Humans, individually, are essentially a neural network. We are each a foundation model. Most of the data, in my opinion, is inherited through our genome. We are booted by our parents genomes, which has all of the necessary adaptions (hopefully) to survive through the environmental changes through our lifetimes.
When we are young, we are taught by our parents, our institutions, and our science about knowledge, how to act, what needs to be created, and what needs solving in that specific period of time. However, I personally believe that the learning part, and whatever other behaviors we learn, is simply fine-tuning what we inherited.
If this is is the case, then we have to think about engineering the body in a different way. A simpler way. If we can make live edits in the body, one that is already almost full formed from what we inherited, then this is just fine-tuning the model that is our body.
Meaning we need to capture as much information in our model of each individual as possible, so we can deliver the correct tuning strategy for whatever is wrong in the body- whatever system has grown to be beyond controllable of our inherited management systems (immune system, senses).
If we then create this model, we can tune it, like we would with a reinforcement learning agent on top of an always-updating neural network. Though the math of RL is fairly crude and simplistic, each element of it (policy, agent, environment, reward function), will become their own frontier of ever-expanding variable capture via mathematical description.
I wonder how these two fields will merge.