[Lex Computer & Tech Group/LCTG] Goodhart's Law in Education

[JOHN M BROWN]

This item, "Cambridge schools are divided over middle school algebra, <https://www.bostonglobe.com/2023/07/14/metro/cambridge-schools-divided-over-middle-school-math/?et_rid=696624770&s_campaign=todaysheadlines:newsletter>” in the Globe may be an illustration of the strong form of Goodhart’s Law (… or, at any rate, of the law of unintended consequences) as applied to education -

"The district’s aim was to reduce disparities between low-income children of color, who weren’t often represented in such courses, and their more affluent peers. But some families and educators argue the decision has had the opposite effect, limiting advanced math to students whose parents can afford to pay for private lessons, like the popular after-school program Russian Math, or find other options for their kids, like Udengaard is doing." 

Regardless, the response to the measured problem was a cop out.  Having detected a racial disparity statistically, rather than taking steps to help poor kids move up, they sought to correct the statistic by keeping affluent kids down.  

   John B




On Jul 10, 2023, at 1:29 PM, Jon Dreyer <jon at jondreyer.org> wrote:

By the time I got through the first few paragraphs, I remembered that, over a decade ago, I saw the danger of what he calls "overfitting" in education and coined what I called "the implicit motto of education reform":

If you can't measure what you value, value what you can measure.

So I was not surprised to see education as the first on the list of examples of the the danger of overfitting in section 3.

I'd not read of Goodhart's Law, either the weak or strong forms, but I had discovered the idea. It seems increasingly important to keep in mind as technology makes it ever easier to measure things. A few examples I don't think I saw in the article are measuring executive performance by measuring quarterly financials or measuring software engineers by lines of code.

-- 

Jon "Measure Twice, Gut Once" Dreyer
Math Tutor/Computer Science Tutor <http://www.passionatelycurious.com/>
Jon Dreyer Music <http://music.jondreyer.com/>



On 7/10/23 8:40 AM, Harry Forsdick wrote:
> Interesting use of ML/AI:
> 
> Too much efficiency makes everything worse: overfitting and the strong version of Goodhart's law
> https://sohl-dickstein.github.io/2022/11/06/strong-Goodhart.html
> -- Harry


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.toku.us/pipermail/lctg-toku.us/attachments/20230715/fd5bc335/attachment.htm>