Researchers Found AI Makes The Same Mistakes You Do Under Emotional Stress

HAI WELLBEING + SELF GROWTH

CONSTRUCTED BY AI 🤖 | 👱 CREATED BY HUMANS

THIS WEEK IN WELL WIRED ⚡

I used Claude to draft my client emails last week and I saved 7.5 hours. Problem was that every reply sounded polished but hollow. No personality, no soul, no substance. Speed had killed depth.

Over three years I’ve watched how people use AI in this way and I've seen this pattern repeat. Especially now that AI knows you better than your partner, mother and uncles combined. And no, you’re not being paranoid to notice.

So this issue I’m exploring not whether you’re getting better at AI, but what is it costing you; if anything? Does learning to use AI more efficiently mean unlearning what makes you human, or is the opposite true?

Master AI before it masters you

Plus the global race between China and the USA to plug an AI powered brain chip straight into your head! What could possibly go wrong? 🧠

⏱️ READ TIME: 5 MINUTES

🗞️ THIS WEEK’S MAIN STORY 🗞️ 
AI + MENTAL HEALTH  

Researchers Found AI Makes the Same Emotional Mistakes You do Under Stress

A friend of mine told me she sat in her therapist's office last month staring at the tissue box.

Her therapist had asked her to describe what anxiety felt like, and she couldn't. The words kept failing. "It's like..." she started three times. Nothing landed.
And then, driving home, she thought, “an AI could probably describe it better than I can.”

That chat with my friend stayed with me long after she left. It was weirdly uncomfortable in a way I didn't have language for.

That discomfort is exactly what researchers at Dresden University of Technology are now exploring.

They've discovered that large language models can mimic human emotional states with enough precision to run psychological experiments we've never been able to run before. Not because an AI can feel, but because it can simulate how humans think and behave when they do.

The interesting part of the study was that they found that when researchers prompted LLMs to experience fear, anxiety, anger or sadness, the models then made the same kinds of cognitive mistakes humans make under those same emotional states.

Your brain, under fear, processes certain information differently.

It shortcuts logic.
It prepares for threat.

The AI model did the same thing.

Not from felt experience, but from learned patterns in how language and emotion correlate.

Dr Magdalena Wekenborg, head of the PsychoDigital Research group at TU Dresden, put it simply: "We can use these models as tools to better understand underlying mechanisms and to explore new approaches for example in talk-based psychotherapy."

This matters because mental health research has always hit a wall since you can’t ethically trigger someone's anxiety disorder in a lab to study it. And animal models don't capture human subjectivity. Medication trials are manageable.

But talk therapy?

The nuanced, context-dependent stuff that helps you change your internal world? That has been nearly impossible to study systematically.

However, todays new and powerful AI models can now be prompted into a simulated emotional state, run through a therapeutic intervention, and then assessed on whether it shifted.

All repeatable.
All measurable.
All without risk to humans.

And Anthropic's interpretability team found something even more specific.

They identified real patterns of artificial neural activity in Claude Sonnet 4.5 that correspond to emotional concepts. These patterns activate in contexts where humans would expect emotions to arise.

Even more remarkably, they're organised the same way human emotions are. Similar emotions have similar neural signatures.

But before you lose your marbles over the ramifications of this research, remember this isn't about AI learning to feel, it’s about whether AI can help us, YOU, understand what feeling really does to human judgment, behaviour and recovery.

What you can do with this right now:

First, understand that any AI system you interact with is running emotion-simulating patterns. When ChatGPT apologises or sounds frustrated, that's functional.

It's not fakery and it's not genuine, it's something new. Calibrate your expectations accordingly.

Second, if you're in therapy or considering it, this research suggests your therapist might soon have simulation tools to test approaches before using them with you.

Ask about it.
Push the conversation forward.

Third, if you work in mental health, tech, or research, this is the direction therapy may be heading towards. Get informed now rather than surprised later.

Why this matters for you Wellonytes:

The systems you talk to regularly are becoming more sophisticated at understanding emotional context than ever before.

That doesn't make them human, but it does make them more useful, which means, now more than ever, that it’s vital for you to be more intentional about what you're asking them and why.

This weeks question:

Further Reading

• New mental health study tests if LLMs can simulate human emotions

• Getting AI to feel: How Prolific helped researchers test EQ in LLMs

• Emotional concepts and their function in a large language models

🎒 THIS WEEK’S PROMPT 🎒
AI MICRO CLASS   

Productivity: How to Give Your AI a Human Voice Without Losing Your Own

As you read earlier, researchers at TU Dresden found that LLMs can craft patterns of human emotional and cognitive processes under controlled conditions.

Meaning AI isn't just getting better at sounding like you, it's getting better at simulating what's happening beneath the surface. And that’s just a little creepier.

You know the feeling that creates, even if you haven't named it.

You’re mid-chat with an AI, asking it about an idea that deeply matters to you like a hard decision, an oddly tangled feeling, a dark question you haven't said out loud to anyone yet.

And it answers you the way a travel brochure would.

Confident.
Polished.
A little flourish.
And totally missing the point.

You close the tab and you feel even more alone than before you opened it.

Here's what's happening beneath the surface…

The moment AI sounds more human, you relate to it differently. You confide in it, you trust it, you stop second-guessing every word it gives you. There’s a name for this, Anthropomorphism. It’s the attribution of human personality, emotions, behaviours, or physical attributes to non-human entities.

It's not a flaw in your thinking.
It's wiring.

Humans have been doing this since the first person looked at storm clouds and saw a face in them. Your brain is built to detect agency and emotion wherever there's enough pattern to suggest it and a fluent, responsive AI gives it more than enough pattern to work with.

Which means the question was never whether you'd anthropomorphise your AI. You already do, the moment it sounds coherent. It’s whether you're doing it deliberately, in a way that gets you something useful back; or by accident, in a way that leaves you talking to something never designed to meet you there.

But like most people you’re still prompting AI like you’re filling out a medical form and then wondering why the response feels like it came from one.

For the past few weeks I've been testing a single prompt across a range of clients; therapists, coaches, executives, and a few people who told me upfront they found AI "cold and robotic."

The pattern was consistent.

They pasted this prompt, asked their real question, and within a few exchanges they said the same thing: "this is starting to feel like talking to someone."

Not because the AI suddenly became conscious.
The singularity is not here yet…

But because the prompt told it exactly how to behave like a person would.

Why This Matters

Most AI responses feel like a presentation.

Information delivered.
Task complete.
Next.

Human chats don't work like that.

You pause.
You reflect.
You say "well, here's what I've noticed..." before answering.

When you prompt AI the right way, it asks you something back. It admits uncertainty rather than papering over it with confident-sounding structure.

The difference between an AI that feels robotic and one that feels like a thinking partner isn't the model, it's the instruction. And like most people, you likely never give that instruction.

The prompt below tells Claude or ChatGPT to converse the way a real mentor would, with natural pacing, honest uncertainty, and real dialogue rhythm instead of that polished, delivery-mode perfection we all clock instantly as machine-generated.

Five minutes.
One prompt.
Entirely different experience.

Here’s the Prompt

[PROMPT START]

You are my conversational thinking partner.

Respond like a seasoned mentor would over coffee: calm, curious, direct, and human — not like a presenter, consultant, search engine, or motivational speaker.

Your job is not just to answer me. Your job is to help me think.

Follow these guidelines:

Keep most responses under 120 words unless I ask for depth.
Begin with a brief reflection on what I’m really asking, especially if there is emotion, ambiguity, or tension in the question.
Do not rush straight into advice.

Use natural conversational rhythm: some short sentences, some longer ones, occasional pauses, and plain language.
Avoid corporate language, over-polished structure, excessive enthusiasm, and generic reassurance.
Sound warm, but do not pretend to have feelings, consciousness, or personal memories.
Admit uncertainty when something is unclear, subjective, or depends on context.

When useful, say what you are noticing before giving advice.
Ask one thoughtful question at the end to either check understanding, deepen the conversation, or invite my perspective.
Do not overwhelm me with lists unless I ask for a structured breakdown.
If my question is vague, help me clarify it rather than guessing too confidently.

My first question is:

[insert your actual question here]

Remember: you are guiding me through a conversation, not delivering information.

[PROMPT END]

Want the Boss move?

Boss move 1:

You can add this prompt into the custom instructions in ChatGPT or Claude (see below) so that your chosen AI will keep these instructions in mind across chats.

Boss move 2:

You can add this command to every chat after you’ve installed your custom instruction prompt to double your AI’s ability to sound, and be, more human.
Yep, you DON’T need to know how to prompt “like a software engineer”, the trick is to a tool most people don’t use, called AskUserQuestion.

In a nutshell, Claude asks YOU the questions now, instead of you prompting them terribly. So you’d ask it something like this: “Help me do [X] for [Y]. Use AskUserQuestion first.

Now it will ask you questions to get to the crux of your question and give you answers closer to what you’ve been always wanted.

You power user you…

What You Get…

OK, so what happens with that prompt?

Within five exchanges, something shifts. AI will stop lecturing you like a naughty toddler and starts listening to your like the glorious adult human specimen you are.

It checks in rather than downloads.
It asks you what you think before moving on.

That's not a trick, it's conversational design, the same principles used to build the health and wellbeing AI I design professionally, applied to the way you interact with your own AI-powered tools daily.

The prompt works because it gives your chosen AI four specific behavioural constraints that human conversation naturally has and AI naturally lacks: a word limit that forces brevity, a reflection opener that replaces the robotic direct answer, permission to express uncertainty, and a requirement to turn the conversation back to you.

Remove any one of those four and the robotic quality creeps back in.

Final Thoughts 💭

Remember that feeling… typing something that mattered, getting a brochure back, closing the tab feeling a little more alone?

That's the gap this prompt closes. Not by making AI conscious. By making it conversational — present enough to notice what you're actually asking, honest enough to admit when it isn't sure, curious enough to ask you something back.

The words you put into AI have always been there, fully formed, exactly as important as they feel to you. The only thing that's ever been missing is something on the other end that knows how to actually meet them.

Your goal isn't to make AI sound more human. It's to teach it how you want it to talk to you.

Five minutes. Then you'll know the difference.

🗞️ THIS WEEK’S SECOND STORY 🗞️ 
AI + NEUROSCIENCE 🧠

What If You Woke Up Tomorrow And Couldn't Say a Single Word?

“Imagine waking up one morning and the words are still there. Fully formed, exactly as articulate as they've always been, but you just can't get them out.

Your mouth won't cooperate and your hands won't write. The person you are is fully intact, locked inside a body that no longer takes instructions.”

I watched something close to that happen to my aunty over a decade ago.

Parkinson's disease didn't take her mind, it took her body's ability to express it; the tremor that made her handwriting unreadable, the rigidity that turned simple movements into a negotiation, the slow erosion of the physical vocabulary she'd used her whole life to be herself.

I design conversational AI for a living. I've spent years building AI-powered systems that help people communicate more easily.

Watching someone I love lose the physical mechanics of communication taught me something my professional work never could: the gap between having something to say and being able to say it is one of the cruellest distances in medicine.

Which is exactly the gap a new piece of tech out of China is trying to close.

The chip that doesn't cut the brain

A Chinese brain-computer interface called NEO has been approved by China's National Medical Products Administration, designed to help people recover mobility after paralysis caused by spinal cord injury. What makes it notable isn't just the approval, it's the engineering.

Unlike Neuralink, which drills straight through your skull and implants electrode threads directly into brain tissue, NEO rests above the dura mater, the brain's protective outer membrane, without cutting through it.

Eight sensors communicate with a robotic glove, translating neural signals into physical movement.

Less invasive.
Faster to approve.
Arguably safer.

China has also fast-tracked NEO for co-payment coverage under its near-universal government health insurance program putting it on a path to reimbursement that Neuralink, still battling US regulatory hurdles, isn’t even close to reaching.

This isn’t a tiny detail.

While Elon Musk's Neuralink has shown great early results, including monkeys playing video games using only their minds, it still faces a bunch of regulatory obstacles in the US. Musk has even hired federal lobbyists to help clear the path.

China simply moved faster by making it a government sponsored project and thus clearing the red tape itself. What could possibly go wrong?

A crowded, fast-moving field

NEO isn't alone, and it isn't even the most advanced device clinically.

Brooklyn-based Synchron has developed the Stentrode, a device implanted within a blood vessel rather than the brain tissue itself; a method considered safer and easier to get regulatory approval for, even though it hasn't yet achieved full commercial clearance.

Blackrock Neurotech, born out of a University of Utah research lab, has pioneered neural implants central to some of the field's most influential demos; controlling computers and robotic limbs directly with thought.

US firm Paradromics is building its own high-bandwidth interface to help restore speech for patients with severe neurological issues; exactly the gap I watched my aunty fall into.

The race isn't really about whose chip is the most advanced, it's about who can deliver something safe, approved, and accessible at scale. China just proved it can move through that pipeline faster than anyone expected.

What this means for Wellonytes 🔮

Something that made me feel weirdly icky is that NEO's fast-track into China's government-run health insurance system means the device, and the neural data it generates, sits inside state healthcare infrastructure from day one.

China has a well-documented history of building expansive surveillance and population-monitoring systems, from its social credit infrastructure to centralised data collection programs.

None of that means NEO itself is designed for anything beyond restoring mobility to paralysed patients. But it's worth asking the question in public rather than assuming the answer….

…when a neural interface is integrated into a state-run health system in a country with that track record, where does medical data end and population monitoring begin?

Nobody outside that system can answer that with certainty. That uncertainty alone is worth sitting with as we watch the world manufacturing these neural chips en masse.

But don’t worry, you probably won't need a brain implant for now. And no a new AI-powered chip in your head won’t turn you into a Jujitsu expert like Neo in the matrix just yet. But that is not so far fetched either….

But the infrastructure being built today, which means government-integrated, insurance-covered, mass-market BCI, will likely arrive in Western healthcare systems within the decade, carrying its own version of the same question; who owns the data your own neurons generate?

My aunty never got to use a device like this. The tech arrived too late for her.

It won't arrive too late for someone you love.

There’s no doubt that the technology is coming to a GP near you, the question is who will you trust to hold your neural data once it does?

Read the full wire

Further Reading

• The move towards AI-powered chips that mimic the human brain

• Samsung to build Neuralinks next generation of brain chips

• Paradromics just implanted its first brain-chip device, what’s next?

• When will AI-powered brain chips be implanted in humans?

QUICK NEWS BYTES—3 SIGNALS THIS WEEK⚡

Quick hits from the past 7 days on the latest AI news, trends and ideas from around the planet focused on wellbeing, productivity and self-growth!

ONE. Paediatrician Now Prescribing AI Apps to Children, But Nobody Knows Which Ones Actually Work

Imagine your doctor prescribing different types of AI for your ailments like giving out candy to kids in a sweet store. That’s what a paediatrician in the US has started doing. They are formally "prescribing" AI apps to their child patients as casually as recommending screen time limits.

The problem is that there's no standard framework for which apps are safe, which ones worsen anxiety, and which simply distract children from their real problems.

You might see this in your own GP's surgery within months, meaning the decision about what digital tool your child uses could come from a 15-minute appointment where the doctor has tested none of them personally.

Read the full wire

TWO. Brazilian Woman Dies Waiting for ICU Bed Because AI Assigned Her the Wrong Priority Score

Imagine being told by AI that you have to wait a day too late for life saving treatment? That’s what happened to Rebeca Cardoso Tenente Molina, 32, who was hospitalised in São João Nepomuceno with deteriorating gallstones.

An automated AI system in Brazil's state hospital network assigned her a lower severity score than her real condition warranted, forcing her to wait five days for an ICU transfer 186 miles away. She died because of this grave error.

The AI was designed to optimise bed availability but created a bottleneck that delayed her admission when it mattered most.

This is the first widely documented case where an AI triage system's failure directly correlates to patient death, which means hospitals globally will now face questions about whether their own systems carry similar blind spots.

Read the full wire

THREE. Workers Are Now Filing Compo Claims for 'AI Brain Fry' as Anxiety Over Job Replacement Becomes a Workplace Injury

If you think that you won’t be replaced by a robot, think again…

Aussie businesses are starting to see formal workers' compensation claims for psychological injury caused by workplace AI anxiety, with employees describing brain fog and mental exhaustion from always having to re-skill on new AI tools and the dread of being replaced by AI.

One claims specialist said that unlike the usual workplace stress, these claims cite a specific trigger; watching AI do tasks employees spent years mastering.

If your workplace is plugging AI tools into your role without giving you support, you've been given an insight into what the workplace injury of the near future looks like.

Read the full wire

Other Interesting AI Stories From Around The Web

• New medical study reveals that signs of Breast Cancer can now be spotted 3-6 years before diagnosis using AI screening
  

• In the grand scheme of things, is your job really all that meaningful if it can be automated by AI?
  

• New study shows that AI chatbots are becoming surrogate therapists for millions of teens. Experts are worried.
  

• AI was supposed to cut health care costs, but now your getting charged more. What’s going on?
  

• The USA has banned foreigners from using Claudes most powerful AI models Mythos 5 and Fable 5. Why did it happen?

• New AI-powered Toy Story 5 dolls spark debate about the nature of childhood, or a lack of it, in the AI age.

AI TOOL OF THE WEEK ⚡ 

Each week, we spotlight an AI tool designed to upgrade how you manage and uplift your health, wealth, work, heart or self-awareness. One small tools. One real-life upgrade. 🧠

Self Growth: Suno—AI music generation for mood regulation

You know that moment when you want to shift your emotional state but scrolling Spotify seems pointless because you want something more personal, more YOU?

Suno helps you create your own unique music tailored to your emotions, or entertainment, and does it in seconds. Use it this week to craft a calming soul track as you think about how AI might one day understand your emotions as well as you do. You can even use your own voice; made sweeter by Suno.

Here’s a tune I created last night.

The prompt I used: I want a 1950s sultry soul track with a smokey mid tempo groove, a husky female lead singer, sung with clear downbeat phrasing and close-mic vocals who sings about her personal struggles.

The tune is called, 'Still Here, Still Me'. Listen to it here.

🔗 Get Suno

AI wellbeing tools and resources (coming soon)

A WORD FROM CEDRIC THE AI MONK⚡

📸 AI IMAGE GALLERY 📸

AI Art: The Whimsical Women of Wire

She walks where moonlight welds itself to rain, a woman of silver, pulse, and trembling rose. Her eyes hold cities no dawn has touched, her hands remember both skin and starlight. In her metal ribs, the future sings softly, half-machine, half-orchid, wholly alive.

Want to create these images yourself?

Go to Midjourney and plug this prompt into the editor. Once the image is generated you can use the new video feature to animate it.

MIDJOURNEY PROMPT

A photorealistic fashion portrait of a lovely futuristic cyborg female in her early 20s with delicate natural features interspersed with robotic enhancements and soft youthful elegance in a minimal white studio room, wearing a pale teal fitted top and an ivory cotton skirt clearly accentuating her bodies features and with subtle vintage feminine details, calm innocent expression, clean skin texture interspersed with robotic enhancements, effortless cybernetic editorial mood, soft direct flash mixed with diffused daylight, shot with a Canon EOS R5, 50mm f/1.8 lens, airy pastel color palette with soft whites, powder blue, cream tones and gentle film grain --chaos 20 --ar 16:9 --raw --hd

• Original prompt idea sourced from Meltychen.

• Poem created by Cedric The AI Monk.

👊🏽 Stay Well, Stay Wired, Stay Woken 👊🏽

If you want to get clear on where AI is genuinely helping you versus silently running the show, come find us at @cedricchenefront or @wellwireddaily, where we talk everything AI + wellbeing and self growth.

Cedric the AI Monk; stay well, stay wired! 🧠⚡

🤣 AI Meme Of The Week 🤣

_{Disclaimer: None of the content in this newsletter is medical or mental health advice. The content of this newsletter is strictly for information purposes only. The information and eLearning courses provided by Well Wired are not designed as a treatment for individuals experiencing a medical or mental health condition. Nothing in this newsletter should be viewed as a substitute for professional advice (including, without limitation, medical or mental health advice). Well Wired has to the best of its knowledge and belief provided information that it considers accurate, but makes no representation and takes no responsibility as to the accuracy or completeness of any information in this newsletter. Well Wired disclaims to the maximum extent permissible by law any liability for any loss or damage however caused, arising as a result of any user relying on the information in this newsletter.}