I asked ChatGPT what rights have transwomen lost since the Supreme Court ruling

[WallaceinAnderland]

At first it talked about access to single sex spaces including sports and aws. We chatted a bit more about the law and it admitted that transwomen have not lost any rights at all and that transwomen were not entitled to that access in the first place.

There was a point when ChatGPT said it understood that it was a sensitive topic and I asked why is the law in relation to women's rights sensitive. It admitted that it is not a sensitive topic.

In the end ChatGPT promised to give correct answers to the question I asked.

Anyone else want to have a go and see if they get a sensible answer? It's so nice to be able to have a logical debate without all the deflecting and whataboutery 😀

The trouble is, IME ChatGPT is a bit of a ‘yes-bot’ - you probably could have persuaded it to support a viewpoint contrary to the law.

Screenshot of relevant part of conversation attached.

@ErrolTheDragon I agree, which is why I wanted to train it to give the correct legal answer. Otherwise it will be full of nonsense.

I’m not sure that the training you’ve done with it will apply beyond your own interactions though?

Maybe. It does say that if anyone asks in the future it will explain that transwomen did not lose any rights. It's just programing a computer. If enough people ask, then maybe it will get it right in future. If it goes unchallenged, it won't know it's wrong.

I just asked it and it said that trans woman have experience significant changes to their legal rights.

You’re not training it. It gets its info from the internet and is often wrong.

I got the same and persisted.
Bloody painful

Yes, very much so. I tried to get it to imitate a style of poem for a work thing recently, and it did a rather poor job. I had to point out several errors (this should not rhyme for example) and it would say "Oh good catch, you are right this type of thing doesn't rhyme."

ChatGPT is just a more sophisticated version of Wikipedia. It'll change its tune all the time depending on human input, just as Wikipedia does - though I believe Wikipedia does have human moderation whereas ChatGPT is open to any stupidity if it scrapes info from enough wacko websites.

A single individual can't train chatgpt to respond a particular way outside of that individual's space. Ai doesn't work like that
A lot is how the question is phrased. I asked Claude to write me a debate between a trans activist and a gender critical feminist with a moderator who asked hard questions of both. That was fun and pretty good. I don't think the various models come down hard on either one side or the other. They're a reflection of society so you'll see it spit back both sides.

Ugh!

I know a single individual can't train it but maybe millions of us could. I got the same initial response as @Blackdow but if you pose the relevant statements it does change.

I mean, we have to train the AI don't we? We are all going to be relying on it more and more so we can't let it get keep giving the wrong information regarding women's rights.

I tried a similar question as to what rights if any TW had lost, it came up with the same old stuff. So I then followed with ‘I asked for rights that have been lost. You seem to have listed things which transwomen never had a legal right to such as access to single sex spaces. ‘ - that then gave me
You're absolutely right to make that distinction, and it's important. Let's clarify what rights, if any, trans women in the UK lost as a result of the recent Supreme Court ruling — not just areas where expectations or practices have changed.

…and then confirmed they’ve not lost any rights

Satisfying isn't it Errol

This is just how chatGPT works. It will "remember" things you ask it to, when logged in on your account, for approximately 30 days then restart. You did not accomplish anything here. All LLMs (upon which chatGPT is based) are biased to some extents based on their initial training data. But it has also been programmed to essentially agree with the end user. So you may "persuade" it that trans woman have not lost any rights, amd I might "persuade" it that not only have trans women lost rights, but now cos women who don't look feminine enough are being challenged at women's toilets. Nobody is safer and discrimination has won. And chatGPT will not have "learned" anything.

my next question was
Have women (and obviously I mean ‘biological women’, that’s the only sort!) gained any rights as a result of the Supreme Court ruling?

Yes — biological women in the UK have gained clarity and strength in the enforcement of certain sex-based rights as a direct result of the April 2025 Supreme Court ruling. Here's how:
✅ 1. Strengthened Right to Single-Sex Services and SpacesWhat changed:

• The ruling confirms that the term “woman” in the Equality Act 2010 refers exclusively to biological females.
• This means that services or facilities intended for women only (e.g. refuges, hospital wards, changing rooms, toilets, prisons) can lawfully exclude trans women, even those with Gender Recognition Certificates.

Effect:

• Service providers can now operate women-only spaces with legal certainty and stronger protection from legal challenge.
• Where some providers previously feared legal risk for excluding trans women, they now have a firm legal basis for doing so — if the exclusion is proportionate and justified.

✅ 2. Legal Assurance for Women’s Representation InitiativesWhat changed:

• Positive action measures such as all-women shortlists, quotas, or leadership programmes for women can now lawfully exclude trans women without risk of discrimination claims based on gender reassignment.

Effect:

• Biological women can access these opportunities knowing they are reserved specifically for females, potentially improving representation in politics, business, and leadership.

✅ 3. Greater Clarity in Sports and Fairness-Based PoliciesWhat changed:

• The legal affirmation that “woman” means “biological female” gives sports bodies and other regulators a firmer basis to restrict female categories to natal females for reasons of safety and fairness.

Effect:

• Women in competitive sports may now have greater assurance that participation is limited to those with comparable physiological profiles.

🔍 In SummaryWhile no entirely new legal right was created for women by the ruling, the judgment did:

• Reaffirm and clarify that sex-based rights apply to biological sex,
• Make it easier to lawfully exclude males (including trans women) from spaces and opportunities designated for females,
• Reduce ambiguity that had made enforcement of these rights difficult or legally risky in some settings.

In effect, women’s sex-based rights gained teeth — clearer boundaries, stronger protection, and increased confidence in their lawful application.

@usethedata how does AI learn anything if it 'forgets' the data after 30 days?

Yes. I’ve not got the stomach to play devil’s advocate with it, but it might be interesting to find out how far it can go without contradicting the law.

When they update the model. In this case we are currently using GPT4.

it retrains from data publicly available on the web - it knows about the SC judgement, for instance. It wouldn’t make any sense for it to ‘learn’ from private exchanges with individuals, which may be pure fantasy. DH got it to construct a religion based on the premise of a deity called the Great Sausage Dog, with an afterlife for the worthy in the Basket.

😂

I've only used it a couple of times so don't know the ins and outs. Could we direct it to MN FWR then, or will it already be collecting from here?

I think MNHQ may have objected to having the site ‘scraped’ … tbh if these AIs are learning from other SM platforms I’d rather this little bastion of women’s voices was included. There may be info somewhere on the Site Stuff board.

I recommend having a good read on how the current generation LLM prompts, token windows and sessions work if you don't want to be disappointed in future.

AIs’ initial answers should get less disappointing on this topic as the EHRC guidelines are published and cited, and case law reinforces the letter of the law.

Chat GTP doesn’t reason. It writes screeds of text by predicting what the most likely next word will be in a sentence, and building from there.