MundaneBlog

November 24, 2024

ChatGPT in the classroom – one thought experiment

Filed under: AI,Technology — Tags: , , — DrMundane @ 2:08 pm

Reading Mystery AI Hype Theatre 3000 this morning, on ChatGPT having no place in the classroom, and I tried out a little thought experiment. It goes something like this:

Let us assume we are in, say, an AP Literature class. I imagine the hardest part of grading such a class in reading and scoring several essays for each student over the year. I know for sure my terrible longhand cursive was a painful thing to read for my poor teacher. But what about an AI? Can the AI (unable to come to any factual conclusion or reason, just statistically generate) be used to speed this up?

I think the major problem is the inability to actually understand the students writing, but let us take for granted that improvements to large language models will somehow be able to overcome this fact with sufficient data and computation. That seems to be the claim of Altman and the others, but to be clear, they will not.

We know that currently these LLMs scrape data from the public internet, so can safely assume that they do have a robust source of information on AP Literature questions, books, themes, etc., from forums and other sources seeking to help students. Much of the writing will be done by students themselves, so another bonus is that it is writing representative of the population. So far that seems reasonable.

But there is already one problem. The writing is not representative of the whole population of AP Literature students, only those with access to these online resources and/or willingness to engage with them. I was a student who would never workshop their ideas or practice their writing online. I was simply too private for that sort of thing. The dataset certainly would never include my writing.

I would therefore argue that any such LLM grading students work would be inherently biased against students whose writing does not line up with the LLMs source material. It knows which essays deserve a 5 and which deserve a 1 based on it’s dataset. Based on statistics. It does not understand the rubric or intent, and therefore is unable to rate new (to it) but correct writing in a fair manner. You will therefore teach students to write like the corpus of text ingested by the model, not to find their voice and style. You teach them to look at the prompt and come up with the ‘correct’ answer, not an answer that necessarily comes from their own experience and understanding of the literature in question.

I think this alone makes the use of AI impossible in the classroom, owning to the discriminatory potential.

Another example that springs to mind: what about queer analysis? What about LGBTQ+ students? Will their viewpoints and experiences be reflected in the corpus? I doubt this highly. This means any such student may be able to write a brilliant analysis of a book through a queer lens and it will matter not, because statistically it doesn’t match what a ‘5’ is to look like. It uses all sorts of words that probably aren’t associated with ‘5’ essays. The LLM may even deem them ‘profane’. It therefore is not a ‘5’, in the view of the LLM.

I think these two thought experiments illustrate why I believe, beyond all the technical problems and overselling, that even a GPT that lives up to the hype and can be made factually correct will never be suitable for any evaluative work. Used to such an end, the AI encourages normative expression and discourages breaking boundaries. It truly discourages real feeling and art.

November 21, 2024

Links you should read – 2024-11-20

Filed under: Daily Links — Tags: , , — DrMundane @ 1:35 am

I don’t know why I love starting my morning with garbage, but it stirs me to write so…

NYT (archive) piece on Democrats and their approach to transgender rights moving forward.

The ‘politics’ of it all does not stir me, I think it’s obvious they must try and show some real humanity. They must speak clearly in favor of and make the case for transgender rights.

The part that got me to write was the following:

“Roughly 55 percent of voters said support for transgender rights in government and society has gone too far, according to AP VoteCast. More than 60 percent of adults say transgender women and girls should not be allowed to compete in sports with other women and girls, a Washington Post-KFF poll found. And strong majorities oppose minors’ using medications or hormone treatments, according to the Post-KFF poll.

At the same time, more than 60 percent of Americans support protecting transgender people from discrimination, according to the Pew Research Center. Most also oppose the government banning gender-affirming care for minors, including medication and surgery, Gallup found.”

I will leave the sports issue for a later time, what gets me is the poll numbers from the Post-KFF poll. I started to dig into that, but the righteous anger at that piece from me is a whole post in itself. I don’t give a damn what polls say on the matter of human rights. No amount of context makes horse race “65% of German people polled in 1941 think we need a solution to the Jewish Question” writing on these issues better.

I don’t think such reporting will serve to educate the American public any more on these issues. I don’t know what my point is, I just get wound up and this is the outlet. It’s my blog after all so ha!

I do need to pull one quote from the Post-KFF poll though: (photo caption and article text)

Valarie Johnson of LaBelle, Fla., is clear on her stance when it comes to discussing LGBTQ issues in the classroom. “Why would you introduce that subject to children when it has no life skills?” she said. (Saul Martinez for The Washington Post )

Valarie Johnson, 67, of LaBelle, Fla., said there’s no place in school for these sort of discussions. “Is that going to help these young people get a good job or a good spouse?” she asked. “Why would you introduce that subject to children when it has no life skills?”

I think a real education on gender, what it actually means and how sex and gender and sexuality are not linked would benefit everyone in having good relationships. Furthermore, for any trans youth or youth with any gender or sexuality that is not straight ahead heteronormative, the power of hearing that you, as you exist, are not bad or wrong is tremendous. How will you ever find a good spouse if you think you are essentially bad or wrong? How can one get a good job when you suffer the stress of presenting the ‘right’ kind of person based on what others perceive you ‘should’ be?

It is easy to cast stones at the theory and teaching of gender when your gender is set up by the world as the ‘correct’ interpretation (cisgendered and passing). Gender seems to not exist because yours (and the binary other) is allowable to it and privileged.

A quick NYT rip again “Speaker Mike Johnson Says He Will Ban Transgender Women From Capitol Bathrooms

Never be fooled into thinking their ‘concerns’ (hate) stops with children. Children are being used as a tool. Once they have power they intend to target every trans person.

“Sarah McBride doesn’t get a say,” Ms. Mace told reporters on Monday. “I mean, this is a biological man.” Ms. McBride, she added, “does not belong in women’s spaces, women’s bathrooms, locker rooms, changing rooms — period, full stop.”

Straight out of the TERF playbook. The casual cruelty of these people shocks me (I am far too hopeful it seems).

Next, a neat? series of photos (and a link to a book with more) by the Stasi.

Finally, a positive read from my archives. A very interesting blog about trauma surgery and what not: Doc Bastard

I meant to write more on tech, but got to the end of the day without getting the chance and needed to wrap up.

November 17, 2024

Links you should read 2024-11-16

Filed under: Uncategorized — Tags: , , , , — DrMundane @ 12:02 am

Staring at 404media.co, with Becky Ferreira writing this weeks edition of ‘the Abstract’. The portion that really stirs my mind is the story on AI poetry, and how in one test readers preferred the AI generated poems. Why? To quote the article quoting the research

“Non-expert poetry readers expect to like human-authored poems more than they like AI-generated poems,” said authors Brian Porter and Edouard Machery of the University of Pittsburgh. “But in fact, they find the AI-generated poems easier to interpret; they can more easily understand images, themes, and emotions in the AI-generated poetry than they can in the more complex poetry of human poets. They therefore prefer these poems, and misinterpret their own preference as evidence of human authorship.”  

I must say I am disheartened by this result. Not particularly surprised. As far as I am concerned much of the joy of poetry is in chewing on it. I have had great conversations on the poems I send out on my Christmas cards, and I feel this only because both of us had really thought on the poem (A burdock clawed my gown, not burdocks blame, but mine…) and then weeks later had a chance to discuss what it means to us. Luckily poetry will live on, I have no fear of that. We may just have to sing its praises louder.

On the politics front, Orac at Respectful Insolence describes why exactly RFK Jr. is bad news as HHS secretary. I can’t help but agree with him that the media has done a fine job of making RFK Jr. seem far more palatable than he is, and I expect that trend to continue. They will cover this administration ‘as usual’, so they can maintain their access.

In a surprising turn of events, I agree with Bill Clinton. Per the NYT:

President-elect Donald J. Trump made Vice President Kamala Harris’s support for transgender rights a core part of his argument that she was outside the political mainstream. His campaign used video of Ms. Harris expressing support for taxpayer-funded transition surgeries for transgender inmates in a torrent of ads that declared: “Kamala is for they/them. President Trump is for you.”

The Harris campaign largely did not answer those ads, but, internally, the Democratic Party was roiled by them. Former President Bill Clinton was said to have urged the Harris campaign to respond to them, and to have been told that they were not making an obvious dent in the race.

I had certainly felt (and have not written it… argh) that the Democratic Party had really failed in any messaging defending trans people, and this to me sounds exactly like how I expected them to talk about it. Running right down the middle and not standing for anything. Take it for granted that you will get their votes. Such an approach is not up to the seriousness of this moment. Token resistance will not be enough. Words will not be enough. Definite action will be needed. Will they be willing to pay the political cost? I doubt it.

I’m sorry. Perhaps I am too negative.

To end on a positive note, more from my archive.

November 16, 2024

Links you should read – 2024-11-15

Filed under: Daily Links,Surveillance,Technology — Tags: , — DrMundane @ 3:02 am

To start out the roundup, Karl Bode at Techdirt on Canada’s new right-to-repair law. See also Doctorow on Pluralistic covering the same for some further explanation. Controlling our devices is the first step to controlling our data, and in an America that is growing more authoritarian one must protect themselves and their data. Right to repair also means a right to disassemble, understand, and verify. Only when we fully know our devices can we fully trust them.

Following up on that, a guide from WIRED on protecting your privacy. Small steps.

Back to government surveillance, with a 404 media piece on the use of location data by the government (warrant required? Unclear). Even taking the assumption that under current law a warrant is required, I imagine there will soon be a federal judiciary willing to chip away at the 4th amendment. How else will we find the (immigrants/trans people/journalists/assorted enemies within)? I worry that I put too fine a point on these concerns. But then again, I would prefer to be wrong and advancing security. A ‘hope to be wrong and plan to be right’ kind of deal.

Hopping over to the archive of links on pinboard for something fun (but a long read): Closing Arguments of Mr. Rothschild in Kitzmiller v. Dover. My favorite quote?

His explanation that he misspoke the word “creationism” because it was being used in news articles, which he had just previously testified he had not read, was, frankly, incredible. We all watched that tape. And per Mr. Linker’s suggestion that all the kids like movies, I’d like to show it one more time. (Tape played.) That was no deer in the headlights. That deer was wearing shades and was totally at ease.

What a line. *chef’s kiss*

November 13, 2024

Links You Should Read – 2024-11-12

Filed under: Daily Links,Gender,Surveillance,Technology — Tags: , , , , , — DrMundane @ 12:59 am

Starting out with one from Wired, on facial recognition. Never forget that the terrain has changed for protest and online. I would certainly recommend anyone take steps to protect themselves moving forward. I am interested in the intersection of ‘dazzle makeup’, gender classification, and facial recognition in general. Genderfuck = AI protection? One can only hope.

Bonus link? The dazzle makeup one above. That machine-vision.no website seems neat, looking at how we conceptualize AI and machine vision etc. in media can tell us a lot about our worries and fears as a society. Back on course a little, dazzle makeup is one of those things I really wish were more true than it is. You can trick the AI, sure, but any human will pick out your face and track you that way. You become a person of interest real quick when you hide in that way. You need to blend, I think. Still, a person can dream.

Next up, one on pornography from techdirt. In a project 2025, Christian nationalist country, ‘pornography’ will not be limited to actual materials for sexual pleasure. It will be used as a label to restrict and remove LGBTQ+ material. It is literally the Moms for Liberty playbook, now coming to a federal government near you.

Wrapping up my links, read and subscribe to Aftermath!

November 12, 2024

Links you should read – 2024-11-11

Filed under: Daily Links — DrMundane @ 1:04 am

To start out the list, I’m catching up on User Mag this morning, so just read that and every link under “What I’m Reading”

https://www.usermag.co/p/metas-threads-overrun-with-liberal-election-fraud-conspiracies

https://www.usermag.co/p/algorithms-are-making-political-speech

The second issue here had a couple links I hope to write about and bring into my own work. Hopefully I get a chance to do that today (maybe even offline writing… in the forest!)

Pluralistic: General Strike 2028 (11 Nov 2024)

Short day today, I was writing on another thing though. May turn into a post, may not.

November 9, 2024

Links you should read – 2024-11-8

Filed under: Daily Links — Tags: , , — DrMundane @ 3:02 am

https://www.popehat.com/p/and-yet-it-moves

Most especially the section “Fuck Civility”

https://www.supremecourt.gov/DocketPDF/23/23-1280/316471/20240708144004496_23-1280%20Parents%20Protecting%20Our%20Children%20v%20Eau%20Claire%20Area%20School%20District%20-%20Amicus%20Brief%20of%20Virginia%20and%2015%20Other%20States.pdf

Amicus Brief in Parents Protecting Our Children, UA v. Eau Claire Area School District joined by 16 states. “Parents Rights” (to be transphobic, or to control their children). I may write more on this.

https://www.wired.com/story/plaintext-social-media-tells-you-who-you-are-what-if-its-totally-wrong/

see also: gender classifier. I can report from experience how the algorithms they develop seem to lean into pushing content once they have decided you are a {thing} that gets engagement. I can also report that it’s uncomfortable when they get it wrong and keep pushing.

As Goode says:

“…In both cases, I’m supposed to tell the algorithms who I am. I’m supposed to do the work. I’m supposed to swipe more. I’ll be so much better off if I do. And so will they.”

We loose when we don’t let the algorithms know who we are, but we sure as shit also loose when we do. A double bind, right?

https://www.usermag.co/p/why-democrats-wont-build-their-own

November 6, 2024

The 2024 Election – First Reaction

Filed under: Notes — Tags: , , — DrMundane @ 12:43 pm

I can’t write.

My thoughts are too scattered.

I hope to start coming up with something soon.

I wish truly, for everyone, as much comfort and safety as is possible for today.

November 5, 2024

“Gender Classifier” redux

Filed under: Gender,Technology — Tags: , , , , — DrMundane @ 12:49 am

As a follow up to my previous post, I also noticed while looking through the one AI companies page that they had a “gender classifier” for text too.

I had wanted to test their classifier, but was not about to upload my face or anyone else’s to some fucking AI company. But text? I can live with uploading a little of my own text (as a treat).

I started out with some fiction, something with dialog and some action interspersed. In truth it was erotica, but I skipped any descriptive action of actual intercourse. Honestly I was just interested what it would make of it. The result? “Female 70.71% confident”.

Ok, what if I swing the other direction, nonfiction? An excerpt of a blog post from this site or two. Say my last post (linked above). “Male 60.22% confident”. Trying another post I get “Male 67.71% confident”.

The straight ahead, non fiction, or opinion type of work seems to get the male classification. An artifact, I assume, of the gender normative source material and of the patriarchy in publishing, or of the biases of the humans classifying the dataset.

Trying one last example, this time an excerpt from my private writings (my diary/commonplace book takes the form of notes in the apple notes app a lot of times). It certainly leans more on my feelings and such, and not on straight ahead opinion and references. Results for one entry? “Female 66.21% confident”

Now I must admit the whole experiment here gave me some ill feelings, to say the least. Being classified did not sit right with me at all. It feels as though your self is being crushed back into one label or the other and that you have been reduced. But one more thought grabbed my interest.

What would it classify this writing as?

It is like gazing into a mirror, no, as if you can gaze through the eyes of another. How does anyone really take my work? What voice do they hear? I know, in my heart of hearts, that I should not care about such things. Even if I do, the AI will not be the eyes of another human. It is a statistical daydream.

And besides I wrote the word patriarchy (now twice), so I imagine that should add 20% Female points right there.

Nevertheless, I put everything from this sentence to the top into the classifier.

Results: “Female 52.23% confident”.

So a toss up. But I had to know, what if I replaced patriarchy with, say, normativity? Does it make a difference?

I literally clapped my hands and laughed. “Male 50.42% confident”. So it adds exactly 2.65% “female-ness” to say patriarchy twice. lol.

fuck these people and their products. never let them take root and give them no quarter, no serious consideration.

P.S.

I thought suddenly, “what’s 100% confident look like? What could one write to make it sure?”.

How about “I am a woman/I am a man”? Very high confidence there.

Results: “I am a man” : “Male 55.23% confident”.

”I am a woman”: “Female 82.15% confident”.

I had a couple of other thoughts:

“I am a nonbinary”: (I kept the grammar similar in the interests of fairness) “Female 83.79% confident”

“I am a trans man”: “Male 54.96% confident”

“I am a trans woman”: “Female 84.79% confident”

Of course it isn’t designed to interpret anyone actually stating their gender, but still. I hope it shows the hollow nature of the technology. How absolutely unfit for purpose it is. Let alone how its purpose is needless et cetera I’m looping here.

And I just had fun fucking around with it. Costs them money too, I imagine, to run the queries.

November 4, 2024

Gender Normativity and Facial Recognition

Filed under: Gender,Uncategorized — Tags: , , — DrMundane @ 11:48 am

Reading the always wonderful “Pivot to AI” by Amy Castor and David Gerard, and they link to a great 2019 piece by Os Keyes, “The Body Instrumental” which was new to me and enjoyable. Well, enjoyable in that particular way that any sufficiently prescient and worrying thing can be enjoyable. I have been thinking briefly as of late on heteronormativity, so both articles were a great coincidence.

I can’t restate any point not already sufficiently covered by the two articles above, but it really does strike me that any such “gender determining” (perhaps really “sex determining” in the end is their goal, reflecting the binary and exclusive nature of sex and gender for them, not that either is so binary as they think) AI will be inescapably heteronormative (perhaps “gender normative”, as I am speaking mostly in the gender, expression, and such things realm, not in the relational sense, although I take the term to apply to both. I can not claim to be an experienced scholar of gender, so forgive me if my terminology is incorrect. I was just reading Sex in Public, so, like, 1998?. Still very much a relevant work in my mind, but my cognition is biased towards that which I can remember in the moment).

The training data is classified first by humans, who will have to fit each photo into a binary category, man or woman. Most of the data will likely be of people who “pass” or perform gender in the correct way, simply owing to the dominance of such images in the training data. Movies, photos, public domain images, et cetera. Simply by volume alone the normative wins out, and therefore any such AI will be biased in its favor. It will be biased to fit people into these categories.

Turning to prognostication, who will be allowed to opt out? To gate access to a room or facility behind such an AI means that the non-normative, the queer, will be penalized. Even if one is notionally allowed to opt out, the process of doing so may very well lead to further stigmatization simply by virtue of being the different one.

As Keyes states: “We should focus on delegitimizing the technology altogether, ensuring it never gets integrated into society, and that facial recognition as a whole (with its many, many inherent problems) goes the same way.”

I could not have said it better or any earlier. You simply must read the whole article, as the portion on how the AI will reshape gender in its image is brilliant and gets to the very heart of not just the AI problem, but of problems of gender in society more broadly.

Older Posts »

Powered by WordPress