1. 23
  1. 3

    I’ll bite. I predict full AI for self-driving cars and radiologists will arrive at the same time Linux overtakes Windows on the desktop.

    I’ve never liked the term Artificial Intelligence because it communicates the wrong message. Neural Network is a better term, but if course it doesn’t sell as well.

    This is all a repeat of the hype around AI back in the 80’s.

    1. 3

      At least this time there’s no Lisp Machines to kill…

    2. 4

      If you plan for anything on an infinite timescale you will always be right. Everything is inevitable eventually. I think this article is extremely premature. Things are just getting started. The reason why there was an AI winter last time is because the path was a logical dead end, the same can’t be reasonably evidenced of modern techniques. In fact the author never even makes that claim, they just make the claim that the present implementations are insufficient.

      1. 8

        The reason why there was an AI winter last time is because the path was a logical dead end

        That itself is an overstatement. A lot of the techs, from rule-based systems to computer vision, are still in use across the globe. They could do a lot for businesses. The AI Winter happened anyway. The reason was overpromoting what the AI’s could do with massive investments into specific techs and companies without ROI justifying that. That’s happening in current market. So, it could tank. Preventing it would just take them making more honest claims about incremental progress in these situations with these constraints or future work needed. They might get less funds or smaller buy-outs that way, though. ;)

        1. 1

          That’s a fair interpretation of my statement, but by logical dead end I meant that it became obvious that a very different approach would need to be taken to move forward, this is of course what the author is claiming about neural nets but I still think it’s a bit premature to say nothing new will be invented. A lot of these ideas haven’t even had enough time to be trialed. I agree that expectations management should be a thing but marketing will be marketing. I still think it’s extremely early to say the winter is on its way. Some of the things that neural net based techniques can do are honestly impressive. I sincerely doubt we’ve come anywhere near the peak of the hype cycle. If anything we’re up the first 100 feet from the foot of the mountain.

          All being said, winter always comes so unless there’s any kind of time frame it’s a bit of a truism.

          1. 1

            Interesting point on timescale. Well, the prior field was built on the time when computers were gaining in power and reach in businesses. That was fueled by large sums of money by groups like DARPA willing to loose on big bets. Businesses kept buying the stuff with it not doing what it said or only marginally better than other methods. Then it collapsed.

            The current situation is both computers and Internet are in pervasive use. No need for slow buildup. There was also a Big Data push with data warehouses and stuff that came before this. The analysts with their tools are already doing what Deep Learning is expected to do better. Proponents are vastly overselling that. Im not sure how long the cycle would take if it’s a bubble going to bust. That’s the lead up for you, though.

            1. 1

              I think while the future is unknowable, AI is really really hard to predict right now. Lots of milestones that were not thought to be passed in the near future have been. Good applications have already arose in computer graphics with things like outstanding denoising results. Self-Driving cars have come a very long way, but it’s also an extremely hard problem. The extreme difficulty and extreme progress make self driving cars in my opinion very hard to call. The proponents are overselling but that’s not really a predictor of success or failure. Proponents will always oversell. When we get self-driving cars the proponents will say “It drives 10 times better than a human”. We’re already seeing brand new military research getting started. So it’s possible this is just the beginning of the hype cycle

      2. 1

        Well yes there are limitations… I think you can work around most of them though. So maybe there was too much hype. But AI is here to stay. It will change everything because it allows more automation. It’s just really far away from a general AI, or sentient AI. Pattern recognition on steroids would be a better name, though not so sexy as AI.

        1. 0

          Feels like a weak argument. I wonder if the author will agree to eat their hat if Waymo gets cars driving the general public around in Phoenix by the end of the year…

          1. 2

            Bold claims require bold hats being eaten.

            1. 1

              Italic hats taste better