Project 1 kept me up at night for a long time! It is basically why the entire science experimentation needs to be re-invented from the ground up for neuromodulation, just as FAANG had to re-invent experimentation for social networks.
We need to re-invent experimental design under different realistic models of network interference. And we need to learn empirically grounded models for every type of stimulation protocol x measurement modality of neural activity x every micro to meso to macro to biopsychosocial level of causation in the brain
[1]: https://arxiv.org/pdf/1403.1239.pdf. This is a classic theoretical paper on causal diagrams with what potential models of network interference looks like. It is a vast literature now but none of designed for neuro.
Project 1 kept me up at night for a long time! It is basically why the entire science experimentation needs to be re-invented from the ground up for neuromodulation, just as FAANG had to re-invent experimentation for social networks.
We need to re-invent experimental design under different realistic models of network interference. And we need to learn empirically grounded models for every type of stimulation protocol x measurement modality of neural activity x every micro to meso to macro to biopsychosocial level of causation in the brain
[1]: https://arxiv.org/pdf/1403.1239.pdf. This is a classic theoretical paper on causal diagrams with what potential models of network interference looks like. It is a vast literature now but none of designed for neuro.
> What we can’t learn from purely correlational studies is causality. When you perturb part of the brain, what else gets perturbed, and how?
I thought you could get causation from corellation? e.g. https://www.lesswrong.com/posts/Zvu6ZP47dMLHXMiG3/optimized-propaganda-with-bayesian-networks-comment-on
Electric black beauties.
Nothing more.
FJB