If I disagree with CoPilot, it's me that's wrong

[thornbury]

DH is utterly obsessed with CoPilot. He now seems paralysed to make decisions unless CoPilot validates them first. Everything must be put to CoPilot, and if I (with my 58 years of life experience and two Masters degrees) disagree with the output, it is me that's wrong. He currently has a rash, it could be shingles or a reaction to medication, but CoPilot says it's due to sun exposure/sun lotion allergy, therefore my views are inaccurate.

He will ask me something and then check my response with CoPilot, or ask me something and tell me that CoPilot has a different view. He presents CoPilot output as valid opinion, without admitting that the expert he is supposedly quoting is actually CoPilot. It's driving me nuts.

I’m a Dr and peoples bloody AI diagnoses are becoming such a problem. No one can tell you their actual symptoms these days without supplying what AI says they mean. People are confidently diagnosing themselves with all manner of very rare diseases and demanding reams of tests and treatments on that basis. Before anyone links to that Welsh lady who did have a rare genetic disorder diagnosed by AI that was news because it’s the exception to the rule.
Common things are common.
AI is massive fuel to the flames of health anxiety and medicalisation of normal life.
Tell him to get a proper human being to look at the rash and ask the right questions to put it in a medical context instead.

Well at least it's keeping him away from actual real live women.

Tell him to ask CoPilot how reliable it is

On a side note, CoPilot failed me today, apologised for being unable to properly complete the task.
I had two diagrams showing flow lines throughout network systems. I wanted it to combine the common elements of the two diagrams and to colour the flow lines of number one in red / amber and the flow lines of number two in green / blue
It apologised that it could not perform the full task and offered to do what it was able to do.

I accepted that option with the intention of then tweaking prompts to work some way to get what I was after

What it gave me was one combined image, with two separate diagrams, it had redrawn both as almost identical to the original but jumbled half the text annotations into gibberish, and its solution to not being able to colour the flow lines was to draw a red line straight through the top one and a green line straight through the bottom one

Ask him to ask Copilot who is mostly likely to be right - an intelligent educated human being or a non-thinking machine?

@Shrinkhole he says he's found it very helpful in explaining medical terms, interpreting reports, etc. His oncologist is the best in the hospital and he lets CoPilot explain things he should be asking her.

Of all the AIs to defer to he chose Copilot? I’d be pissed too, copilot is rubbish 😂

Well it’s fair enough to ask it ‘what is stage 4 metastatic disease’ or something like that but not ‘is this chemotherapy regime right for me?’ and then imagine that it knows better than an experienced oncologist. I have patients asking for medications that don’t exist in the UK because their AI info is US based or it just agrees with what they want to hear and then they are reinforced in what might be a wrong opinion. Eg if you input a ton of ASD symptoms in it’ll say you have ASD but it is not assessing the persistence, pervasiveness, disruption of those symptoms or any evidence to the contrary that you will not be adding as you don’t want to hear it. Then the person has a real assessment, is not diagnosed and the assessors are wrong cos ChatGP says I am autistic.

This probably isn't much consolation OP, but the two links below appear to confirm what you're saying. From my point of view, I'd be hesitant to enter personal information because of security and privacy concerns.

One in seven Brits swapped their GP for ChatGPT, study finds.
Patients are using chatbots for medical advice.
Site: www.theregister.com/ai-ml/2026/05/16/one-in-seven-brits-swapped-their-gp-for-chatgpt-study-finds/5240514

Ask Don't Tell: Reducing Sycophancy in Large Language Models
Our new paper finds that the phrasing of user inputs affects levels of model sycophancy.
"Consumers are increasingly turning to Large language models (LLMs) for advice on consequential life decisions, from health and relationships to career choices and wellbeing. They are also common tools for information-seeking, on political topics, for example."
Site: www.aisi.gov.uk/blog/ask-dont-tell-reducing-sycophancy-in-large-language-models-2

I use AI in a personal sense but hold a neutral position on the subject as I can see both sides of the coin. If possible, get your husband to read the following book. It's opened my eyes and worth a read however you see AI. I have no financial relationship with the authors or publisher and receive no compensation or affiliate payments for recommending this book.

Book:
The AI Con: How to Fight Big Tech's Hype and Create the Future We Want
written by co-authors Emily M. Bender and Alex Hanna
Published by Vintage Publishing under The Bodley Head imprint.

The book is a sharp, witty take-down of the modern artificial intelligence industry. Linguist Emily M. Bender and sociologist Alex Hanna argue that "AI" is a misleading marketing term used by massive corporations to obscure ordinary automation, concentrate power, and drive corporate profits.

Hi, Copilot. In 400 words or less, give me the pros & cons of preferring Copilot's advice on everyday life questions over the opinion of one's spouse.

Pros of preferring Copilot

• Speed and convenience: Copilot gives instant, researched answers for routine choices (recipes, schedules, product specs).
• Consistency and breadth: It can summarize many sources and offer neutral comparisons without fatigue.
• Scalability and privacy for low‑stakes queries: Useful when you want a quick second opinion without involving your partner.
• Helpful for information gaps: Good when your spouse lacks knowledge on a specific topic.

Cons of preferring Copilot

• Erodes human connection: Relying on AI for personal support can weaken interpersonal bonds and reduce opportunities for mutual problem‑solving.
• Limited emotional nuance: AI’s empathy is simulated and may not challenge you or reflect shared values the way a spouse can.
• Context and values mismatch: AI lacks access to your private history, household norms, and tacit knowledge that shape good personal advice.
• Risk of over‑reliance and isolation: Excessive outsourcing of everyday decisions to AI can reduce social skills and increase loneliness.

Practical guidance and risks

• Use Copilot for low‑stakes, factual, or time‑sensitive queries (e.g., “best route,” “how to fix a leaky tap,” quick recipes).
• Use your spouse for values, feelings, and decisions that affect both of you.
• Combine both: Ask Copilot for options, then discuss the shortlist with your spouse so decisions reflect both evidence and shared values.
• Watch for signs of harm: If you notice less conversation, more unilateral decisions, or emotional distance, rebalance toward human interaction.

Bottom line: Copilot is a powerful tool for efficiency and information; your spouse is irreplaceable for empathy, context, and relationship health. Use both deliberately rather than substituting one for the other. American Marketing Association Harvard Business Review

I find it creepy that the AI says it values human empathy and relational bonding, but assumes its factual information to be superior! It sees itself filling human 'information gaps', with no clue that a human might have better information than the AI.

Wonder where it thinks all the information it 'knows' came from?

Everything or medical questions? Set a boundary

What happens if you straight out ask Copilot how reliable it is at giving factually correct answers?

That sounds awful. Out of nosiness, what sort of emails were they? I would be furious if someone was feeding sensitive work information into a shadowy AI machine - the information doesn’t just float into space, it gets stored somewhere.

"I want to be married to someone who thinks for himself, not an AI"

At its harshest

AI cannot diagnose a rash! I have actually tried it. It got it completely wrong every time.

Also it’s worrying that he’s relying on it so much. He’s giving his critical thinking away to a machine.