Is There a Smarter Path to Artificial Intelligence? Some Experts Hope So | Tech News

I think ML is doing amazing things, but it will never generalize. ML is really advanced function search, that’s it. There are achievements being made in ML transfer, but nothing too impressive. The technology is limited in its scope of application. It will be faster better cheaper, but never really general.

Will ML generate ASI, yes. It already has (it is killing in all forms of non-temporal patter detection). Will it generate Super-AGI, I don’t think so. I don’t think it has the right ingredients. ML techniques need loads of data. They do not learn like protein information structures do, (our benchmark of intelligence). They don’t learn in hierarchal patterns. They use minima search across vast number of interrelated factors. In a few years we’ll stop talking about ML as AI, just as we stopped talking about search as AI. Like Kurzweil says, once it works we stop calling it AI. ML will just be a cool tool to generate excellent pattern matching results.

Perhaps the hardware will improve enough that the huge function space required to achieve temporal pattern recognition and prediction won’t be the tech killing bottleneck, but I also doubt that as well.

There are better approaches to AGI. My focus is on SDR’s. I think Chris Elaismith is really on to the right path with his Semantic Pointer Architecture. Numenta seems to be playing the right game too.

It’s all fun to learn at watch. No matter what, I do not see an AI Winter coming anytime soon. ML still has a 100 fold improvement to be made with better GPU’s, plus funding in other AI Research areas is exploding, mainly thanks to ML successes from Deepmind, Goodfellow, Bengios, LeCunn, Ng and others.

Edit: A word. Plus I’m learning LISP so I can know what else was happening. If anyone else is working though SICP, holler.

You might also like More from author