Some Very Random Links

Peter Gray on Self-Directed Education; Peter Turchin tries to define a society; OpenAI on hallucinations; Ben Landau-Taylor on weapons and democracy

Peter Gray writes,

The difference between progressive education and Self-Directed Education lies in the understanding of how such whole-person education occurs. To the progressive educator it emerges from a collaboration between the child and a benevolent, extraordinarily competent teacher, who gently guides the child’s energy and shapes the child’s raw ideas in ways that serve the child’s and society’s long-term good. To the advocate of Self-Directed Education, it emerges out of children’s natural drives to understand themselves and the world around them and to use whatever resources are available in their environment, including knowledgeable and skilled others, to achieve that end.

To the advocate of Self-Directed Education, it is the child’s brilliance, not the teacher’s, that enables excellent education. The job of adults who facilitate Self-Directed Education is less onerous than that of teachers in progressive education. In Self-Directed Education adults do not need to have great knowledge of whatever a student might want to learn, do not have to try understand the workings of every child’s mind (a truly impossible task), and do not have to be masters of pedagogy (whatever that might be).

Rather, they simply must be sure that the child is provided with an environment that allows the child’s natural educative instincts to operate effectively.

I think that most of us would say, “self-directed education for me, but not for thee.” That is, we doubt that other people are motivated to learn what we think that they need to learn.

Peter Turchin writes,

Cooperation is a controversial concept because of the tension between group-level benefits (such as production of public goods) and individual-level costs. As a result, cooperation can easily unravel because of the temptation by individuals to free-ride on the contributions of others. How evolution overcomes this free-riding problem is explained in Ultrasociety (and the theory is then extensively tested and empirically supported in The Great Holocene Transformation).

My definition of society, then, is a collection of individuals that cooperate to achieve some common goal.

I balk at the phrase “common goal.” That sounds more like a team. Instead, I would describe a society as a set of people who are interdependent, without necessarily having a common goal. And I think of the issue of cooperation vs. defection as arising at two levels. The first level is the individual vis-a-vis the sub-Dunbar group, where people are watching you and judging you to make sure you cooperate. The second level is the group vis-a-vis the super-Dunbar organization or society, where a small group that cooperates well internally may be acting contrary to the well-being of the larger entity. The Biden entourage cooperated effectively to protect him, but that was not good for the rest of us.

Adam Tauman Kalai and others write,

language models are optimized to be good test-takers, and guessing when uncertain improves test performance. This “epidemic” of penalizing uncertain responses can only be addressed through a socio-technical mitigation: modifying the scoring of existing benchmarks that are misaligned but dominate leaderboards, rather than introducing additional hallucination evaluations. This change may steer the field toward more trustworthy AI systems.

The tyranny of metrics strikes again. If the metrics reward AI’s for making lucky guesses, then they will sometimes make bad guesses. The metrics instead should reward AI’s for admitting when they are guessing.

Pointer from Rowan Cheung.

Ben Landau-Taylor writes,

In Weapons Systems and Political Stability, the master historian Carroll Quigley explained that a society’s form of government is downstream of the balance of military force. When the best weapons system requires expensive equipment used by highly-trained specialists, like a medieval knight with his horse and lance and heavy armour, this concentrates military power into the hands of a small group. On the other hand, when the best weapons system is cheap and can be used effectively by amateurs with relatively little training — for example, muskets or rifles — military power is spread more evenly throughout the population.

…By the late 1700s, guns became widely affordable. With the rise of these amateur weapons, the new balance of military power was much more widely distributed, so over the 1700s and 1800s, revolts and reforms throughout Europe and the Americas shifted political power towards the people.

…In the 20th century, this phase ended. The weapons system based on amateur-friendly guns was supplanted by a series of weapons systems based on specialist equipment like airplanes and tanks and rockets.

…The 20th century decline of popular military power, then, led to the gradual but thorough transfer of power from elected representatives, who are somewhat constrained by the popular will, to bureaucracies and agencies able to act however they wish. In other words, within Western states which describe themselves as democracies, the people — the demos — have lost much of their power — their kratos. Elections have become less important in determining the government’s policy, and more ceremonial.

Hmmm.