In November, I’m presenting a paper tentatively titled “The Electric Kool-Aid Turing Test” at this conference in Brighton, England. My argument is that emerging paradigms in research on the use of psychedelic drugs as psychotherapeutic tools problematize machine learning.
To be more specific: the recent resurgence of psychedelic drug research has, generally, privileged quantitative and empirical methods over qualitative and interpretive methods. There’s nothing inherently wrong with this. My suspicion, though, is that quantitative methods may not be sufficient to understand trip reports in a way that contributes to workable medical knowledge — that the data gathered from the profoundly subjective experience of a psychedelic trip requires as much theorization, sensitivity to context and (quite frankly) creativity to parse in order to draw meaningful conclusions from them. This disturbs the capacity for them to be operationalized in an algorithm, especially a machine-learning algorithm that relies on inference and prediction to extrapolate beyond finite training data.
This is pulled from my abstract:
The paper draws from interdisciplinary scholarship that uses qualitative methods to interpret research on psychedelics as psychotherapeutic tools. I combine precepts of machine learning with developments in psychedelic research to explore the complexities of generalizing findings, which includes accounts from those undergoing “ineffable” and difficult- to-predict experiences — for data modeling. In doing so, I demonstrate that the use of qualitative methods in psychedelic drug research may offer useful insights to the field of machine learning… [later] I explore axioms of machine learning that emphasize the ways in which generalization and inductive reasoning are used to build algorithms that effectively “predict” the future. Here I partially draw on the work of Pedro Domingos, whose research explores how machine learning generalizes “beyond” finite data sets. Joining emerging paradigms from psychedelic research and machine learning, I offer that the former can help the latter a) account for difficult-to- predict phenomena and b) understand its possible limitations.
The turn toward interpretation and qualitative approaches is the “emerging paradigm” in psychedelic research to which I refer. Scholarly work has been published very recently that foregrounds qualia and subjectivity in clinical trials with psychedelics. Furthermore, some scholars emphasize that philosophy is implicated in the outcome of these trials, and that the answer to why psychedelics are effective at treating mental illness may be entangled with traditionally philosophical concerns.
I’m not unique in connecting psychedelics to machine learning / machine consciousness. Andrew Smart’s book Beyond Zero and One attempts to theorize machine consciousness with the question: could a robot [an artificially intelligent mind] trip on acid [some sort of digitally modeled version of LSD]? Beyond Zero and One came out in 2015 and I suppose I could be accused of stealing some of Smart’s ideas, at least on the surface. Yoinks, a connection between machine consciousness and psychedelic drugs!
Actually there’s a funny story here: I was working at the publishing company behind Beyond Zero and One in the months leading up to its publication — I helped with the publicity for it. Despite a long-standing interest in psychedelic studies and computer science, the book seemed too much like a pop-science head trip to me to bother reading it and I basically forgot it existed until I started doing this work. (I could have even snagged a free copy, but I didn’t … sorry to Andrew Smart and my former employers at O/R Books…). At any rate, not accounting for the possibility that it somehow got lodged in my unconscious mind, I can say that Andrew Smart and I arrived at the same connection independent of one another. With questions about the mind, consciousness, associative thinking, and so on central to both psychedelic studies and machine learning, this seems fairly plausible.
At any rate, I’m fully in the throes of this project now. It may become part of my dissertation — I hope it does, but as folks in higher ed know, some of that is beyond my control. Honestly, the fact that psychedelic research is so controversial will make this already complicated work all the more difficult. For the most part, I’ve been fairly quiet about this interest to avoid raising eyebrows. But since I’ll be “outing” myself by giving a psychedelic conference talk anyway, a possible new way of confronting the controversy is to own it — to be open about it, stand behind what I’m doing. To that end, I’ve been tweeting a lot more about psychedelic stuff, testing the waters I suppose, and talking to some sympathetic colleagues.
There’s a lot more to say, interesting connections between books and papers I’ve consulted. I’ll write it all down somewhere. One day.
The Virginia Tech library doesn’t have an extensive amount of material on psychedelic science, but they do have a copy of Albert Hoffman’s LSD: My Problem Child that I’m going to check out before I leave here today…