Chat GBT and Ai support for Hypothyroid

[Danielj1]

I dialled into the new Google artificial intelligence software today to ask whether it new how to address hypothyroid symptoms.

I was somewhat shocked to be told that part of the solution was to increase iodine consumption

I wrote back explained the error and asked the robot to reconsider the advice

Very quickly there was an apology as it had searched the data bank and identified that this was a wrong recommendation

I redid the search and to be fair the robot changed the answer to remove the iodine reference

So try to do my big as they say - but moral of the story is the robot is only as intelligent as the feedback provided by the human!

Be very careful to not assume the robot suddenly had new powers of insight …

[Realtiger]

Wrong information in, wrong information out! Great story - thanks for sharing!

[BootsOn in reply to Realtiger]

Back in my IT days, we used to call it GIGO. Garbage In, Garbage Out

[Stourie in reply to BootsOn]

😂😂😂

[Realtiger in reply to BootsOn]

That’s exactly the expression I was looking for - it escaped me - thank you for the reminder! 😂

[helvella]

The "clever" bit is largely that it takes input in ordinary language, and replies in ordinary language.

What it says is often woeful.

Further, your correction today might stand for a few seconds, hours, days. But next release might wipe it all out, leaving it then to unequivocally recommend iodine again.

There is a real possibility that some version will, at some time, come up with associations and links of real value. But how will we, or it, identify that real value? And how will it be "published" and spread to practitioners?

Its getting towards twenty years since some papers started to say that taking levothyroxine at bed-time was a reasonable option. Yet the majority of literature (guidelines for treatment, product Patient Information Leaflets, etc.) simply assumes taking in the morning without any suggestion of it being a choice.

How will anything one of these AI systems comes up with actually get used?

(And I am quite sure some people had been taking at whatever time they thought best for them for decades before. With those who need help doing so at the time that help was available regardless any recommendations.)

Am also completely convinced it has a built-in "if client disagrees, apologise and feign learning" mechanism.

Without the ability to ask it to detail its reasoning, any "recommendations" should be treated as no better than being of potential interest but needing to be thoroughly worked through.

[FancyPants54]

Artificial Intelligence is the wrong term for AI really. It isn't intelligent. It is not sentient. It can only respond with the words and phrases it has already been given to work with. I would not recommend AI like Chat GPT for medical advice! It would be different if the AI was specifically programmed for some medical field or other. But it can never reason or consider. Only regurgitate.

[ThyroidLadyLondon]

can you post exactly what the response was?

[helvella]

And do remember, a major part of the project is to advertise effectively.

They could bias results to promote particular things - and, unless they highlight them, it would be very difficult to be sure whether something was promotional or "genuine".

Imagine, as already happens in much USA discussion, they mention Synthroid and Cytomel repeatedly. This could be simply because they replicate what is already being done - often by patients - using brand names rather than recommended international names. Or it could be AbbVie and Pfizer paying to promote their major brands.

If you notice, and raise the issue, the system might then effectively make a note against you as a user not to do so in future - just talk about levothyroxine and liothyronine. And you relax. But the system might continue promoting Synthroid and Cytomel by name to everyone else. And you'd be unlikely ever to find out.

Applying any form of control to such systems is an extremely difficult thing to achieve.

[Danielj1]

very good point helvella. Sorry FP, do not have the wording back saved but you can login and ask questions like is there a cure for hashimoto etc