gem from a “CIA spy” this August: “In my assessment, Vladimir Putin is winning. Russia is winning. They’re winning in Ukraine but they’re also winning the battle of influence with the West. They’re winning in the face of economic sanctions. They’re winning…[Russia will take over all of southern Ukraine, including Mariupol, the Kherson region and Odessa] and into Moldova. I believe all of that will happen before the fall…Fall of this year.”
https://youtu.be/T3FC7qIAGZk?t=756

Henry Kissinger talks about the Ukraine War, and also the threat of AI:

“Ukraine has become a major state in Central Europe for the first time in modern history. Aided by its allies and inspired by its President, Volodymyr Zelensky, Ukraine has stymied the Russian conventional forces which have been overhanging Europe since the second world war. And the international system – including China – is opposing Russia’s threat or use of its nuclear weapons.

This process has mooted the original issues regarding Ukraine’s membership in Nato. Ukraine has acquired one of the largest and most effective land armies in Europe, equipped by America and its allies. A peace process should link Ukraine to Nato, however expressed. The alternative of neutrality is no longer meaningful, especially after Finland and Sweden joined Nato. This is why, last May, I recommended establishing a ceasefire line along the borders existing where the war started on 24 February. Russia would disgorge its conquests thence, but not the territory it occupied nearly a decade ago, including Crimea. That territory could be the subject of a negotiation after a ceasefire.”
https://www.spectator.co.uk/article/the-push-for-peace/

The reality of war: Drone footage of a Russian soldier religiously crossing himself during an enemy artillery barrage.
https://youtu.be/XjWFryBJMvs

These calculations show how a small nuclear weapon detonated a few hundred feet above the ground would produce almost no radioactive fallout–no significant amounts of radioactive dust would drift far from the site. If Putin became desperate, such a nuclear strike against a cluster of Ukrainian military units could make sense to carry out.
https://www.thedrive.com/the-war-zone/just-how-radioactive-are-low-yield-nuclear-weapons

Ukraine is struggling to maintain its huge force of foreign-made weapons (including Russian captures and donations from numerous Western countries) due to non-interchangeable spare parts, lack of familiarity with maintenance procedures, and other human factors. A military AI like Skynet wouldn’t have these problems, it would know how to maintain, fix and use every kind of weapon, or could figure it out quickly and never forget. It would also keep an up-to-date inventory of all weapons, equipment and parts it had. Logistics that human armies find impossible or too costly to support would become possible. Future machine armies might be LESS standardized than human armies.
https://www.thedrive.com/the-war-zone/ukraine-situation-report-kyiv-struggling-to-keep-captured-weapons-in-the-fight

I roll my eyes at people who claim old weapons aren’t obsolete so long as you only use them in roles that respect their limitations. This has come up a lot during the Ukraine War, with some troops getting WWI bolt action rifles and 60 year old tanks. An important criterion for being obsolete is whether your weapon has gotten so old that its value on the antique market is higher than what it would cost to buy a newer, better weapon intended for the same role. For example, a WWII T-34 tank that Ukraine has laying around could be sold to an international collector for $230,000, which is more than enough to buy a T-72. The same is true for the WWI bolt action rifles. Selling one would bring in enough money to buy a full auto AK-47.
https://www.rbth.com/defence/2017/04/26/how-to-buy-an-old-soviet-tank-for-the-price-of-a-fancy-car_750701

The suicide drones that Azerbaijan used against Armenia two years ago were cruder than I thought. They were obsolete biplanes, fixed up just enough to do a final flight, and packed with explosives. Human pilots flew them into the air, lined them up with their targets in Armenia, and then jumped out and parachuted to the ground. Russia is planning to copy this idea, but they actually have the money and technology to fit the biplanes with computer autopilots.
https://youtu.be/z7I3Illsuqg

The A-10 is badass, but also obsolescent: It is too vulnerable to newer air-to-ground missiles, and its mighty 30mm gun isn’t strong enough to punch through the advanced armor of modern tanks. The new idea is to use it for attacking ships, which it could do, albeit inefficiently. The recent moves by Azerbaijan and Russia to convert obsolete biplanes into expendable drones makes me suspect the same will be done to planes like the A-10 once the technology is better and cheaper. Sending old A-10s and 1960s tanks into battle against poor odds might make military sense if the vehicles are piloted by machines whose lives mean nothing.
https://www.businessinsider.com/a10-warthog-trying-a-new-role-decoying-enemy-air-defense-2022-12

The sad saga of Russia’s sole aircraft carrier continues: it caught on fire while in port.
https://www.thedailybeast.com/russias-only-aircraft-carrier-admiral-kuznetsov-catches-fireagain

The last Boeing 747 rolled off the production line, ending an era in passenger plane flight.
https://www.thedrive.com/the-war-zone/the-last-747-ever-built-has-rolled-off-boeings-production

After discovering large reserves of natural gas under the North Sea, Britain’s government instituted a national program to upgrade all appliances to use the fuel. Most of them had been designed to run off of “town gas,” which is made from coal. Over just eight years, 40 million appliances belonging to 14 million customers were modified. Once the global worker robot population gets into the tens of millions, major changes to infrastructure like this will become financially possible, such as changing national railroad gauges and electrical outlets. The path dependencies humans emplaced won’t last forever.
https://www.resilience.org/stories/2020-03-10/share-the-great-switch-lessons-from-when-14-million-homes-and-businesses-changed-fuel-in-less-than-a-decade/

Facebook has invented an audio compression algorithm that is 10x better than MP3. It allows high quality audio to be transmitted on low bandwidth channels.
https://ai.facebook.com/blog/ai-powered-audio-compression-technique/

‘[Researchers] from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL), Autodesk Research, and Texas A&M University came up with a method to automatically assemble products that’s accurate, efficient, and generalizable to a wide range of complex real-world assemblies. Their algorithm efficiently determines the order for multipart assembly, and then searches for a physically realistic motion path for each step.’
https://youtu.be/2Xw01yyg5So

A group of AI researchers found a way to beat a superhuman computer Go program, “KataGo,” by using an unusual gameplay strategy.
https://arstechnica.com/information-technology/2022/11/new-go-playing-trick-defeats-world-class-go-ai-but-loses-to-human-amateurs/

A shockingly advanced chatbot, “ChatGPT”, has been unveiled. It can quickly make short-essay-length responses to complex human questions. At this rate of advancement, the Turing Test will be passed this decade.
https://www.theguardian.com/technology/2022/dec/04/ai-bot-chatgpt-stuns-academics-with-essay-writing-skills-and-usability

DeepMind unveiled a new algorithm, called “AlphaCode,” which can write computer code as well as a novice human programmer.
https://www.science.org/content/article/ai-learns-write-computer-code-stunning-advance

An essay on how language models could lead to the creation of superintelligent AI. I like this passage: ‘The totality of humanity’s recorded knowledge about the world — our shared world model — is a lower bound on what language models can learn in the limit[3]. We would expect that sufficiently powerful language models would be able to synthesise said shared world model and make important novel inferences about our world that is implicit in humanity’s recorded knowledge, but which have not yet been explicitly synthesised by anyone[4].’
https://www.lesswrong.com/posts/MmmPyJicaaJRk4Eg2/simulators-thesis

This guy’s predictions about the next ~30 years of AI development are worth reading. I don’t agree with all of it, but it’s interesting and plausible.
https://www.lesswrong.com/posts/qRtD4WqKRYEtT5pi3/the-next-decades-might-be-wild

The U.S. government announced a “breakthrough” in fusion power technology–the achievement of “fusion ignition.”
https://www.energy.gov/articles/doe-national-laboratory-makes-history-achieving-fusion-ignition

Analysts say that, while the development was significant, its importance was exaggerated. Commercial fusion power plants are still decades away, and in fact may never be built.
https://news.yahoo.com/controlled-fusion-power-little-nearer-165242667.html

The technical impediments to making commercially viable fusion reactors are formidable. It’s not the miracle power source people assume it is.
https://thebulletin.org/2017/04/fusion-reactors-not-what-theyre-cracked-up-to-be/

A group of scientists developed a plan to slowly convert an asteroid into a hollow, cylindrical habitat with spin-based gravity.
https://newatlas.com/space/space-habitat-ring-plan/

For most of the 20th century, a sizeable minority of geophysicists believed that the Earth’s diameter had grown over billions of years. Among them, the most plausible explanation involved slow changes to the gravitational constant: If gravity gets weaker, then all the tiny rock particles that make up the Earth are not attracted to each other as much, so they spread out more, causing the whole planet to get wider.
https://www.chemeurope.com/en/encyclopedia/Expanding_earth_theory.html

Did the 1995 movie Species steal the plot of a 1961 British TV series A For Andromeda?
https://en.wikipedia.org/wiki/A_for_Andromeda
https://en.wikipedia.org/wiki/Species_(film)

Authoritarians on the left and right wings have essentially the same mindsets: Intolerance for people different from themselves, a lust for power, an instinct for what people on their side are thinking and which people in the group are powerful or weak, and an abnormal willingness to engage in political violence.

“It’s a mistake to think of authoritarianism as a right-wing concept, as some researchers have in the past,” he says. “We found that ideology becomes secondary. Psychologically speaking, you’re an authoritarian first, and an ideologue only as it serves the power structure that you support.”
https://phys.org/news/2021-09-left-wing-authoritarians-key-psychological-traits.html

There are “impossible colors.”
https://en.wikipedia.org/wiki/Impossible_color

When a male [Anglerfish] finds a female, he bites into her skin, and releases an enzyme that digests the skin of his mouth and her body, fusing the pair down to the blood-vessel level.[26] The male becomes dependent on the female host for survival by receiving nutrients via their shared circulatory system, and provides sperm to the female in return. After fusing, males increase in volume and become much larger relative to free-living males of the species. They live and remain reproductively functional as long as the female lives, and can take part in multiple spawnings.[4] This extreme sexual dimorphism ensures that when the female is ready to spawn, she has a mate immediately available.[27] Multiple males can be incorporated into a single individual female with up to eight males in some species, though some taxa appear to have a “one male per female” rule.[4]
https://en.m.wikipedia.org/wiki/Anglerfish

Britain’s NHS will start offering free genetic sequencing for outwardly healthy newborns, to screen for 200 genetic disorders. It’s a step in the right direction, but the tests should really be done before the babies are born, so the parents can know about any disorders beforehand and abort pregnancies the realize they can’t handle.
https://www.bbc.com/news/health-63906892

Jonas and Wyatt Maines are identical twin brothers. However, Wyatt gender-identifies as female, and used hormone therapy during their teens to help transition into a transwoman. Wyatt now goes by the first name “Nicole.” Their example shows that gender identity is at least partly rooted in brain development, and not exclusively in genetics, and their differences in appearance make for an interesting study of sexual dimorphism.
https://time.com/4074959/see-how-twin-boys-became-brother-and-sister/

Ear shapes are unique to people, and a new algorithm can identify people by their ears with 97.25% accuracy.
https://news.uga.edu/new-facial-recognition-technology-scans-your-ear/

Different races of people tend to have different brain shapes. There’s even shape variation within races, allowing a person’s ancestral region of origin to be determined with fair accuracy.
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4786069/

Pharmacists are still making remarkably poor use of automation. There’s no reason why highly-paid humans should still be spending their time counting pills instead of leaving the task to machines.
https://www.wsj.com/articles/cvs-tries-out-remote-system-to-help-fill-prescriptions-11670108339

The causal link between saturated fat consumption and cardiovascular diseases has always been based on shaky science.
https://journals.lww.com/co-endocrinology/Fulltext/2023/02000/A_short_history_of_saturated_fat__the_making_and.10.aspx

This Bill Gates COVID-19 prediction from August 2021 would have been right had it not been for China’s bungling: “You have to admit there’s been trillions of dollars of economic damage done and a lot of debts, but the innovation pipeline on scaling up diagnostics, on new therapeutics, on vaccines is actually quite impressive. And that makes me feel like, for the rich world, we should largely be able to end this thing by the end of 2021, and for the world at large by the end of 2022. That is only because of the scale of the innovation that’s taking place.”
https://www.wired.com/story/bill-gates-on-covid-most-us-tests-are-completely-garbage/

Posted on

We should let machines choose jobs for us

In the last few months, I’ve posted links to a few articles with related implications:

In summary, when it comes to picking fields of study and work, humans are bad at doing it for themselves, bad at doing it for each other, and would be better off entrusting their fates to computers. While this sounds shocking, it shouldn’t be surprising–nothing in our species’ history has equipped us with the ability to perform these tasks well.

Consider that, for the first 95% of the human species’ existence, there was no such thing as career choice or academic study. We lived as nomads always on the brink of starvation, and everyone spent their time hunting, gathering, or caring for children. Doing anything else for a living was inconceivable. People found their labor niches and social roles in their communities through trial-and-error or sometimes through favoritism, and each person’s strengths and weaknesses were laid bare each day. Training and education took the form of watching more experienced people do tasks in front of you and gradually learning how to do them yourself through hands-on effort. The notion of dedicating yourself to some kind of study or training that wouldn’t translate into a job still payoff for years was inconceivable.

For the next 4.9% of our species’ existence, more career options existed, but movement between them was rare and very hard. Men typically did what their fathers did (e.g. – farmer, merchant, blacksmith), and breaking into many career fields was impossible thanks to restrictions on social class, race, or ethnicity. For example, a low-caste Indian was forbidden to become a priest, and a black American was forbidden admission to medical school. Women were usually prohibited from working outside the home, and so had even less life choice than men. The overwhelming majority of people had little or no access to information or ability to direct their courses of their own lives.

Only in the last 200 years, or 0.1% of our species’ existence, have non-trivial numbers of humans gained the ability to choose their own paths in life. The results have been disappointing in many ways. Young people, who are naturally ill-equipped to make major life choices for themselves, invest increasingly large amounts of time and money pursuing higher education credentials that turn out to not align with their actual talents, and/or that lead to underwhelming jobs. In the U.S., this has led to widespread indebtedness among young adults and to a variety of toxic social beliefs meant to vent their feelings of aggrievement and to (incorrectly) identify the causes of such early life struggles and failures.

The fact that we’re poor at picking careers, as evidenced by two of the articles I linked to earlier and by a vast trove of others you can easily find online, isn’t surprising. As I showed, nothing in our species’ history has equipped us with the skills to satisfactorily choose jobs for ourselves or other people. This is because nowhere near enough time has passed for natural selection to gift us with the unbiased self-insight and other cognitive tools we would need to do it well. If choosing the right field of study and career led to a person having more children than average, then the situation will be different after, say, ten more generations have passed.

Ultimately, most people end up “falling into” jobs that they are reasonably competent to perform and for which they have modest levels of passion, a lucky few end up achieving their childhood dreams, and an unlucky few end up chronically unemployed or saddled with jobs they hate. (I strongly suspect these outcomes have a bell curve distribution.)

As I said, the primary reason for this is that humans are innately mediocre judges of their own talents and interests, and are not much better grasping the needs of the broader economy so they can pursue careers likely to prosper. In the U.S. I think the problem is particularly bad due to the Cult of Self-Esteem and related things like rampant grade inflation and the pervasive belief that anyone can achieve anything through hard work. There aren’t enough reality checks in the education system anymore, too many powerful people (i.e. – elected politicians, education agency bureaucrats, and college administrators) have vested interests in perpetuating the current dysfunctional higher education system, and our culture has not come around to accepting the notion that not everyone is cut out for success and that it’s OK to be average (or even below average).

And I don’t know if this is a particularly American thing, but the belief that each person has one, true professional calling in life, and that they will have bliss and riches if only they can figure out what it is, is also probably wrong and leads people astray. A person might be equally happy in any one of multiple career types. And at the opposite end of the spectrum are people who have no innate passions, or who are only passionate about doing things that can’t be parlayed into gainful employment, like a person who absolutely loves writing poetry, but who also writes poor-quality poetry and lacks the aptitude and creativity to improve it.

Considering all the problems, letting computers pick our careers for us should be the default option! After all, if you’re probably going to end up with an “OK” career anyway that represents a compromise between your skills and interests and what the economy needs, why not cut out the expensive and stressful years of misadventures in higher education by having a machine directly connect you with the job? No high school kid has ever felt passionate about managing a warehouse, yet some of them end up filling those positions and feeling fully satisfied. 

Such a computer-based system would involve assigning each human an AI monitor during their childhood. Each person would also take a battery of tests measuring traits like IQ, personality traits, and manual dexterity during their teen years, performed multiple times to compensate for “one-off” bad test results. Machines would also interview each teen’s teachers and non-parent relatives to get a better picture of what they were suited for. (I’m resistant to relying on the judgements of parents because, while they generally understand their children’s personalities very well, their opinions about their children’s talents and potential are biased by emotion and pride. Most parents don’t want to hurt the feelings of their children, want to live vicariously through them, and like being able to brag to other people about their children’s accomplishments. For those reasons, few parents will advise their children to pursue lower status careers, even if they know [and fear] that that is what they are best suited for. )

After compiling an individual profile, the computer would recommend a variety of career fields and areas of study that best utilize the person’s existing and latent talents, with attention also paid to their areas of interest and to the needs of the economy. At age 18, the person would be enrolled in work-study programs where they would have several years to explore all of the options. It would be a more efficient and natural way to place people into jobs than our current higher education system. By interning at the workplaces early on, young adults would get an unadulterated view of important factors like work conditions and pay.

And note that, even among highly successful people today, it’s common for their daily work duties to make little or even no use of what they learned in their higher education courses. Some argue that a four-year college degree is merely a glorified way of signaling to employers that you have a higher than average IQ and can stick to work tasks and get along with peers in pseudo-work settings reasonably well. Instead of charging young people tens or hundreds of thousands of dollars for those certifications, why not do it earlier, less obtrusively, and much cheaper through the monitoring and testing I described?

While I think a computer-based system would be better for people on average and in the long run, it would also be psychologically shattering to many teenagers who got the bad news that their dream career was not in the cards for them. However, it is also psychologically shattering to pursue such dreams and to fail after many years of struggle and financial expenditure. Better to get over it as early as possible, and to enter the workforce faster and as more of an asset to the economy, with no time and money wasted on useless degrees, dropped majors, and career mistakes.

Finally, the same level of technology and of its integration into the workforce could raise the value of capital throughout each person’s career arc. AI monitors would detect changes to each person’s skill sets and knowledge bases over time, as old things were forgotten and new things were learned. Having an up-to-date profile of a worker’s strengths and weaknesses would further optimize the process of linking them with positions for which they were best qualified. And through other forms of monitoring and analysis, AIs would come to understand the unique demands of each line of work and how those demands were changing, and to custom tailor continuing education “micro-credentialing” for workers to keep them optimized for their roles.