1. 60

  2. 23

    As a programmer, I read this and think, yes, that’s an unsurpassable moat. How will Apple–much less OpenStreetMaps–ever catch up?

    As someone who worked with Sanborn fire insurance maps in a previous career, I’m not so sure. Sanborn maps had even more detail than today’s Google Maps–they showed not only building footprints but building function, number of floors, construction materials, and other features of interest to fire insurers (sometimes interior structural walls were indicated). They existed for big cities but also for places like Circle City, Alaska and Adrian, Michigan. I have a hard time imagining the level of effort that went into creating and maintaining these maps (they were literally updated by mailing out sheets with bits of paper you would cut out and paste over your existing map, usually at the level of individual buildings). But people managed to do it without aerial or satellite imagery, or ML image recognition, or any of the other tools available to us today. It’s hard to imagine that Apple couldn’t–if it wanted to–replicate something a much smaller company (the Sanborn Map Company employed 700 people at its peak) was doing a hundred years ago.

    1. 10

      The Sanborn example is a really good one. As a fire insurance company, they invested in good maps because they were potentially liable for a lot of unexpected costs from inaccurately-estimated risks. The maps were literally Sanborn’s business. Apple sells phones and computers so they just need a good-enough map to keep people from jumping ship to Android or allowing GMaps an enclave in iOS territory. What does Google need this level of detail for? Ultimately they sell ads, and they’ve been very creative in figuring out ways to expand the potential surface area for ad sales and improve consumer data flowing back to them.

      1. 3

        I agree with you, it’s all about who the detail is for. It’s not unsurpassable. It really depends what you’re trying to do.

        Mapping is one of the most subjective pieces of data you can offer: a map’s value is only what the reader gets from it. That’s why we have so many… hiking maps, road maps, the fire maps you point out. Is the information that the article notes others don’t have really that valuable? Not to me. I’m sure it’s valuable to some. I’ve tried Apple Maps again because it worked nicely with the iPhone X out the box (note to app developers: you don’t get second chances, you gotta be there at the beginning) and it seems fine for road maps, which is what I need it for. I also like the Yelp reviews that are embedded. I remain skeptical about the traffic information, though.

        Waze is a really good example of a map that’s hyperfocused on a single use case: driving. You got roads. You got traffic. You got where you can get a donut. You don’t need to know the shape of a building.

        I guess I just don’t find the idea of moats all that compelling. I think we’ve seen time and again in tech and elsewhere that when people see a moat, what you usually have is a very broad offering which leaves opportunities for very focused offerings to do better (even Google was this at the beginning, Yahoo had all the content out the wazoo, and everyone thought you couldn’t compete with that, and Larry and Sergey just built a very good search engine that rocked the one Yahoo had).

      2. 3

        I sat next to a Apple maps engineer on a flight recently. Was told it’s a 1500 person org. Kind of shocked me.

        1. 1

          Most of those people are acquiring and processing data, similar to other mapping orgs.

      3. 15

        looks at map

        Someone actually named their hotel “Dick’s Halfway Inn.”


        1. 3

          As someone who worked in the mapping industry, I saw Google doing this years ago and investigated. First thing I noticed was that google was using satellite imagery to generate these building footprints. I doubted they used any 3D information because many things that were not building, like shadows on a driveway, were picked up as buildings.

          The footprints were so hilariously bad compared to the real ones I concluded workers don’t look at them outside large city centers. Yes google made buildings out of computer vision algorithms, however it was clear to me the algorithms were crude. Even in rural areas where they have 3D reconstructed data, they don’t use that to compute the building footprints, at least not yet.

          I went ahead a took a random place and compared the 2D footprints with the 3D reconstructed one and it is clear google isn’t using that information yet. See Here

          I took a 2d screenshot and used image processing to pull out the footprint and layed it on top of the satellite image. The footprints are WAY off and missing major features. So yeah, they use a crude shape detection algorithm based on 2D satalite imagery.

          However, the 3D reconstructed buildings are very nice, and likely touched up by the thousands of special cheap labor they use abroad that works on google maps.

          1. 1

            Yeah, it was quite obvious that they’ve done this by processing satellite imagery once the author showed that it’s done across all population centres.

            The idea of matching positions of shops with polygons and then constructing a hull for multiple polygons to define an AOI doesn’t sound revolutionary either. It seems the article is a bit too breathless. Some of the UI discussion and the development retrospective is certainly interesting, but I’m not quite sure why this can’t be reproduced by anybody else.

            1. 2

              Even more curious is the author works for Apple…

              1. 2

                I think he doesn’t anymore.

          2. 1

            By the way, it seemed to me that Google uses planes to collect high resolution 3D imagery in some places.

            1. 1

              Impressive to me is the fact that it appears to be a product roadmap dream: e.g. this milestone makes the next one possible, forever and ever. I can’t point to very many places where you can look at releases and see how they could all fit together (as this post clearly illustrates for maps), and I have to wonder why?

              Is it simply (ha) the case that, since cartography has been studied for ages, we simply understand the problems in greater depth and just need to catch up with better technology (which Google has successfully invented purely via looking at data differently)?

              But, if it’s actually just raw product talent, and not “well understood problem domains”, why is it that Google can do such a great job with product on maps, but can’t successfully build a social network?

              1. 3

                Social networks live & die by the network effect. Doesn’t matter how good the software is, if everybody and their grandparents is already on Facebook, good luck trying to get them to move off it unless Facebook makes enough terrible mistakes to drive them away (like MySpace once did).

                1. 2

                  Yep, there is 0 friction to switch maps apps so you will use the best one you know. G+ always seemed like better software but no one you know uses it.