AI Algorithms Should Attempt to Understand the Human Condition
Currently, it seems impossible to escape the dichotomy that we’re both healthier and more advanced than ever as a collective of civilisations and a species, and at the same time, more disturbed and explosively unstable than ever. The political landscape alone is terrifying the people of countries where political stability and unwavering democratic-socialist thinking have ruled for generations, but are now experiencing radicalism on all fronts. Even those who take refuge in the wise words of philosophers like Atwood (if you don’t think she’s a philosopher, drop the copy of Handmaid’s Tale you’re inevitably reading right now and read her poetry), where she advises us that “It’s happened and un-happened many times in history.”, will feel uneasy about the rhetoric being used by politicians all over the world.
The rise of anxiety as an epidemic has been well covered in Alex William’s wonderful New York Time’s piece, outlining its rise and profound effect on our society. As he puts it, we are constantly assailed by “Push notifications. Apocalyptic headlines. Rancorous tweets.” pushing anxiety-causing news at us, from the failing Arctic shelf to political turmoil to the renewed threat of nuclear war.
At the same time, we read news of life expectancy soaring past 100 years, with new medicine advances making it possible to survive previously fatal disease, and our working lives extending decades past previous retirement age.
The systems of support that attempt to help human beings manage and heal their minds, when anxiety overwhelms them and becomes disease, are failing us through overprescription of benzodiazepines, for which this is a lack of accepted methods of assessing estimates of effectiveness in clinical practice, and which we know cause severe drug dependence, as well as adverse effects on cognitive function, physical health, and mental health itself.
Other support systems, such as Cognitive Behavioural Therapy, have up to 18M waiting lists in frameworks of national healthcare such as the NHS in the UK, and are often unavailable in less socialist systems such as the US, where it’s much easier and more common to resort to Xanax prescriptions.
In the meantime, the mindfulness industry (yes, it’s prevalent enough as a product family and platform to call it an industry) peddles ‘meditation’ and ‘mindfulness’ that rely on people having the time, focus, and mental acuity to understand the basic principles and to apply them assiduously over a long enough period of time (thousands of hours) to internalise its benefits. Meditation can and does help, but needs profound practice. My own work at BioBeats and that of others tries to counter this by introducing biofeedback and data perspective into anxiety management : a form of digital therapeutics that we will need more of as we go deeper into crises (and back out of them).
But for digital therapeutics to work, we need to imbue the algorithms that enable our interventions with a sense of the human condition. This may seem like a facile aggrandising statement, but it’s a simple call to action: without understanding, at the emotional intelligence level, how humans process conflict and suffering, algorithms that hope to analyse mind and body data (psychometric, physiometric and cognitive) will need to understand how to speak their human clients, and even more importantly, when.
In Spike Jonze’s insightful movie, “Her”, Theodore (the lonely writer who buys a new AI operating system), at one point coos to the OS: “I love the way you look at the world”. This is what makes that OS (and its potentially disastrous implications for their relationship) useful: its perspective is finely attuned to that of its human client, from a phenomenological perspective. She tunes her perception to an understanding of the emotional landscape Theodore embodies.
This is called Affect Modelling, or simply Emotion Modelling, and has a long tradition in AI for games, and good overviews of the subject point at affective loops between AI and player that I believe we should be emulating in digital therapeutics.
Digital therapeutic AIs, and the machine learning models they feed on, such as the one we built here, need to understand how human beings feel in order to take us through this increasingly long, often turbulent life.
It’s only when we teach our nascent machine intelligences this sort of understanding that we will win over Xanax.