AI and Polling

Posted on by
Reply

A letter-writer in Tuesday’s WSJ Letters section thinks he sees a use for AI in improving the quality of political polls, which for the last several years have been badly off the mark both in prediction election winners and in predicting the margins of victory when they did get the winner right.

He’s sort of right, and he’s wrong.

It [AI] can help researchers reach a more representative mix of people, examine results faster and track changes in opinion more often.

AI can certainly speed up “examinations” of results if it’s properly trained on truly balanced techniques. That, though, requires humans being diligent about scrubbing as much as possible their own biases, with those humans involved in the training coming from a variety of political bents so as to counterbalance remaining biases. This is a requirement too often honored in the lack.

Additionally, researchers already know how to reach a more representative sample. However, they’ve been, instead, deliberately oversampling one party over others in order to reach a predetermined outcome. AI will not address that; it cannot correct it if it could address it.

Finally, there’s not much need to track opinion changes more frequently than is already done; most of us already have polling fatigue and are not interested in being confronted by even more polls, even if we just decline to participate.

In the end, all AI is likely to do, at least until we get better trained human oversight, is speed up the rate at which the polls get the politics wrong.

An Appropriate Move

Posted on by
Reply

The Trump administration is adjusting its interpretation of Title IX as it concerns sexual assault, including rape. The change greatly reduces the requirement for colleges and universities to themselves investigate sexual assaults, including rape, where those occurred off campus. Critics, of course, complain that the move will simply increase the likelihood of further moves like the alleged Cornell student who lately has filed a civil suit against the school and the seven students she alleges raped her at an off-campus party.

The move may well encourage fee-seeking lawyers and deep pockets civil suits by alleged victims, but the plain matter is centered on who has the jurisdiction for investigation and criminal charges.

If the assault occurs off campus, even if in an off-campus fraternity or sorority house, it’s still primarily a city police matter, not a campus police mater. The fraternity and sorority house is off campus almost exclusively because it’s not supported or even sanctioned by the school, perhaps even evicted from the school’s campus. That severely limits campus police authority regarding those places, and campus police have no authority at all anywhere else off campus. Beyond that, off-campus means off campus, and so outside the jurisdiction of school authority of any sort.

Outside school jurisdiction, though, does not mean the alleged victim has no protections against assaults or no means of redress after the assault as some of the more hysterical criticisms would have it. Those protections and means of redress remain where they have always been: in the local, State, and Federal laws against any crime, including sexual assaults of any form, in the hands of local and State police to investigate, and in the hands of the State and Federal criminal trial courts.

AI Development and Testing, and Humans

Posted on by
Reply

OpenAI has decided not to release its latest iteration of ChatGPT, which it had labeled GTP 6.1 Astra. The company had safety concerns that researchers raised during internal testing.

The model was more capable than the company’s previous models in completing challenging tasks from end-to-end without human assistance, as well as writing.

But.

GPT-6.1 Astra regressed in two areas. Compared with its predecessor, GPT-6 Astra, the model performed poorly on tests measuring alignment, or how well the model adheres to what humans would like it to do. Specifically, GPT-6.1 Astra showed higher levels of deception: it wasn’t always honest about telling users of the actions it did or didn’t take.
Another issue was what OpenAI calls “scope authorization,” meaning that GPT-6.1 Astra would push ahead on a task without asking the user for permission, and would at times reach for external tools and services even if it might be unsafe.

This tale demonstrates pretty conclusively the criticality of humans having controlling involvement in AI sandbox construction, isolation, and enforcement and in testing. It also demonstrates the importance of corporate culture—i.e., corporate human managers—in creating and then enforcing the internal culture that supports and drives dedication to that human control of and involvement in development and testing.

It also proves that we humans are up to the task if we care enough about our responsibilities—closely akin to Madison’s and Franklin’s remarks that our republic requires a moral, virtuous people in order to survive.

A Misnomer

Posted on by
Reply

In a WSJ article centered on how to take advantage of higher bond yields, the news writer had this:

[Bond] Investors get a coupon payment that helps offset the decline in value the bond might experience if market yields rise. In other words, even if a bond’s price drops, it can still end up with a positive return.
…
Currently, 2-year Treasurys are yielding over 4.9%. That isn’t much less than what the 10-year Treasury is paying at over 5.2%.

That’s the misnomer. The news writer is using yield and coupon interchangeably. They aren’t. Yield is simply the dividend payment relative to the bond’s then-current market price. Coupon is the dividend payment the bond is required to pay under the terms of its issue, and it’s fixed—independent of the bond’s market’s price. Yield will rise and fall with the movement of the market’s pricing of the bond.

Neither 2-year Treasurys nor 10-year Treasurys nor any other Federal government debt instrument (or any other bond, whether corporate or municipal or…) pays yield. The pay only the coupon’s dividend. Calling the yield what the bond is paying is badly misleading.

 

Guys like this article’s writer really ought to know better, though.

Progressive-Democrats, Education, and Assimilation

Posted on by
Reply

California’s Progressive-Democrat governor, Gavin Newsom, is illustrating how thoroughly the Progressive-Democratic Party and the education of American children are an oxymoron, and this doesn’t even involve teachers unions.

Nearly 1 million California students designated as English learners will no longer have to pass a set of tests and benchmarks to exit the program, under a new law signed Sunday by [Progressive-]Democrat Governor Gavin Newsom.

Newsom’s—Party’s—move also illustrates the disdain that they have for American culture. Language is at the heart of any nation’s culture. It informs and strongly influences how a nation’s citizens think about themselves and how they understand their nation’s culture. By dumbing down English learning in California’s schools, Newsom and Party are deprecating the very thing that makes our nation American—our culture. Party doesn’t care that it’s very hard for immigrants to assimilate into our culture when they’re not required to become even reasonably conversant, much less proficient, in our language. But Party doesn’t care about American culture, either, so there’s that.