Altering the Mind: Philosophical Challenges for Decoded Neurofeedback

Jorge Morales


With the advent of modern neuroimaging technologies, an impressive capacity has emerged to directly measure brain activity and associated biomarkers indicative of perceptual, cognitive, and affective neural processing. They allow scientists to correlate brain activity with a wide array of psychological constructs, but typically they cannot establish causal relations per se. It is finding causal relations, however, that often provides understanding of the mechanistic nature of a system. Philosophers of science have regarded interventions as a hallmark for establishing causation [1-3], and yet, neuroimaging technologies are usually limited to reading out neural activity without affecting it. Thus, neuroimaging techniques seem quite limited in what they can teach us about neural mechanisms.

Neural interventions can go beyond correlations and establish causal relations. However, neural interventions in humans are either anatomically specific but invasive (e.g. ablations of cortical tissue), or non-invasive but with anatomical limited resolution (e.g. TMS). Most of these interventions are functionally non-specific. Decoded Neurofeedback (DecNef) is different: it allows making functionally-specific, unconscious and non-invasive causal interventions that go beyond the usually limited correlational claims allowed by neuroimaging techniques.

DecNef is a new neuroimaging technique that leverages fMRI and machine learning to create non-invasive, functionally targeted interventions of neural activity [4]. Aided by external reward, DecNef allows subjects to unconsciously self-regulate targeted multi-variate patterns of their own BOLD-activity. Thus, a subject can be trained to self-activate patterns of neural activity indistinguishable from activity corresponding to seeing images of some target stimulus type (e.g., spiders). Importantly, this training does not involve seeing any actual (depictions of) spiders. Through trial and error and a sophisticated online readout of their neural activity paired with a reward system, subjects learn to create “spider” activity patterns in the absence of spiders.

DecNef has obvious clinical applications. A popular treatment of phobias and post-traumatic stress disorder (PTSD) is exposure therapy. Subjects become less reactive to a triggering stimulus (e.g., spiders) by being continuously exposed to (depictions of) them [5]. While this approach is somewhat successful, it takes an enormous toll on patients who have to endure being exposed to their source of fear and trauma. With DecNef, because patients are unaware of the target, they can be unconsciously “exposed” to patterns of neural activity similar to those of their feared target effectively reducing their phobic symptoms without undergoing the emotional labor [6]. DecNef has also been used to reduce depression symptoms, increase performance and alter confidence in a task, create associations, and change preferences [4].

Despite its promise for cognitive neuroscience and for the clinic, DecNef's mechanisms of action remain mysterious. This raises philosophical questions about our understanding of neural representations and neural intervention. In this work, I discuss three challenges for DecNef from the philosophy of neuroscience and the ethics of clinical psychology that call into question whether its interventions are really functionally-specific, unconscious and non-invasive:

(1) Voxels are not representations: A problematic inference on which we are based General criticisms raised against the status of the neural representations supposedly targeted by multi-voxel pattern analysis (MVPA) can be raised against DecNef too [7].

(2) There is no one-to-one mapping between neural activity and patterns of voxel activation. This makes DecNef's impressive functional-specificity mysterious. Despite the matching voxel patterns between trained and target representations, we cannot be sure that the underlying neural activity is matched.

(3) DecNef's interventional nature raises ethical questions. Because of its relative lack of specificity and its potential long-lasting effects (e.g., two months [8]), it is important to question whether unbeknownst to subjects, and perhaps even to the experimenters, harmful or undesirable patterns of activity are activated. Could new phobias be instilled, precious memories altered or erased, undesired associations created?

1. Woodward, J. (2003) Making Things Happen. Oxford University Press, Oxford.
2. Cartwright, N. (2003) Two theorems on invariance and causality. Philosophy of Science 70(1):203-224.
3. Pearl, J. (2009) Causality. Cambridge University Press, New York.
4. Watanabe, T. et al. (2017) Advances in fMRI Real-Time Neurofeedback. Trends in Cognitive Science, 21 (12):997-1010.
5. Craske, M. G. et al (2008) Optimizing Inhibitory Learning During Exposure Therapy. Behaviour Research and Therapy 46:5-27.
6. Taschereau-Dumouchel, V. et al. (2018) Unconscious psychological treatments for physiological survival circuits. Proceedings of the National Academy of Sciences, 115(13):3470-3475.
7. Ritchie, J.B., Kaplan, D., Klein, C. (2019) Decoding the Brain: Neural Representation and the Limits of Multivariate Pattern Analysis in Cognitive Neuroscience. British Journal for the Philosophy of Science, 70:581-607.
8. Megumi F et al. (2015) Functional MRI neurofeedback training on connectivity between two regions induces long-lasting changes in intrinsic functional network. Frontiers in Human Neuroscience, 9:160.

Poster for the Philosophy of Science Association Meeting 2020/2021