The Precision Of Sensory Evidence

[semi-necessary prerequisites: Surfing Uncertainty, Friston On Computational Mood]

In earlier posts, I've expressed confusion about two competing models of depression. In one - supported by an analogy to mania and various forms of sensory and motor disturbance - it's inappropriately low neural confidence levels. In the other - supported by common sense - it's a highly-confident global prior on negative perceptions and events - a bias to interpret incoming information in a threat-related way. Both of these models had a lot going for them. But they didn't really fit together. Van der Bergh et al’s Better Safe Than Sorry: A Common Signature Of General Vulnerability For Psychopathology, in the October issue of Perspectives In Psychological Science, tries to tie the pieces together into a more ambitious theory of negative emotionality, including depression, anxiety and trauma.

Its solution is that these conditions are marked by a processing style that assigns unusually low precision to sensory evidence. All perception and cognition is the combination of evidence and priors. But in depression and otherwise neurotic people, the evidence is only a weak signal and the priors are a much stronger one.

Before we get into their argument, a clarification on sensory evidence - yes, this includes normal senses like vision and hearing. But it's also your "internal" senses, like your ability to vividly experience memories. Or your ability to detect and interpret your own emotions. Maybe even an inferential sense, as in when you "sense" that somebody is angry at you by reasoning from their actions. In VDBEA's "negative emotional" people, all of these senses are "low precision" - your brain artificially down-weights their signal, marks them as suspicious and untrustworthy.

This model where negative emotionality reduces sensory precision is supported by lots of oddities we've already talked about so far. Like:

- When depressed people say the world looks gray, they're being literal; visual tests show that their color perception is genuinely dulled. Likewise, recovering depressives say the world seems bright and colorful again, and manic people say the world seems oversaturated, even brighter and more colorful than usual. The association between bright colors and happiness is found across cultures.

- The more depressed you are, the worse your sense of smell, as measured by something called (sigh) the Sniffin' Sticks Test.

- In The Body Keeps The Score, Bessel van der Kolk writes about the deep link between trauma and poor bodily awareness. For example, trauma patients do much worse than normal on tests of stereoagnosia, where you have to identify an object (eg a key) by touch alone. He gives the example of a trauma patient undergoing massage therapy who was so out-of-touch with his body that he didn't even notice the massage had started. Trauma patients tend to say they "can't feel their bodies" or "they're living in a fog"; on successful treatment, they say they "feel alive again" or "I'm back in my body".

- Autobiographical memory is impaired in depression and trauma. This is obvious when eg people can't remember traumatic events at all, or have holes in their memory. But a vast literature shows that depressed and traumatized people have worse autobiographical memory in general. When asked to remember an emotionally neutral event, they will be less specific than a healthy person; this doesn't seem linked to general decline in memory.

Since sensory evidence gets lower weight, the brain is more likely to defer to its prior, which in traumatized/depressed people includes a global expectation that things will be bad.

VDBEA point out that this is a self-reinforcing cycle. Suppose you expect everything to be bad. You go on a date. The date goes fine. But you can only assess how it went using your brain - and that means combining the evidence and a prior. But you downweight the evidence and defer to the prior. That means that unless the date went amazing, you probably conclude that the date went badly. And since you think the date went badly, now you have additional new information supporting your prior that things are always bad. End result: even though your date went fine, you become more certain of your theory that everything is always bad. Your depression deepens.

(I was a little bit resistant to this part. Suppose we rank the goodness of events on a scale from -10 to 10. In reality they average 0, ie completely neutral. You're depressed, so you have a prior that they'll average -9, ie quite bad indeed. And you assess things based 66% on your prior, 33% on evidence. You go on an average date, ie 0. Given your prior of -9 and your weighting, you perceive it as having a valence of -6. You expected things to be -9 bad, but they were really only -6 bad; it seems you were overly pessimistic. You update your prior some bit of the way between -9 and -6. Do this enough times, and you should be able to get your prior all the way back to 0, ie eliminate your depression. Right? The people I asked about this didn't think it would work - it can't possibly be this simple, maybe "bad" and "not bad" are binary states and once your brain decides something is bad it fails to process that it was slightly less bad than expected. Or maybe it is this simple - most cases of depression eventually resolve. I don't know. But it's easy to see that the more heavily you weight your prior compared to sensory evidence, the harder and longer the resolution process will take.)

The way this might feel to you is - you go on the date. Your date looks bored (because your priors make you hypersensitive to cues that they are bored) and the food tastes subpar (because that's how you're expecting the food to taste, and your priors are stronger than whatever stream of real sensation is making it through your taste buds - it can happen to the best of us). Looking over the date after you get home, you hyperfocus on the one really embarrassing thing you said, and any good parts of it mysteriously slip your mind. You conclude that the date went terribly, that you're a useless person who can't do anything right, that the world has it out for you, and you were dumb to think you could ever succeed at dating. Now you're primed to be even more miserable next time around.

But the same thing can happen when recalling and reconsolidating memories. VDBEA link this to trauma processing. Usually when something scary happens, you gradually "process" it in a way related to the normal habituation response. That is, the first time you bungee jump you're probably pretty scared. But you bungee jump a few times, nothing bad happens to you, you learn on a preconscious level that it's not dangerous, and then you're fine with it. In theory you should be able to do the same with traumatic memories: you think about the traumatic event, but since the trauma's over and thoughts alone can't harm you, you're fine and you habituate to the trauma and don't mind thinking about it anymore. In practice - in real PTSD - that's not how it works. VDBEA think it's because the sensory channel is underweighted compared to the priors channel. You think about your trauma (using reconstructed memory perceived through the sensory channel), and that's however bad it is, but you also have your strong prior that this is terrible and thinking about it is horrifying and awful, so the system that should be habituating to the trauma is actually "learning" to become more and more afraid of it.

I've been kind of imprecise here, lumping together depression and trauma and causes and effects and so on. The more rigorous version is that the tendency to underweight sensory evidence (and correspondingly overweight priors) is what VDBEA call a "better safe than sorry" processing strategy. In threat-detection, avoiding false negatives is more important than avoiding false positives - if you mistake a tree stump for a tiger it's mildly embarrassing, but if you mistake a tiger for a tree stump, you're dead. So in a threat-laden environment, you want to both adjust your prior to be that things are threats, and over-weight your prior relative to sensory evidence. Some people could have this processing strategy because they're genetically programmed to have it, and these people will be more vulnerable to depression, and more likely to get PTSD for any given level of trauma. Other people will get traumatized into it - something will happen that ups their estimate of how threat-laden the environment is and makes this cognitive strategy more attractive.

If this is true, what does it imply about treating conditions of negative emotionality?

First, it justifies various psychotherapies - EMDR and coherence therapy seem to come out especially well here - that focus on trying to get the brain to pay unusually precise attention, under unusually safe conditions, to something it usually represses. If dysfunctional negative emotions come from a situation where the attention you're able to focus on a negative belief/memory is so low-precision that you can't override your prior that it's terrible, you need to find a way to focus high-precision attention on it and hold it there even though that will start out being hard and aversive. Trauma-focused therapies seem particularly designed to do that - and maybe something about the eye movements in EMDR specifically increases the available precision of your attention?

Second, it justifies some of Dr. van der Kolk's writing on somatic therapies. Bodily sensation seems to be an unusually important form of sensory evidence. Anything that gets you more "in your body", convinces you to pay attention to your body, trains you to expect bodily sensation to be pleasant rather than aversive, or helps you interact with your body more precisely should increase the precision of sensory evidence in general. So massage, yoga, tai chi, etc.

Third, what if you tried to cut out the middleman, and just focus really really hard on your sensory channel, trying to experience awareness in as much precision as possible? This seemed like an interesting enough lifehack that I tried it for a few minutes before realizing I had accidentally just reinvented meditation. Darn. But meditation is still pretty good and it's nice to have an idea of one of the ways it might work. Meditators talk about a stage of meditation where they feel that their traumas and preconditioning just dissolve under the light of increased awareness, and the idea of increasing precision of sensory evidence might explain what this means.

Fourth, what do we know about all of this pharmacologically? Priors seem to be encoded in NMDA receptors and their relative strength modulated by 5-HT2A receptors, so if you wanted to downweight priors (and so relatively upweight sensory evidence), you would want NDMA antagonists or 5-HT2A agonists. That would mean ketamine and psychedelics, which is a good match for ketamine-assisted and psychedelic-assisted therapies where you take the relevant drug, then explore a trauma or memory that you're "stuck" on, then find that your explorations have "unstuck" you much more than they would have without the drug. Sensory evidence seems to be something something AMPA receptors, so maybe ampakines would also be helpful here, but I don't know of any sufficiently good ones, except maybe ketamine again.

So this model, where inappropriately narrow sensory evidence channels create a bottleneck that makes it impossible to process sufficiently traumatic memories, ties a lot of things together. It gives me the understanding of the trauma-somatization link that The Body Keeps The Score never quite managed. It helps tie together the mechanisms of action for psychedelics, meditation, and therapy. And it resolves the apparent dichotomy between depression as low confidence and depression as negative prior that's been bothering me for so long. This is the most exciting paper I've read so far this year and an important addition to my understanding of predictive processing and psychiatry in general.