But Hinton suggested that, to get to where neural networks are able to become intelligent on their own, what is...

But Hinton suggested that, to get to where neural networks are able to become intelligent on their own, what is…

Reading Time: 1 minute

But Hinton suggested that, to get to where neural networks are able to become intelligent on their own, what is known as “unsupervised learning,” “I suspect that means getting rid of back-propagation.”

Originally shared by Ward Plunet

Artificial intelligence pioneer says we need to start over

In 1986, Geoffrey Hinton co-authored a paper that, four decades later, is central to the explosion of artificial intelligence. But Hinton says his breakthrough method should be dispensed with, and a new path to AI found. Speaking with Axios on the sidelines of an AI conference in Toronto on Wednesday, Hinton, a professor emeritus at the University of Toronto and a Google researcher, said he is now “deeply suspicious” of back-propagation, the workhorse method that underlies most of the advances we are seeing in the AI field today, including the capacity to sort through photos and talk to Siri. “My view is throw it all away and start again,” he said….. In back propagation, labels or “weights” are used to represent a photo or voice within a brain-like neural layer. The weights are then adjusted and readjusted, layer by layer, until the network can perform an intelligent function with the fewest possible errors. But Hinton suggested that, to get to where neural networks are able to become intelligent on their own, what is known as “unsupervised learning,” “I suspect that means getting rid of back-propagation.” “I don’t think it’s how the brain works,” he said. “We clearly don’t need all the labeled data.”

https://www.axios.com/ai-pioneer-advocates-starting-over-2485537027.html

12 comments

  1. Yeah, he’s much more fond of things like Boltzman machines.

  2. Wow my FIRST official job was in ai in 86,I was in high

  3. True !!! Gideon Rosenblatt​

    Wash Out The Contamination from #PampasAspirinBoxes

    Like FaceBook and USBigBanks …

    Regular Folks will Make it Happen …

    But First Purge The Cancer !!!

  4. I agree, back propagation is krap – start over, model the brain, model evolution …

  5. Gideon Rosenblatt the question of AI and Machine Learning (and the two don’t always overlap) is interesting in terms of how machine learning, right now, is set up. Without labelled sets to begin with there is no correction and adjustment phase and even so-called “unsupervised learning” requires a check of output quality tested against a data set whose value are known.

    True AI would self-learn the way we do in the real world where our labelled data set is provided in an informal way by our lengthy, supervised interaction in the real world as we are brought up.

    Trust (a fluid, finite, mutating value) plays a pivotal role in the assignment of any AI the ability to make critical decisions. Trust not just of us in them but also of the AI itself towards any other reporting agent. You will find this talk (and attendant links) fascinating: https://goo.gl/xh4R7A

  6. Thanks for that link, David Amerland​. Sending it to myself for later reading as it does look like something that is be very interested in.

    I just love seeing a pioneer encourage the next generation to question his work for the sake of progress. Very admirable, actually.

  7. After a light study of the problem areas in AI, I think more work will need to be done to AI’s model building ability. The human brain seems to have an intense ability not only to model the world, but itself. Our ability to interact with the world seems to be guided by what most would call “simulations”. Only parallel processing computers (i.e. “supercomputers”) seem to be able to do anything like the human brain does in terms of modeling. I think the next task would be to “loosen” the modeling abilities of these supercomputers to make them better at general AI. Best guess scenarios would be better for general AI than the current scientific models built around massive data sets.

  8. I agree, AI in the future needs to be more “workwithable” , as for Siri saying I can’t isn’t good enough, after all give what customers want.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Sign up here for the latest articles. You can opt out at any time.


Subscribe by email:



Or subscribe by RSS:

Subscribe-by-RSS---Orange-Background
%d bloggers like this: