Biologically plausible SGD is a hard BCI roadblock
sgd relies on non-local error signals which neurons do not seem to use, making brain-computer interfaces that simulate extra neurons difficult to implement.
the post explores using ai to simulate additional neurons for intelligence augmentation, but identifies the lack of a biologically plausible equivalent to stochastic gradient descent as a major barrier. deep learning relies on backpropagation, which propagates error signals across the entire network, whereas biological neurons appear to update based on local signals and firing statistics. despite observing some gradient-like signals in biology, there is no scientific consensus on a global learning rule matching sgd. the author concludes that discovering a compact, specifiable algorithm for this process is extremely difficult, casting doubt on near-term bci capabilities for cognitive enhancement.