Hotz: “The last people with jobs will be AI programmers” (+ comments on neural nets for ‘knowing how’)

A fascinating interview with George Hotz: http://www.bloomberg.com/features/2015-george-hotz-self-driving-car/ He does remarkable things.

Final edit, as the story behind this post continues to unroll: Tesla’s reaction.

Hotz has a different approach to ‘programming how to drive’ than (according to him) others that try to achieve the same. What he does has strong connections with epistemological theories on ‘knowing how’. Basically, what he claims is that the knowledge involved is hard to ‘propositionalise’ and is better learned through experience. The same is true for Ryle’s (or Polyani’s) famous example from the philosophical literature: ‘knowing how to ride a bike’.

Another interesting fact: that neural nets outperform other AI-techniques in this context was already known in the context of racing games: an interesting example of knowledge transfer from virtual to real environments.