Objective: Can machine learning (ML) enable data-driven discovery of how changes in sentiment correlate with different psychoactive experiences? We investigate by training models directly on text testimonials from a diverse 52-drug pharmacopeia.
Methods: Using large language models (i.e. BERT) and 11,816 publicly-available testimonials, we predicted 28-dimensions of sentiment across each narrative, and then validated these predictions with adjudication by a clinical psychiatrist. BERT was then fine-tuned to predict biochemical and demographic information from these narratives. Lastly, canonical correlation analysis linked the drugs' receptor affinities with word usage, revealing 11 statistically-significant latent receptor-experience factors, each mapped to a 3D cortical Atlas.
Results: These methods elucidate a neurobiologically-informed, sequence-sensitive portrait of drug-induced subjective experiences. The models' results converged, revealing a pervasive distinction between the universal psychedelic heights of feeling in contrast to the grim, mundane, and personal experiences of addiction and mental illness. Notably, MDMA was linked to "Love", DMT and 5-MeO-DMT to "Mystical Experiences" and "Entities and Beings", and other tryptamines to "Surprise", "Curiosity" and "Realization".
Conclusions: ML methods can create unified and robust quantifications of subjective experiences with many different psychoactive substances and timescales. The representations learned are evocative and mutually confirmatory, indicating great potential for ML in characterizing psychoactivity.
Keywords: large language models; machine learning; psychoactives; subjective experiences; testimonials.
© 2023 The Authors. Human Psychopharmacology: Clinical and Experimental published by John Wiley & Sons Ltd.