Bridging Machine Learning and Psychoanalysis

Artificial Intelligence + Psychoanalytic Research 

Walking the Tightrope: Bridging Machine Learning and Psychoanalysis

By Ilana Gratch

At first blush, machine learning might seem to have little to do with the subjectivity, individuality, and humanness that characterizes psychoanalysis.

But machine learning — a subbranch of artificial intelligence in which statistical algorithms perform tasks independently by learning from patterns in data — is, in essence, about pattern detection. Part of what makes ML so powerful is that it can discern patterns that often escape human awareness.

Similarly, a couple of years ago, I heard Mark Solms (a South African psychoanalyst and co-founder of the International Neuropsychoanalysis Society, well-known for integrating psychoanalytic theories and methods with modern neuroscience) liken this sort of data-driven pattern detection (as opposed to more standard, hypothesis testing done in traditional inferential statistics) to “evenly suspended attention” — a state of open-minded “alert passivity” which Freud proposed in his 1912 book “Recommendations to Physicians Practicing Psycho-Analysis.” In both computational pattern detection and analytic listening, we are attending to data without imposing agendas or biases, hopefully opening novel avenues for exploration.

Almost 100 years ago, Freud wrote: “The use of analysis to treat neuroses is only one of its applications; perhaps the future will show that it is not the most important (Freud, 1926). He continued: “The future will probably judge that the meaning of psychoanalysis as a science of the unconscious far outweighs its therapeutic significance” (Freud, 1926).

In the decades since, researchers across disparate disciplines, from economics to social psychology, have devised creative empirical studies demonstrating the power of unconscious, out-of-awareness processes (e.g., Soon et al., 2008; Bargh et al., 1996). Yet within psychoanalytic circles, this sort of empirical research happens mostly on the margins.

Fortunately, a number of pioneering psychoanalytic researchers are broadening the circle through their exciting and thought-provoking work. For instance, Marianne Leuzinger-Bohleber has documented changes in dream content over the course of psychoanalysis (e.g., Leuzinger-Bohleber, 1989) and has conducted outcome studies of psychoanalytic treatments that examine not just symptom reduction but also changes in patients’ inner worlds, i.e., structural change (Leuzinger-Bohleber et al., 2019). Mark Solms’ gap-leaping work in neuropsychoanalysis endeavors to “correlate our psychoanalytic models of the mind with what we know about the structure and functions of the brain” (Solms & Turnbull, 2011). And Beatrice Beebe employs video-tape microanalysis to capture and study nonverbal aspects of mother-infant communication that are largely out of awareness (e.g., Beebe & Lachmann, 2020).

It is understandable that these kinds of investigations are the exception rather than the rule. Studies aiming to examine complex processes theorized to be central to the analytic endeavor are incredibly difficult to carry out. They not only require what Leuzinger-Bohleber and Target (2002) have referred to as “tightrope-walking;” they are also time-consuming and expensive, necessitating resources often unavailable.

But this kind of empirical research is, in my view, more than just ground-breaking; it is, alongside other forms of psychoanalytic research, an essential component of keeping psychoanalysis alive. We have heard a lot in recent months about the threats posed to meaningful traditional psychotherapy from rapid advances in artificial intelligence, including therapy bots and a host of other computational services.  There are good reasons to be skeptical of, and cautious about, the actual capabilities of these technologies. But I believe there is also an unparalleled opportunity. Leveraging advances in machine learning may be one approach for furthering psychoanalysis as the science of the unconscious that Freud envisioned.

There are many exciting possibilities at the intersection of machine learning and psychoanalytic processes. In my own research, I have used machine learning-based software (Hinduja et al., 2023) to characterize nonverbal behaviors of young adults and their clinical interviewers during suicide assessments; the deep learning-based software I use codes facial and head behavior for every frame of video, which, when applied to videos with a frame rate of 30 frames per second, captures moment-to-moment affective information impossible for humans to grasp.

Similar software exists that can characterize vocal prosody and turn-taking in interpersonal interaction – that is, not just the content of what a person says but the way it’s being said. Natural language processing tools can be applied to text data to reveal patterns in language use, such as in transcripts of dreams or session content, e.g., tracking changes in the dreamer’s use of emotion-laden words over time. These tools make possible the characterization of rich and complex information which is incredibly challenging to capture otherwise, and are also often open-source, highly efficient, and freely available.

I want to appeal to everyone interested in psychoanalysis as the science of the unconscious, inside and outside of the consulting room, to consider the promise of machine learning. If we open ourselves up to the possibility – and work using our very human minds to devise rigorous, creative, ethical studies – we may come to find that machine learning can help us discover and rediscover much of what we know to be special about psychoanalysis.

References

Bargh, J. A., Chen, M., & Burrows, L. (1996). Automaticity of social behavior: Direct effects of trait construct and stereotype activation on action. Journal of Personality and Social Psychology, 71(2), 230-244.

Beebe, B., & Lachmann, F. (2020). Infant research and adult treatment revisited: Cocreating self-and interactive regulation. Psychoanalytic Psychology, 37(4), 313-323.

Freud, S. (1912). Recommendations to physicians practicing psychoanalysis. Standard Edition XII. London: Hogarth Press.

Freud, S. (1926). Die Frage der Laienanalyse. GW, Vol XIV.

Freud, S. (1926). Psycho-Analysis. GW, Vol. XIV.

Hinduja, S., Ertugrul, I. O., Bilalpur, M., Messinger, D. S., & Cohn, J. F. (2023). PyAFAR: Python-based automated facial action recognition library for use in infants and adults. 11th International Conference on Affective Computing and Intelligent Interaction Workshops and Demos.

Leuzinger-Bohleber, M. (1989). Veränderung Kognitiver Prozesse in Psychoanalysen. Bd. 2: Fünf aggregierte Einzelfallstudien. Berlin: Springer (PSZ).

Leuzinger-Bohleber, M. E., & Target, M. E. (2002). Outcomes of psychoanalytic treatment: Perspectives for therapists and researchers. Whurr Publishers.

Leuzinger-Bohleber, M., Kaufhold, J., Kallenbach, L., Negele, A., Ernst, M., Keller, W., … & Beutel, M. (2019). How to measure sustained psychic transformations in long-term treatments of chronically depressed patients: Symptomatic and structural changes in the LAC Depression Study of the outcome of cognitive-behavioural and psychoanalytic long-term treatments. The International Journal of Psychoanalysis, 100(1), 99-127.

Solms, M., & Turnbull, O. H. (2011). What is neuropsychoanalysis? Neuropsychoanalysis, 13(2), 133-145.

Soon, C. S., Brass, M., Heinze, H. J., & Haynes, J. D. (2008). Unconscious determinants of free decisions in the human brain. Nature Neuroscience, 11(5), 543-545.