This approach develops skills based on experiential learning. Computer algorithms include neural networks. Robots have been built to interact with each other and to have learning capabilities.
This approach has been less successful than the first in simulating interesting and challenging skills such as playing chess or competing in Jeopardy. They do illustrate learning, but it is not proven that they simulate how humans learn. So far they have not been used extensively for diagnosing problems and helping to improve skills.
With the development of neural imaging and increasing interest in computational neuroscience, there is a lot of interest and some effort in building artificial intelligence simulations out of neural components.
There very few such models, and practically no experience in using them for skill engineering. Assuming that we can find a working simulation for the kind of skill we are working with, we can focus on changes in the skilled behaviour due to a sequence of experiences. We can add to the list of questions.