I have previously written about skin on this blog, in the context of transdermal absorption, and how we might be able to pipe all kinds of chemicals down to the bloodstream. I think skin is underrated and pretty interesting because it’s the barrier between us and the rest of the world.
Something we discussed at Conduit last night is “can you get me out of my head, and put me back in again?” — i.e. can you help me have a subjective experience located somewhere other than behind-my-eyes without breaking psychological continuity? This is a long-term goal of mind-uploading research. The more proper question is “can you get me out of my skin?” — since cognition takes place all over the body, not just in the brain.
I was cooking Christmas dinner a few days ago, and predictably burned myself. I’m very proud of my body for healing up so quickly.
But look — this patch of healed skin is paler. What’s going on?
I’m not a biologist, but GPT-5.2 directed me to “Wound repair: a showcase for cell plasticity and migration”. Here we can start to understand the dynamics:
A skin wound requires several cell lineages to exhibit considerable plasticity as they migrate towards and over the site of damage to contribute to repair. The keratinocytes that re-epithelialize the tissue, the dermal fibroblasts and potentially other mesenchymal stem cell populations that repopulate damaged connective tissue, the immune cells that counter infections, and endothelial cells that re-establish blood supply and facilitate the immune response – all of these cells are ‘dynamic’ in that they are activated by immediate wound cues, they reprogram to adopt cell behaviours essential for repair including migration, and finally they must resolve. In adult tissues, repair is unique in its requirement for dramatic cell changes and movements otherwise associated only with development and disease.
Looking deeper within the article:
One of the key tissue movements of any skin wound healing episode is re-epithelialisation. In a skin wound context, the triggers for rapid keratinocyte activation are numerous, and include damage signals such as H2O2 and calcium, changes in mechanical tension, pathogen sensing, loss of electrical gradient, and serum exposure [30, 31, 32, 33]. Features of activation include induction of stress signalling cascades leading to immediate early gene activation (e.g. Fos and early growth response genes, EGRs [34, 35]), which in turn mediate vast transcriptional changes. The dramatic epithelial response to injury is not only protective through promoting the immune response and limiting DNA damage [36, 37••], but also it actively drives repair by initiating a transient reprogramming of the edge cells [38, 39], a phenotypic change that has been equated to a partial epithelial-to-mesenchymal transition (EMT) [40, 41]. This renders the wound-edge cells migratory and invasive [42], immunogenic (and thus self-limiting) [43], and proliferative [44••]. These migrating cells are clearly very vulnerable en route because of the loss of protective cornified and pigmented stratified layers of normal skin and their exposure to increased stressful stimuli including inflammation-triggered ROS at the wound site. To cope with these stresses, the epidermal cells activate a series of interacting glutathione-NRF2-thioredoxin pathways that together enhance keratinocyte viability during healing [37••].
I of course have an interest in how we can ask the body to act differently, so it is interesting to start learning a tiny bit about cell reprogramming, and think about how cellular activations contrast with neural network activations.
The closest ML analogue to cellular activation is not ReLU—it’s temporarily changing the optimizer, learning rate, and regularization regime in response to distributional shock. We don’t call that activation, but maybe we should. (GPT-5.2)
In the end, everything links back around to control theory. Encounter some unexpected stimulus, inhabit a different mode, gradually re-equilibrate. The economy, the skin, some elements of ML we haven’t pinned down yet…cellular reprogramming as blueprint for continual learning. We have a case in the body where restoration prevails when we might actually like more durable reprogrammability (for eczema, psoriasis, etc.), and a case in ML where we’d love to be able to count on restoration / re-equilibration-or-deactivation in response to shocks. Circling homeostatic ML, etc..