The lever

The argument

Most help works one problem at a time: one therapy for anxiety, another for conflict, a third for procrastination. If Claims 1 to 3 are right, many of these problems share one upstream cause, which is the way the self-model handles its own evidence. Changing that cause should then help in many places at once, including places that were never trained.

This is the boldest claim in the thesis, and we hold it as a belief. We are not claiming the self-model is the biggest single cause of suffering. Poverty, illness and institutions matter enormously. We are claiming it may be the most widely connected lever, because it also slows how fast people and groups correct their mistakes. Fix the upstream error and you do not just reduce pain; you speed up learning.

What the evidence shows

The first piece of evidence is that psychology has already been moving upstream. For decades, treatments were built one diagnosis at a time. Then it became clear that a few processes run through most of them. Repetitive negative thinking appears in depression, anxiety, insomnia and more, and predicts later symptoms across the board (Ehring & Watkins, 2008; McLaughlin & Nolen-Hoeksema, 2011). Treatments that target the shared process instead of the diagnosis work. In a randomized trial of 223 patients with four different anxiety disorders, a single transdiagnostic protocol matched the disorder-specific treatments on outcome, and patients were three times more likely to complete it (Barlow et al., 2017). One lever moved four conditions.

The second piece is that the shared processes point at the self. The strongest candidates have a common shape: a change in the relationship to one’s own thoughts rather than a change in their content. Decentering, the ability to see a thought as a mental event rather than as reality, is now considered a core mechanism across several effective therapies (Bernstein et al., 2015). Self-distancing, which means examining a painful memory as an observer rather than reliving it, lets people analyze negative experiences without spiraling into rumination (Kross & Ayduk, 2010). Psychological flexibility, the capacity to shift mindset when it stops serving you, is described as a fundamental aspect of health and is missing in many forms of psychopathology (Kashdan & Rottenberg, 2010). Self-compassion, which removes the verdict from failure, correlates with lower psychopathology at r = −0.54 across 14 studies, a large effect (MacBeth & Gumley, 2012). Each of these is a way of loosening the grip described in Claim 1. Each helps across many problems at once. We think that is not a coincidence.

The third piece is that interventions aimed at the self already show spread. Mindfulness programs, whose proposed mechanisms include a change in perspective on the self (Hölzel et al., 2011), produce moderate improvements in anxiety, depression and pain in the same program (Goyal et al., 2014). The effects are modest in size, which we take seriously below. Brief self-affirmation exercises, which let people hold a threat to one part of the self without defending the whole, improve outcomes in school, health and relationships, often long after the exercise (Cohen & Sherman, 2014). Intellectual humility, the ability to recognize the limits of one’s own beliefs, is now studied as a possible antidote to societal acrimony, with documented benefits for personal decisions, relationships and science (Porter et al., 2022), tying the lever back to the group-level problem in Claim 3. In the brain, the network most active during rumination is the same self-referential network involved in thinking about who we are (Zhou et al., 2020), which is at least consistent with one hub rather than many.

Now the honest caveat. All of these effects are moderate. Meditation program effects on anxiety and depression are around d = 0.3 (Goyal et al., 2014), and self-affirmation effects on wellbeing are also small, around 0.2 to 0.3 across 129 tests, though they last (Zhang et al., 2025). The existing literature establishes that pulling on the self produces spread. It does not establish that the spread is large. Our bet is that current methods loosen the grip without changing the update rule, so the gains are real but partial. Claim 5 is about whether a deeper change is trainable.