Wednesday, April 26, 2017

How quantum effects could improve artificial intelligence

How quantum effects could improve artificial intelligence

That's the title of a recent article in which Chris Altman pointed us to via twitter:

My comment:

This topic has been discussed before, albeit in a form which is more general and thus harder to understand. Reinforcement learning and approximate dynamic programming (RLADP) is not just one of three forms of neural network learning, but a general paradigm for constructing intelligent systems and modeling general intelligence in brains. It includes many designs or methods from many disciplines, the most popular of which are relatively limited (except with clever ad hoc preprocessing), some of which have important applications and some of which pose serious global instability risks. The best overview is the book Handbook of RLADP edited by Frank Lewis and Derong Liu, IEEE press/Wiley 2013. How to implement the most powerful and general forms to fully exploit quantum computing was discussed in Dolmatova and "Analog quantum computing (AQC) and the need for time-symmetric physics”, Quantum Information Processing (2015): 1-15. A six slide overview of larger implications is posted at

To exploit the full capability of quantum computing in this domain, one needs design simulation models which reflect the full degrees of freedom in design, which in turn require new experiments as the next important order of business. The AQC paper gives one option. On this blog I have suggested another, a bit messier but easier -- not an alternative, just a complementary way to get more badly needed data.  

No comments:

Post a Comment