I’ve worked clinically in mental health and suicide prevention for the better part of four decades.
asserted
I → work → decades
And I now believe the emergence of generative AI could be the most positive opportunity to ever emerge in the field.
uncertain
emergence → believe → field
I know that sounds jarring.
asserted
that → know → ?
The potential for chatbots to harm those experiencing mental health issues has been widely canvassed.
asserted
chatbots → harm → issues
Chatbots can promote engagement with sycophantic interactions that support harmful behaviour.
asserted
that → promote → behaviour
They can also state falsehoods or dismiss warning signs clinicians would catch.
asserted
clinicians → state → signs
Given this, some are keen to stuff the AI genie back in the bottle.
asserted
some → give → bottle
But young people who don’t want to consult a human or can’t make contact for practical reasons are turning to chatbots in droves.
asserted
who → want → droves
They function 24/7 in any location.
asserted
They → function → location
When someone is desperate at 3am in a remote community, empathic and evidence-informed help is now available in a variety of commercial and other forms.
asserted
help → inform → forms
Now, I’m not suggesting that turning to a generic LLM like ChatGPT or Claude for mental health care is always a good option.
asserted
turning → suggest → care
But I’m also not suggesting that we should accept the outcomes our current youth mental health system is delivering.
asserted
system → suggest → outcomes
Previous Brain and Mind Centre research showed youth care services are fragmented and poorly co-ordinated.
asserted
services → show → ?
Those in greatest need are often the most neglected.
asserted
Those → neglect → need
We also know the story isn’t necessarily much better for those who attend primary care-based early intervention services.
asserted
who → know → services
In fact, our 2021 research showed nearly two-thirds of those attending these services found their daily functioning – think attending work or school regularly – stayed poor or got worse.
asserted
functioning → show → work
So what are we getting wrong?
asserted
we → get → ?
Well, by practical necessity, most traditional mental health care systems have relied heavily on fairly blunt diagnostic concepts and broad averages to allocate care.
asserted
systems → rely → care
But AI opens up exciting new alternatives.
asserted
AI → open → alternatives
We can now, for example, use smart tools to implement outcome-based care.
asserted
We → use → care
Young people can effortlessly record and report their mood, sleep, activity or other symptoms, risk behaviours and level of functioning.
asserted
people → record → functioning
With privacy and confidentiality strongly respected, we can then use this data to make critical adjustments to treatments we might otherwise have missed.
uncertain
we → respect → treatments
Recording data has been available for some time – but the “intelligence” of AI adds something extraordinarily useful into the mix.
asserted
intelligence → record → mix
Young people don’t neatly summarise their experiences in conventional diagnostic categories or the reductionist language intrinsic to questionnaires.
asserted
people → summarise → questionnaires
They might say they’re “wired” or that their brain “zaps” and “goes burr” or is “full of noise”.
uncertain
brain → say → noise
They might have been on a crowded bus and “freaked out” or had a “meltdown”.
uncertain
They → freak → meltdown
A smart language-based model trained on real clinical encounters and genuine research-based evidence, can process a complex human conversation and turn it into the structured information a clinician needs.
asserted
clinician → base → information
Research is under way to see if this could also prompt less-skilled clinicians to make more specific inquiries that guide decision-making.
uncertain
that → see → making
This kind of guided information-acquisition could turbocharge the capacities of youth services.
uncertain
kind → guide → services
A well-calibrated AI assistant could not only prompt primary-care clinicians and less-skilled youth workers, but bring much of the experience of specialised psychiatrists, mental health nurses and clinical psychologists into the room with the young person.
uncertain
assistant → calibrate → person
This really matters in our outer urban, rural or remote areas where few specialists are available.
asserted
specialists → matter → areas
Most experienced clinicians agree face-to-face assessment and ongoing care are superior to digitally enhanced services.
asserted
assessment → agree → services
Being able to observe key mental state features and participate in social interactions definitely has its upside.
asserted
Being → observe → upside
But we should also be brave enough to step back and ask ourselves if face-to-face is always the gold standard.
asserted
face → step → face
Human interactions are far less consistent than we like to admit.
asserted
we → like → ?
Human judgment is strongly value-laden.
asserted
judgment → lade → ?
When highly trained clinicians assess the same young people independently, they often come to quite different conclusions – not only about diagnosis, but also about treatment or need.
asserted
they → train → treatment
I’m not suggesting AI-assisted care is free from its own biases.
asserted
care → suggest → biases
It’s why some of us are so interested in building and training our own models, and making their assumptions transparent and contestable.
asserted
assumptions → ’ → models
But if we’re serious about assisting many more young people, particularly early in the course of their difficulties, we can’t demand immediate perfection.
asserted
we → ’re → perfection
…and 7 more, not listed.
The Aporia
