It is notoriously difficult to study the effects of psychoactive drugs in the laboratory. Existing studies can be limited by relatively small sample sizes, and many drugs have not yet been investigated with human subjects. By contrast, descriptions of drug effects abound online. In fact, the subjective effects of almost every psychoactive drug in existence have been documented by countless users through natural language reports freely accessible on the internet. Until recently, it was difficult to gain significant insights from these anecdotal reports. However, the rapid progress of Natural Language Processing (NLP) in computer science has introduced revolutionary techniques to analyze large collections of texts systematically.
In this talk, I will present the preliminary results of a new study investigating a corpus of close to 30,000 natural language reports of drug-induced states using NLP with machine learning. I will show that this approach yields new insights into the subjective effects of a very large number of psychoactive drugs, including many drugs that have never been studied in the lab (such as so-called novel psychoactive substances). In particular, it allows for systematic comparisons between the subjective effects of any drug, as well as the systematic classification of subjective effects into distinct categories. I will suggest that these results can inform future research in psychopharmacology, neuroscience and philosophy of mind, and may have further implications for harm reduction and public policy.