Here’s links and commentary for this week…..
1. AI’s balmy winter. Filip Piekniewski has a post arguing AI Winter Is Well On Its Way. It’s an excellent corrective to the past several years of AI hype. Recommended! That said, I didn’t find it completely convincing, though my objections are mostly definitional.
In general, technology follows what’s called the hype cycle:
And for AI the hype cycle gets far crazier. People go bonkers. Thinking machines! Terminators! We’re all doomed! Doomed I tell you! It’s always been the ultimate sexy tech topic, and always will be. Hence the special term for AI investment downturns: AI Winter. Wikipedia defines two major AI Winters 1974-1980 and 1987-1993. If we redrew the hype diagram above for AI, it’d have a much higher hype peak, followed by a trough of disillusionment that hit zero. That’s a winter. When funding for AI dries up completely.
But we don’t have a winter, at least not yet. Google continues to pour billions into machine learning, as do many other tech companies. So I agree with Piekniewski’s main point that AI hype is cooling off. But I’m less convinced we’ll see a real winter, where the trough of disillusionment is noticeably lower than other kinds of hyped tech (think google glass, internet of things, VR, blockchain etc.). For example Piekniewski says “By far the biggest blow into deep learning fame is the domain of self driving vehicles (something I anticipated for a long time, see e.g. this post from 2016).” He correctly cites Uber and Tesla hype and recent crashes and troubles. That’s fine. But Google subsidiary Waymo is the company to keep an eye on. In particular see this post As Uber and Tesla struggle with driverless cars, Waymo moves forward.
I guess we’ll find out for sure in another year or two.
2. Gene editing to fight maralria. Genes have a 50-50 chance of passing from either particular parent to their child. With gene drive, you can trick the reproductive system to always have one parent succeed in passing on their genes. And from there into an entire population. It’s a somewhat scary technology once you think about it. There are two possibilities for using gene drive to fight malaria: 1) make mosquito species which carry malaria resistant, or 2) make mosquitoes species that carry malaria extinct, by using gene drive so they only have males. About half a million people die of malaria each year. So I hope one of these two approaches is approved for trial. I suspect #1 is less risky, since some species of mosquito are already resistant to malaria, so it’d be less of a change. You can learn more by reading this explainer post, which starts out with an excellent first sentence: “Kevin Esvelt wants me to know that if I fuck up this article, 25,000 children could end up dead.”. link
3. SV Angel will no longer raise funds for seed rounds. The first money that founders of a company get is called seed capital, or a seed round. Quote: “When we started SV Angel, there were just a handful of funds investing in seed rounds.” That is to say, SV Angel was one of the first venture capitalist firms to fund seed rounds. But “Today there are thousands of firms and individuals investing in seed rounds.” So it’s big news that SV Angel will no longer raise funds for seed rounds, since they are one of the first to do it. With that context, I really liked the chart below from Semil Shah at Haystack. The next funding round after seed is Series A. And the fraction of companies making it to Series A keeps going down. While the money flooding into seed is going up, though looks like the peak was a few years ago. See Semil Shah’s post for more.
3. Moral Hazards and China. Thoughtful post by Tanner Greer on China’s gulag in Xinjiang, in the northwesternmost part of China . Let me quote this bit on how it’s easier (but a trap) to emphasize only with victims:
I take a similar approach to the history we study. I have studied many of the nastiest parts of modern history with my students. Slavery. Japanese war-mongering. The Holocaust. My approach to these atrocities is simple: it is not enough to empathize with the victims. That is easy. It is also mostly useless. The real challenge is to try and feel the emotions, understand the fears, and take seriously the ideas that lead perpetrators to commit the crimes they did. One must not just sympathize with the tyrannized–one must also try and sympathize with the tyrant.
Why is this necessary? Why focus just as much on the experience and fears of the slaver as the slave? Because you are far more likely to become a slaver than you are to suffer as a slave.
Xinjiang is different. Here a people is crying. They have been subjected to a new and frightening form of despotism, a terrible marriage of terror and technology. To enforce this new tyranny the Party has imprisoned one out of every twelve to one out of every six adults. Each has been subjected to torture (or the threat of it), insult, betrayal (or the threat of it), and an attack on all they hold sacred. Each has been plugged into an Orwellian system of surveillance that rates, rewards, and punishes them for everything they do or identify with. There is nothing else in our world like it.
There are moral hazards here.
Worth reading. link
4. Marshmallow test revisited. The Marshmallow Test paper was published in 1990, testing which children could resist eating a marshmallow while the adult left the room, getting two marshmallows if they waited until the adult got back. Those who waited had more academic and life success. Now a new replication paper has come out revisiting the test. There’s been some bad takes on the new paper, in particular Famed impulse control ‘marshmallow test’ fails in new research and Why Rich Kids Are So Good at the Marshmallow Test. The best review of the new paper I’ve read is by Jason Collins, who hits the key points here:
Given the original authors’ notes about effect size, and the differences in study design, the original findings have held up rather well. For a simple diagnostic, the marshmallow test still has a surprising amount of predictive power. Delay of gratification at age 4 predicts later achievement. Some of the write-ups of this new work have stated that the marshmallow test may not be as strong a predictor of future outcomes as previously believed, but how strong did you actually believe it to be in the first place?
The other headline from the replication is that the predictive ability of the marshmallow test disappears with controls. That is, if you account for the children’s socioeconomic status, parental characteristics and a set of measures of cognitive and behavioural development, the marshmallow test does not provide any further information about that future achievement. It’s no surprise that controls of this nature do this. It simply suggests that the controls are better predictors. The original claim was not that the marshmallow test was the best or only predictor.
Link to Jason Collins post.
5. Scientists estimate the mass of all life on Earth. The new paper has the chart below. Plants dominate. Animals had to be broken out to the right side just to see where we humans fit in. Apparently arthropods and fish are doing just fine.
Also see this Vox piece, which has nice visualizations as well.
And that’s all the links for this week!