1. 17
  1.  

  2. 2

    How does OCR work, AI or automation?

    1. 2

      If you believe the thesis of the article, OCR is AI made necessary only by our civilization’s inability to digitize everything, while automation would do away with both papers and the need for OCR.

      You might be confused because AI is often seen as one kind of automation. I believe the article tries to distinguish them by correctness (without actually using that word). OCR and other “AI” can make mistakes and require human intervention, while “automation” can be proven correct.

      1. 3

        I think it’s self evident that a world without humans is easy to manage by shifting data to memory regions.

        We all know printers are devilspawn.

        Getting computers to work with us instead of us working with them, that’s the trick, and I’m not sure automation solves that problem. In fact I might be so bold as to say AI gets computers to work with us, and automation helps us work with computers.

      2. 2

        It’s a sliding scale. https://en.m.wikipedia.org/wiki/OCR-A

        1. 1

          I defined “AI” for the purpose of this as “using the term to mean some heuristic for learning equations from data or simulations rather than writing them out by hand”.

          So I think OCR fits as AI, unless there’s a way to do OCR without training data.

          Basically what axelsvensson said.

          1. 1

            A while ago I got this problem of the fuzziness of the term “AI” stuck in my head and thought about it a lot. I think “a computer program that takes on the world on its own terms (without the situation being arranged to suit the computer program)” is somewhat accurate. I think this matches your definition of “AI” and the parenthetical fits how you contrast it against “automation”.

        2. 2

          Trains, assembly lines, and planes are not fully automated in the way that self-driving cars are automated; they have human staff on hand who can manage irregularities and small problems. If we use an apples-to-apples comparison, where the cars are attended by humans but centrally coordinated and powered, then the Commercial Railway seems to have opened in 1840. Certainly, by the 1880s, streetcars and trams were appearing in many cities. In general, the history of trams and light rail is part of the history of trains. I think that bridging the gap between this sort of transit and commuters is a last-mile problem, not necessarily an automation problem.

          1. 1

            That’s actually a good point, but the automation is responsible for flagging irregularities for the human operator to look at. So in the case of a car, this could essentially mean “slow down, and stop”.

            Planes and most trains do usually have on-board operators but this seems to be mainly a regulation issue.

            1. 3

              Chesterton’s Fence applies here; the regulations have a rationale. We want heavy machinery to have operators in case of emergency or exceptional situations which go beyond the capabilities of automation.

              1. 1

                I mean, to a point, yes.

                With e.g. trains there are fully automated systems that work just fine.

                For a plane you could indeed argue cost of a pilot is small enough to justify it as conservative safety, same for a train that goes over unusual and varried terrain (as opposed to, say, a metro)

          2. 1

            I wonder if there aren’t examples of the AI->Automation process which have already occured? Maybe that’s just called automation?

            I’m thinking about machine-readable data formats and CPU or database dependant operations gaining acceptance over time (e.g. the standardization of VINs in the 80’s).

            I don’t think that AI is special in this regard… if anything I would expect it to accelerate the process in all cases where it is present.

            1. 1

              I wonder if there aren’t examples of the AI->Automation process which have already occured? Maybe that’s just called automation?

              I can think of: companies introduced AI in their customer support, then realized they could just automate it. For a successful example see Amazon, for a failed one Revolut. At least that’s my reading of the situation but I can’t be sure.

              But it’s not a very good example, I’ll try to look into this.

              I don’t think that AI is special in this regard… if anything I would expect it to accelerate the process in all cases where it is present.

              I agree