Social Learning Links, 2/19/2025
N.S. Lyons on Strong Gods; Jack Despain Zhou and Lillian Tara on education; David Donaho on frictionless research exchange; Tyler Cowen on AI in education
Energetic national populism is, then, a rejection of all the core obsessions and demands of the twentieth century and the open society consensus that so dominated it. The passionless reign of weakness, tolerance, and drab universalist utilitarianism being held up as moral and political ideals seems to be ending. And that means the gerontocracy of the Long Twentieth Century is finally dying off too. This is what Trump, in all his brashness, represents: the strong gods have escaped from exile and returned to America, dragging the twenty-first century along behind them.
The phrase “strong gods” comes from a book by R.R. Reno, which perhaps I should read. Reading Lyons’ post (which is worth doing) makes me think of Trump vs. his opponents as comparable to Randle McMurphy vs. Nurse Ratched. I worry that it will turn out similarly.
Jack Despain Zhou and Lillian Tara write,
We've spent over fifty years with a system whose core goal is not to lift students, but to close all gaps between them. Education schools have aligned policy and research towards that end, dismissing, deprioritizing, and attempting to discredit everything else. But why would you expect that to work? You can’t eliminate individual human variance with policy any more than you can legislate away height differences.
If your goal is to close all gaps and leave every student at the same level, education doesn’t work, and we have a fifty-year policy graveyard reminding us of that.
But what if, instead of pretending differences don't exist, we built an adaptive, responsive system that embraced them? What if our goal is to raise the ceiling, to see just how high each child can reach? If that’s the goal, we’ve barely begun to imagine how well education can work.
They take a view about the potential for policy innovation that is the opposite of what I call the Null Hypothesis. I am afraid that policy innovations tend to not show significant effects. What effects one does find are subject to fadeout—the effect you find in 3rd grade is gone by 8th grade. And effects usually fail to replicate. I think that this failure to replicate is due to social desirability bias—a study that shows a can-do result of an intervention will get a lot of play in the press, but it turns out to be just luck.
I hope that their center for educational progress succeeds. The time may be right for new approaches that are more rigorously tested.
We are entering an era of frictionless research exchange, in which research algorithmically builds on the digital artifacts created by earlier research, and any good ideas that are found get spread rapidly, everywhere…
developers implemented globally accessible code and model repositories, data repositories, and, eventually, globally visible task metric leaderboards. ..
participants are constantly bringing something (code, data, results), and taking something (code, data, new ideas), from the exchange
He argues that a key element in this is the Challenge, meaning a well-defined, objective measure of success.
in the 1980s through the 1990s and beyond, challenges were mounted in speech processing, biometric recognition, facial recognition, and other fields
Of course, I believe we could do this with intellectual debate. Just as AI systems are instantly scored, we could use LLMs to instantly score a public intellectual’s essay on how well it deals with objections to the argument, weaknesses in the argument, and taking a charitable view of those who disagree.
He argues that these three elements make it easy to make rapid progress in digital learning.
Given them some topics to investigate, and have them run a variety of questions, exercises, programming, paper-writing tasks — whatever — through the second or third-best model, or some combination of slightly lesser models.
Have the students grade and correct the outputs of those models. The key is to figure out where the AIs are going wrong.
As of now, one of the most important things for people to learn about is AI. Tyler is thinking of ways to do that.
substacks referenced above:
@
@
Really fantastic NS Lyons post this time. Could hardly get a better description of Yascha Mounk's
" passionless reign of weakness, tolerance, and drab universalist utilitarianism being held up as moral and political ideals" which Lyons claims, and I hope, "seems to be ending."
Lyons gives the steelman ideal: "The crusade for openness took on for itself a great commission to go and deconstruct all nations in the name of peace, prosperity, and freedom." If it had worked, it would have been fine. It failed. Because humans are not equal, and cultures are not equal, and the forced toleration of law violation to support DEI is a clear violation of the Rule of Law.
If "liberals" are unwilling to support laws against perjury, bribery, destruction of evidence, and of course border laws, than we don't have Rule of Law, we don't have L Liberalism. We have crony liberalism & crony capitalism.
Trump: not the symptom, nor cause, but the embodiment of the "End of the Era".
Thank God.
I do.
With reasonable fears about the bad stuff that might follow in the Creative part, after quite a bit more Destruction.
"We are entering an era of frictionless research exchange, in which research algorithmically builds on the digital artifacts created by earlier research, and any good ideas that are found get spread rapidly, everywhere…"
I'm skeptical of this claim. It would be an ideal world for scientific advancement, but from what I've seen, scientists and other researchers are very jealous of protecting their data and methods. If this is true, and stays true, they will learn to obscure the key parts of their research, so that others can't appropriate it easily. Part of this, I'm afraid, is to assure that sloppy methods, p-hacking, and outright fraud are hard to detect.
As one minor example, I remember the "Climategate" incident from 2010. In that incident, some researchers refused to release their data, even though the research was done at publicly-funded universities and the data was owned by the government and should have been disclosed under freedom of information requirements. Phil Jones said, in one of the leaked emails, that he would destroy his data rather than release it to opponents. And then his data disappeared, although he denied deliberately destroying it.