3 ms·

Thanks for the link. After 60 years of promising to make computers think even the fastest machines on the planet with access to Google's entire database still h

by slacka 13y ago

Thanks for the link. After 60 years of promising to make computers think even the fastest machines on the planet with access to Google's entire database still have trouble recognizing a cat from a dog, so I have to agree with the article that yes it was "Ahead of its time...In the early 21st century, many flavors of parallel computing began to proliferate, including multi-core architectures at the low-end and massively parallel processing at the high end." As 5th generation shows small pockets of researchers haven't forgotten evolution has given each of us a model to follow for making intelligent machines. I hope they continue down this road because faster calculators aren't going to get us there in my lifetime. You feel differently? As the article mentions "CPU performance quickly pushed through the "obvious" barriers that experts perceived in the 1980s, and the value of parallel computing quickly dropped", where as 30 years later single threaded CPU performance has gone from doubling every 2 years to 5-10% per year since ~2007. Combine that with the needs of big data, and the time is right to reconsider some of these "failures". Parallel programming is the biggest challenge facing programmers this decade, which is why we get posts like this on callback hell. Isn't it possible that part of the problem lies in decisions made 50 years ago? I'm not saying we need to start from scratch, but with these new problems we're facing, maybe it's time reconsider some of our assumptions. "It is the mark of an educated mind to be able to entertain an idea without accepting it." -Aristotle

thelukester 13y ago
Funny a cryptographer was just explaining why you should never question your field's fundamental assumptions. http://www.extremetech.com/extreme/164050-discovery-may-make-ewncryption-exponentially-easier-to-break http://www.extremetech.com/extreme/164050-discovery-may-make...
seanmcdirmid 13y ago
The problem is that we've been down this route before and it failed spectacularly (we've gone through two AI winters already!). It does not mean that such research is doomed to fail, but we have to proceed a bit more cautiously, and in any case, we can't neglect improving what works today. The schemes that have been successful since then, like MapReduce or GPU computing, have been very pragmatic. It wasn't until very recently that a connection was made between deep learning (Hinton's DNNs) and parallel computing.
- slacka 13y ago
  Yes, I did say continuing down the current path was necessary. From the languages to the libraries, today's tools allow us to work miracles. As Go has shown with CSP, sometimes these initial failures are just ahead of their time. Neuromorphic computing and neuroscience have come a long ways since the last AI winter. The hierarchical model in Hinton's DNNs is a promising approach. My biggest concern is that all of the examples I've seen are built with perceptrons, whose simplicity makes them easy to model but share almost nothing in common with their biological counterparts.

by slacka 13y ago

thelukester 13y ago

Funny a cryptographer was just explaining why you should never question your field's fundamental assumptions. http://www.extremetech.com/extreme/164050-discovery-may-make-ewncryption-exponentially-easier-to-break http://www.extremetech.com/extreme/164050-discovery-may-make...

seanmcdirmid 13y ago

The problem is that we've been down this route before and it failed spectacularly (we've gone through two AI winters already!). It does not mean that such research is doomed to fail, but we have to proceed a bit more cautiously, and in any case, we can't neglect improving what works today. The schemes that have been successful since then, like MapReduce or GPU computing, have been very pragmatic. It wasn't until very recently that a connection was made between deep learning (Hinton's DNNs) and parallel computing.

slacka 13y ago
Yes, I did say continuing down the current path was necessary. From the languages to the libraries, today's tools allow us to work miracles. As Go has shown with CSP, sometimes these initial failures are just ahead of their time. Neuromorphic computing and neuroscience have come a long ways since the last AI winter. The hierarchical model in Hinton's DNNs is a promising approach. My biggest concern is that all of the examples I've seen are built with perceptrons, whose simplicity makes them easy to model but share almost nothing in common with their biological counterparts.