Relaxed Beliefs under Psychedelics
October 05, 2019
Robin Carhart-Harris, whom you might know from Michael Pollan’s book, may have done it again.
In 2014 he wrote the insanely insightful Entropic Brain paper, which everyone, regardless of interest in psychedelics/neuroscience, should read. It’s that good.
Now he’s written the rather less beautifully titled REBUS paper (“Relaxed Beliefs under Psychedelics”), and so far it does not disappoint. Here’s a sample section that I’m still thinking about (BMR = Bayesian Model Reduction, BMS = Selection):
BMR is a particularly intriguing form of BMS that is likely to play a central role in brain development, e.g., in the form of synaptic pruning (Piochon et al., 2016) and formation of small-world architectures (Avena-Koenigsberger et al., 2014). In brief, BMR is the hypothesized mechanism via which high-level models are stripped of their redundancy so that simpler, more refined solutions may be revealed. Again, we see the theme of complexity minimization and compression in play. In this setting, one can refine high-level models or narratives to make them simpler by removing redundant parameters, thereby revealing the underlying core structures and manifolds. Crucially, this mechanism can proceed without the need for new data (fact-free learning) and is thought by some to be the purpose of sleep and accompanying synaptic homoeostasis (Hobson et al., 2014).
In other words he’s arguing that moments of insight come not from new facts, but from the stripping away of detail, to reveal underlying structures. This relates to my interest in induction. I suspect that induction requires a period of exploratory gathering of data, followed by a period of outwardly inactive “stewing” — after which a pattern emerges. Critically, these patterns are recognised by the subconscious, not the conscious, mind, which is why the insights seem to come out of nowhere. And these insights have less information, not more, than the stimuli that provoked them.
I also wonder if a similar process isn’t going on with mastery. Learning seems to involve a process of apprehending that there is something out there to be learned, followed by increasing attention to that object, then eventually a decreasing amount of attention to the object once it is known, with a concomitant loss of detail. These stages each seem quite separate to me. Carhart-Harris divides the first two stages in terms of “explorative search” (which dominates in childhood) and “exploitative search” (preponderant in adulthood, both facts probably relating to the formation of the default-mode-network).
I still need to think about the explorative stage in light of Agnes Callard’s points about aspiration.
Anyway, during learning, an object increasingly dominates consciousness, but eventually the level of detail falls away. Once it is mastered, it becomes “part of the landscape,” which can also mean that it’s harder to articulate. (Think of asking pro athletes precisely how they did something, or the speed with which you recognise a familiar word in your native language.)
Perhaps there is a neurological basis for this: during learning, you’re forming new synapses. But mastery is a stripped back, compressed (therefore more efficient and less entropic) version, having lost most of the detail gleaned along the way. Autopilot; part of the landscape. I also wonder if this isn’t how memories simplify.
Perhaps the types of platitudes he discusses under psychedelics (“Love is everything”) also relate to near death experiences (like when Andrei falls and can think of nothing but “that lofty infinite sky” in War and Peace). And perhaps this loss of detail could also explain the platitudinous ring of some interviews with consummate experts in various non-speech-related fields.
I have a lot more thoughts on this paper, which I’ll probably need to re-read a few times. If you’re in London, we’re discussing it on Monday in Angel.