AI Musings 6/8/2021

Updates on AI rabbit holes, my reading list, and random thoughts that barely make any sense, even to me…

Knowledge Graphs, Sparsity, and the One Format To Rule Them All

  • I last left off talking about George Lakoff and embodiment, and I added a reading list below that reflects my interest in that. Learning about navigation and movement outside of pure AI fields is essential. Mainly because knowledge and language does feel inherently spatial, as well as the way our languages operate.

    Also, I feel like our first language- a common language with all animals, is that of movement: Body language. Ever meet an animal in the wild and communicate in broken interspecies pantomime? Well, there you go…
  • It seems that the missing link in my search for a way to compare things for my NPC AI was Sparse Representations/Sparse Coding. It’s a rather flexible format that you can encode just about any value for. Some good resources for this are found on the HTM School channel on YouTube (RIP Matt). Another one is regarding’s Semantic Folding, which is pretty snazzy.

    Being able to encode a spatial organization for information seems like the next frontier for this particular rabbit hole. And I’m specifically interested in ways to represent data spatially that the AI agent can then interact with. That would allow for an AI to have some rudimentary level of embodiment, even if it were artificial.
  • Interrogative is going to have to be brushed off and updated with what I’ve learned since I hit that wall some years ago…


  • A lot of the reading I’ve been doing below has me thinking about spiritualism and shifting cognitive loads to the world around us. Many cultures named places descriptively, which aids in wayfinding. Yet, we can stuff all of that into a GPS and then subtract from your ability to wayfind by shifting your attention from the path to the product. I’ve since turned my GPS off, except for a short check of traffic in the morning. Aside from that, I’ve gone back to driving as I did in the 90s.
  • Sorry, several-AI-startups, you can’t tell emotions by looking at faces. If you could, then that whole side industry of reading body language and convincing people would evaporate over night.
  • Adding to the above, it’s probably a good idea also to think about if interacting with your product is going to “flatten” people’s behavior into something that the device can understand. In trying to communicate with a device, we may inadvertently limit ourselves habitually, losing some of ourselves in the process. That an AI cannot understand us is not bad- other people cannot understand each other regularly, and that sparks communication, not self-limitation.
  • I can’t wait to get to some conferences.

What I’ve been reading…

If you’re a parent with small kids, then you know that short of packing up and leaving your family behind, there’s very little normal reading you can get done. So, I finally discovered Audible, and now I get to listen to books randomly in the car or while working. And that’s great because even if I did read, I’m slow as hell. So what have I “read” this year so far? Glad you asked:

  • A Thousand Brains by Jeff Hawkins: Sparse Representations for the win.
  • Louder Than Words by Benjamin K. Bergen.
  • Seven and a Half Lessons About the Brain by Lisa Feldman Barrett: I picked this up almost reflexively because I loved her book on emotions, and this was just as good.
  • Wayfinding by M.R. O’Connor: Spatial navigation is inherent in embodiment, because animals need to get around, and spatial navigation is key to memories, intelligence, and our health. A lot of that is laid out here, as well as a ton of detail on indigenous navigation and language usage.
  • Rethinking Consciousness by Michael S. A. Grazino: Goes on about attention, which is the point, and how it may lead to consciousness.
  • The Spike by Mark Humphries: Really nice tour of the brain, and how confusing its operations are, despite all of our progress.
  • The Forgetting Machine by Rodrigo Quian Quiroga: Made a pretty solid case that you don’t want the kind of memory where you never forget anything and remember the minute details.
  • From Here to There: The Art and Science of Finding and Losing Our Way by Michael Bond: Another great book on spatial navigation, intelligence, and effects on our mental health. Seems like a lot of overlap with Wayfinding. There’s a lot of complementary information in it, though, and they really should be read together.

That’s quite a bit- I find it’s a lot easier to get through books using Audible. The obvious downsides are that I can’t write notes in the book or highlight and annotate like with Kindle. I can bookmark sections, but that’s not the same thing, and even that is hard to do when working or driving. Still, better than nothing, and faster than I read anyway.

Leave a Reply