Algorithmikē Psychē
Three stories came across my desk this quarter that, on the surface, have nothing to do with each other. A government report. A pair of computer science labs on opposite sides of the world. A psychologist writing for a business magazine. But read together, they're about the same thing: people trying to work out, in real time, how much trust to place in a machine that's suddenly very good at sounding like it understands you.
This quarter's brief looks at one governance story, one research finding, and one question about what AI is doing to us socially, day to day.
A Government Report Says the Quiet Part Out Loud
Australia published its final report on AI regulation in healthcare this quarter, and it's more direct than these documents usually are. It names informed consent, data bias, and clinical accountability as the things that have to be sorted out before AI can be trusted in medicine and mental health — not as abstract principles, but as unresolved problems.
What stands out is the honesty of it. Rather than announcing that AI is ready and safe, the report reads more like: here is what we don't yet know how to guarantee, and here is what needs to happen before we can. For anyone whose care — physical or mental — might one day involve an AI tool, that's a more useful thing for a government to say than reassurance would have been.
Read the report →Australian Government Department of Health. (2025, July). Safe and responsible artificial intelligence in health care: Legislation and regulation review — Final report. Canberra: Commonwealth of Australia.
Two Labs, One Quiet Fix for a Real Problem
Researchers at TU Darmstadt in Germany and IIT Delhi in India built something unglamorous but useful: a way to train AI models on mental-health data without keeping a copy of anyone's actual, identifiable information. The data is anonymised and partly synthetic, meaning a computer generates realistic stand-ins for real patient records rather than working from the records themselves.
It sounds technical, and it is. But the underlying question is one anyone would recognise: if a system is going to help spot depression or anxiety earlier, does that mean someone's most private information has to sit in a database somewhere, waiting to be misused? This work is a step toward answering "no" — toward AI that can learn the patterns without keeping the paper trail. It won't make headlines the way a chatbot scandal does, but it's the kind of quiet infrastructure work that decides whether the flashier stuff can be trusted later.
Read the summary →MedicalXpress. (2025, October 13). Towards privacy-aware mental health AI models. Summary of research published in Nature Computational Science.
A Psychologist Asks What Happens to Belonging
Greta Bradman — a psychologist who also researches AI — wrote a piece this quarter that stepped back from the usual "will AI take my job" framing and asked a quieter question: what happens to people's sense of belonging and purpose when AI reshapes how workplaces run. Not whether jobs disappear, but whether the texture of work — the small, human parts of it — disappears first.
Her argument is that mental-health policy has mostly been written for a world where technology changes slowly enough to catch up with. That's no longer true, and she's asking employers and policymakers to build wellbeing support into AI adoption from the start, rather than patching it in afterward once people are already worse off. It's a useful reminder that some of the most important mental-health conversations about AI aren't happening in clinics at all — they're happening in HR departments.
Read the piece →Bradman, G. (2025, October 7). How can AI impact population-level mental wellbeing? Significantly. Forbes Australia.
From a government naming what it doesn't yet know, to two labs quietly solving a privacy problem, to a psychologist asking what work is really for — this quarter wasn't about AI doing anything dramatic. It was about the people around it doing something harder: being honest about how much is still unresolved. That honesty is worth more than confidence would have been.
© 2025 The Algorithmikē Psychē — Q3 2025