>>14245186It's not, it's actually an active area of research to implement neuro-mimetic NNs, they're called spiking neural networks. The implementations include temporal inclusion to better simulate how, for example, our parallax vision neurons work (really cool. Certain neurons fire, and if another set of neurons fire at an exact rate a little further in time, it activates the network, essentially using the physical lag between neuron firing and the impulse reaching the end of the neuron as a timer), and researchers have started actually building specific hardware to mimic neuronal structures that these spiking neural networks run on. As it always is, despite the incredibly impressive simulation capabilities of these right now, they aren't AI (just yet). I work in the field of neural development/ML drug discovery, and I've kept my eye on recent advances in that field; the renessaince in ML in the past ~5 years (LSTMs, Seq2Seq networks, attention models, transformers, generative models) have almost entirely been pushed by NLP, but there's been an undercurrent of improvements in neuron/brain mimicking architecture. I'm really sad that memory-augmented neural networks haven't been in the limelight; I thought neural turing machines/differential neural computers would make waves, but they've been largely ignored.
