Very thought-provoking and topical! I appreciate how you explore and expand on complex ideas without talking down to the reader but still discuss in a way that is easy to understand. Looking forward to more writing on here :))
Fantastic article, I loved reading it :) I did have a quick question - this article mostly focuses on the current limitations of AI, but I think it's fair to say that AI will continue to grow in capabilities over time. As AI's diagnostic and multimodal capabilities advance, I'd be curious to hear your take on how you think the dynamic plays out in the long-term between a psychiatrist and AI? Particularly, how can we ensure the dynamic is synergistic and that we don't use the fundamental unit of human interaction in the process
Two threads I glossed over in the article are worth discussing here; ground truth and transference.
Currently, AIs in psychiatry are limited because there is no consensus on what the diagnostic ground truths or biological ones are. ie Good ML clasification studies seem to have accuracies ranging 60-70%, and it’s unclear how this will be solved.
Assuming a world where this problem is solved through extensive outcome data, there might be a world where psychiatric problems are classified as amenable to AI or humans. For instance, medical treatments for predominantly biological disorders will be easily treated by AI, and psychological problems that have distinct manualised ideas (ie CBT). The human-side likely comes from questions of social interventions or deeper psychological problems, where the therapist themselves are part of the treatment (per transference).
Interesting piece! I suppose one important assumption you make here is that our human nature gives us the privilege of exploiting the concert of “senses” to gauge the patient in front of you that is lost in the compressed signal of text. Curious whether you think there is residual signal if we were to collect all such text, audio, video during a consultation.
Separately, re the interpretation of a patient in the context of the interviewer’s on experience of contemporary zeitgeist, would a LLM not arguably have a better handle on this given the breadth of human culture it has ingested?
Certainly, there’s still plenty of residual data left in the physical world (I’m thinking olfactory and pheromones) - but given the dominance of the visual system in perception (how we have an entire occipital lobe dedicated to vision, compared to less space for other senses), I’m curious how much mileage we can get from this alone.
And agreed re: zeitgeist, if LLMs were constantly updated on modern trends that would be true… I’m most interested in however how the cultural zeitgeist shifts generally toward normalising AI-Human interactions (ie The Wild Robot, Big Hero 6), where younger generations are more and more comfortable with this as a natural dynamic. This feels much more divergent to the Blade Runners, Terminators, and The Matrixes of the 80s and 90s!
Very thought-provoking and topical! I appreciate how you explore and expand on complex ideas without talking down to the reader but still discuss in a way that is easy to understand. Looking forward to more writing on here :))
Thank you kindly Alice :)
Fantastic article, I loved reading it :) I did have a quick question - this article mostly focuses on the current limitations of AI, but I think it's fair to say that AI will continue to grow in capabilities over time. As AI's diagnostic and multimodal capabilities advance, I'd be curious to hear your take on how you think the dynamic plays out in the long-term between a psychiatrist and AI? Particularly, how can we ensure the dynamic is synergistic and that we don't use the fundamental unit of human interaction in the process
Appreciate the thoughts :)
Two threads I glossed over in the article are worth discussing here; ground truth and transference.
Currently, AIs in psychiatry are limited because there is no consensus on what the diagnostic ground truths or biological ones are. ie Good ML clasification studies seem to have accuracies ranging 60-70%, and it’s unclear how this will be solved.
Assuming a world where this problem is solved through extensive outcome data, there might be a world where psychiatric problems are classified as amenable to AI or humans. For instance, medical treatments for predominantly biological disorders will be easily treated by AI, and psychological problems that have distinct manualised ideas (ie CBT). The human-side likely comes from questions of social interventions or deeper psychological problems, where the therapist themselves are part of the treatment (per transference).
Hope those half-formed thoughts are interesting!
Interesting piece! I suppose one important assumption you make here is that our human nature gives us the privilege of exploiting the concert of “senses” to gauge the patient in front of you that is lost in the compressed signal of text. Curious whether you think there is residual signal if we were to collect all such text, audio, video during a consultation.
Separately, re the interpretation of a patient in the context of the interviewer’s on experience of contemporary zeitgeist, would a LLM not arguably have a better handle on this given the breadth of human culture it has ingested?
Appreciate the thoughts!
Certainly, there’s still plenty of residual data left in the physical world (I’m thinking olfactory and pheromones) - but given the dominance of the visual system in perception (how we have an entire occipital lobe dedicated to vision, compared to less space for other senses), I’m curious how much mileage we can get from this alone.
And agreed re: zeitgeist, if LLMs were constantly updated on modern trends that would be true… I’m most interested in however how the cultural zeitgeist shifts generally toward normalising AI-Human interactions (ie The Wild Robot, Big Hero 6), where younger generations are more and more comfortable with this as a natural dynamic. This feels much more divergent to the Blade Runners, Terminators, and The Matrixes of the 80s and 90s!