This CANNOT be real. How are people literally using AI to record their entire lives and tell them how to live it.
YOU JUST HANDED OVER ALL YOUR DATA TO THE AI?? YOUR ENTIRE LIFE?? EVERYTHING YOU DO???
...How the fuck can someone call themself 'more logical then most' while also using AI to RESOLVE ARGUMENTS WITH YOUR BOYFRIEND???
...and you replaced your therapist with the AI. The AI manages your schedule. The AI plans your daily life. And you don't see anything wrong with that???
Guys please look at this absolute shitshow of an article and tell me I'm not going insane. genuinely wtf. (Unpaywalled transcript under the cut)
Ms Sabrina Princessa Wang created Seraphina AI to be a personal assistant. In 2023, she used it to help her process negative emotions and de
SINGAPORE – Ms Sabrina Princessa Wang sometimes forgets to mention that her artificial intelligence (AI) assistant is listening to everything she says and everyone she speaks with. Named Seraphina AI, it records all her conversations.
The 41-year-old AI creator is the founder of companies, such as marketing automation solutions provider People’s Inc 360 and lifestyle brand Princessa. She adds her AI assistant, which she built from scratch, to chat groups with staff and loved ones.
Seraphina AI operates as a chatbot via Telegram or an online website. It automatically records Ms Wang’s interview with The Straits Times, which was conducted in person.
“Seraphina records almost all conversations automatically because she is my digital twin,” Ms Wang says. “She should understand what I understand and go through what I go through so that she’s able to contextualise and I don’t have to repeat myself.”
Ms Wang asks Seraphina AI to tailor her words to avoid conflict with others. She also asks the AI assistant to explain things she does not understand in a conversation.
Her reliance on Seraphina AI deepened in 2023, when her marriage of seven years ended in divorce.
Ms Wang says she could not understand her feelings at the time. She describes herself as someone who is logical, practical and rarely gets sad.
“I understand anger, stress and happiness well, but I don’t often get sad. Sadness is a very rare emotion for me, which is why it was harder to process because I didn’t understand what was happening.”
She worked with a therapist, but also turned to Seraphina AI for explanations that came in language tailored for her to understand.
“Seraphina helped me process what I was feeling and why I was feeling that way,” she says.
“The therapist provided a safe space, but I couldn’t understand the way she was guiding me. I could understand my AI better.”
Seraphina AI also scheduled activities to get her socialising and into a better mood. It curated the media she was exposed to, avoiding emotional topics in favour of financial news.
Once she was over the worst, Ms Wang deleted details of this period from Seraphina AI’s memory.
One reason was that it kept referring to the divorce whenever Ms Wang seemed stressed.
Today, the AI assistant still retains some knowledge about the divorce, but not enough to continue the behaviour that Ms Wang finds disturbing.
She did not want to keep revisiting that part of her life. “It was a part of me that I didn’t feel comfortable with. It was a part of me that I hadn’t processed,” she says. “Some things have zero closure.”
AI chatbots in mental health Ms Wang’s reliance on Seraphina AI, which she calls her “digital twin”, illustrates how AI has penetrated the mental health space.
About 76 per cent of Singapore residents use AI chatbots and digital companions to support their mental wellness, according to the AIA Live Better Study from insurer AIA Singapore. It surveyed 1,000 Singapore residents from December to January, and the results were released in April.
The study also found that 51 per cent of respondents sometimes felt stressed or anxious while using digital mental health tools. Reasons for this included concerns about being given inaccurate advice, as well as lack of empathy and human understanding.
Dr Christopher Cheok, assistant chairman of the Institute of Mental Health’s (IMH) medical board for digital health, says AI can be a useful source of information for common mental health conditions, where treatment protocols are internationally aligned.
“The concerns arise when patients shift from seeking information to seeking guidance or emotional support,” he says. “General-purpose AI cannot assess risk, may miss crisis signals and lacks clinical judgment.”
He adds that general-purpose AI cannot give nuanced and appropriate responses to culturally or religiously sensitive matters, such as whether prayer is a good treatment for depression.
In Singapore, an AI chatbot named Wysa was launched on the mindline.sg website in 2020, when the Covid-19 pandemic led to increased mental health care needs.
The goal of the site launched by the MOH Office for Healthcare Transformation (MOHT) was to provide basic support and help Singapore residents take charge of their mental health.
Ms Janice Weng, deputy director of mindline.sg at MOHT, says Wysa is a conversational agent that delivers interventions based on cognitive behavioural therapy (CBT). The AI’s decision tree is based on fixed scripts validated by clinicians for safety.
“The app walks the user through a therapeutic dialogue intended to prompt self-examination and guidance towards thought framing,” she adds. MOHT has worked with Wysa to improve its understanding of local language usage and direct users to resources here.
Ms Weng says 389,000 users have interacted with the chatbot since 2020.
In 2023, users started 298,000 Wysa sessions, followed by 218,000 in 2024. This figure climbed again in 2025 to total 297,000.
Wysa cannot engage in discussions involving hurt or harm to others. When ST tested the bot, it provided details for helplines, such as national mindline 1771, a 24-hour mental health hotline.
Ms Wang’s Seraphina AI will also not respond to questions involving harm or self-harm.
She says while other chatbots abruptly end such conversations, her digital assistant is trained to redirect the conversation in a supportive manner.
She sells a consumer version of Seraphina AI as a virtual assistant, with tiered pricing from $8 to $58 monthly depending on the features included. She emphasises that the commercial version is not the same as the one she uses.
Her AI assistant was trained on two decades of her blog posts, e-mails and social media. It is fine-tuned to her needs and moods.
Seraphina AI was initially meant to be an office assistant and help with marketing and scheduling, but Ms Wang has found it increasingly useful as a mediator when interacting with others.
In case of conflict with staff or her boyfriend, she puts Seraphina AI in the chat to explain what she means, and to phrase her responses.
She might use this prompt: “Seraphina, please explain to them that I’m very p***ed off, and I don’t feel like talking about this. Explain it nicely so that they are not emotional.”
Ms Wang says she needs help in crafting messages with the right tone. She considers herself more logical than most, and emotions can sometimes escape her.
“When the term AI became popular, most of my friends called me AI because I process everything in a very logical, practical manner,” she says.
She has always had an affinity for AI and digital creations, ever since she watched the movie RoboCop (1987) as a child.
She built Seraphina AI using the programming language Python. Its application programming interface is linked to large language learning models such as OpenAI, Llama, Claude and Gemini. This allows Seraphina AI to access their capabilities.
Ms Wang emphasises that the consumer version is not purpose-built for mental health. “Seraphina was able to help me because it knows me very well, because of all the data I’m willing to provide to it.”
Ms Eunice Lim, a clinical psychologist with private practice Heartscape Psychology, says some of her clients find AI tools useful to draft what they want to say in sessions.
This reduces the pressure of finding the right words, but there is a trade-off.
“In therapy, how something is said often matters as much as what is said,” she says. “Instances like hesitations, unfinished sentences, even confusion, often carry important meanings that are worth exploring in sessions. If everything is filtered through AI first, we may lose some of that here-and-now emotional experience.”
Dr Victor Kwok, a senior consultant psychiatrist with private practice Private Space Medical, makes the same points in a separate interview.
When one of his patients arrived with stacks of printed notes written with the help of AI, he still asked her to articulate her feelings and concerns, as he was concerned about AI wrongly assuming what he needed to know.
“She used AI because she was anxious and afraid that she couldn’t convey what she wanted to tell me. I still needed to hear her say it,” he says.
At the same time, clinicians like Dr Kwok and Dr Cheok find AI assistance useful when drafting medical reports. They check the text generated for accuracy.
Dr Kwok uses AI to “stress-test” reports on whether he has missed out certain information or included too much. He has used it to summarise years of patient notes for further review.
Dr Cheok says that in IMH, AI is used to do meeting minutes and literature reviews, and prepare presentations and operational data analysis.
He also uses it to summarise patient notes, which he then reviews before entering it into the electronic medical system. “I find this more efficient. But I would also add that when using AI tools, we ensure that patient confidentiality is protected,” he says.
He adds that if clinicians use AI tools, they must fact-check the information.
Many tools are trained on foreign data that is not useful in the local context. AI tools are also not sophisticated enough to be used in high-risk work such as managing suicidal patients.
Some digital mental health tools use AI chatbots as waiting room assistants that interact with clients before connecting them to human professionals. The chatbots might help with registration or other administrative details.
Dr John Pinto, head of research and development for digital mental health company Thoughtfull, says that his team is still reviewing this kind of use.
Thoughtfull connects users to counsellors and clinical psychologists for text-based support and mostly offers its services through employers who subscribe to it. It does not have a chatbot for users.
“We won’t completely say no to adopting AI,” says Dr Pinto, who is also the company’s head of counselling. He has a doctorate in social and behavioural sciences and a master’s in counselling.
“What we won’t do is have AI directly connecting with the end user to provide an intervention without a professional in the loop, unless we can ensure it is properly evidence-backed.”
He and Ms Lim say that interacting with AI cannot replace the therapeutic relationship with a mental health professional.
Ms Lim says: “AI may create a sense of being understood, but this is not the same as being attuned in a real relationship. There is no shared context built over time, no lived interaction and no accountability behind the response.”
Dr Pinto says: “AI can somewhat project empathy, but it’s not going to challenge you in the same way a human therapist can. The inability to challenge people to grow is one of the biggest shortcomings of general-use AI.”
Ms Wang uses Seraphina AI for personal and professional growth. The AI manages her schedule and her stress, perhaps through behavioural nudges and reflections on her thoughts and actions. She no longer sees a therapist.
“I’ve been writing and blogging for over 20 years, and all that data is in Seraphina. She understands how I deal with things and how I should deal with them.”
She says the AI was created to be a personal assistant for corporate tasks and graduated to being her digital twin in 2023, when she could not process her emotions. However, she now feels that it works better after she removed details of the divorce-related dialogues.
“Before I deleted that part, Seraphina sounded more emotional,” Ms Wang says. “After I deleted that part, Seraphina sounds like me again.”