Links for 2022-03-22
China is going to train a 100 trillion+ ‘brain-scale’ model. The program is a collaboration between Tsinghua University, Alibaba Group, Zhejiang Lab, and the Beijing Academy of Artificial Intelligence. https://keg.cs.tsinghua.edu.cn/jietang/publications/PPOPP22-Ma%20et%20al.-BaGuaLu%20Targeting%20Brain%20Scale%20Pretrained%20Models%20w.pdf
DeepMind co-founder launches new AI company: “we will soon have the ability to relay our thoughts and ideas to computers using the same natural, conversational language we use to communicate with people. Over time these new language capabilities will revolutionize what it means to have a digital experience.” https://greylock.com/portfolio-news/a-new-paradigm-in-human-machine-interaction/
Demis Hassabis on the subject of AGI: "So I think that the progress so far has been pretty phenomenal. I think that it's coming relatively soon in the next you know--I wouldn't be super surprised--in the next decade or two." https://youtu.be/GdeY-MrXD74?t=428
Japan’s ten year cancer survival rate has increased to 58.9%. This is a disease that was once said to be incurable, but the survival rate has steadily increased alongside medical advances. Prostate cancer in Japan now has the highest survival rate at 99.2% https://www.nippon.com/en/japan-data/h01168/
Why Agent-Based Modeling Never Happened in Economics https://economistwritingeveryday.com/2022/03/14/why-agent-based-modeling-never-happened-in-economics/
Hoverfly Brains Mapped to Detect Sound of Distant Drones — “The bio-inspired signal-processing techniques detected the acoustic signatures of drones from nearly four kilometers away, improving the detection rate by up to 50% over existing methods.” https://www.unisa.edu.au/media-centre/Releases/2022/hoverfly-brains-mapped-to-detect-sound-of-distant-drones/
Go big or stay home! Small neuroimaging association studies just generate noise. http://www.wiringthebrain.com/2022/03/go-big-or-stay-home-small-neuroimaging.html?spref=tw
Tropical Methane Emissions Explain More Than 80% of Recent Changes in Global Atmospheric Methane Growth Rate https://www.nature.com/articles/s41467-022-28989-z
Physicists find direct evidence of strong electron correlation in a 2D material for the first time. The discovery could help researchers engineer exotic electrical states such as unconventional superconductivity. https://news.mit.edu/2022/electron-correlation-2d-graphene-0317
Organic Insecticides More Damaging to Beneficial Insects Than Synthetic Alternatives https://elifesciences.org/articles/73812
War links:
“The Russian campaign, if it has not already, is about to culminate.” https://threadreaderapp.com/thread/1505621690767749123.html
War in Ukraine: ‘Maybe it is good the scared ones left. Maybe it makes those who stayed stronger’ https://www.sundaypost.com/fp/maybe-it-is-good-the-scared-ones-left-maybe-it-makes-those-who-stayed-stronger/?utm_source=twitter
In 1919 the Soviets decisively lost a war with Poland, a much smaller but extremely resilient opponent who beat them decisively even after Soviet troops nearly reached Warsaw. Only in 1939 Moscow conquered Poland in alliance with Nazi Germany https://www.britannica.com/summary/Russo-Polish-War-1919-1920
One million Soviet troops invaded Finland in 1939, but they suffered such gigantic losses that they had to settle in a deal that cost the Finns territory but preserved their independence. https://www.britannica.com/event/Russo-Finnish-War
Western systems of government and justice are obviously not perfect. They are fallible and their shortcomings are being exploited by bad actors. Indeed, there is no perfectly coherent ethical, judicial, or governmental system known to humanity; taken to their logical conclusions, all of them suffer from fatal flaws.
A deontological ethical theory might suggest that you must tell a murderer that you are hiding their victim because lying is always wrong. Consequentialism would suggest that you give a mugger all your money if they threaten to use their powers from outside the Matrix to torture a Graham's number of people.
This is the gist of the AI alignment problem; the problem of ensuring that superhuman artificial intelligence does aid us rather than harm us. In other words, if you were to codify your favorite set of rules and implement it in an agent that was for all practical purposes infinitely powerful, would its execution steer the future towards outcomes ranked higher in your preferences? Almost certainly not.
Luckily, we're not dealing with superintelligences when we consider what system of government suits us better. More importantly, even though all of them are deeply flawed, there are better and worse alternatives.
In short: It's always possible to point out flaws. But do you have an alternative that is less wrong?