I’ve been reading a lot about cephalopods such as octopus and squid (annotated bibliography at end). I’ve also been taking machine learning classes, learning about artificial intelligence. These concepts swirled around my brain, making marbleized patterns until, in an instant of “aha!”, I had a thought that combined them.
In machine learning, most things are based on neural networks. They work sort of like our brains do, with many inputs going into a neuron (a little bit of math in software terms, a little bit of goo in biologic ones) and then the neurons have outputs that get combined with more neurons until somehow we build up enough layers to get information out of our messy input data.
The hard part is training these neural networks so they learn what we want them to. Training (usually) happens before we interact with the software neural nets, often involving large computers and thousands of data samples. When we use the trained neural net (in software or in our brain), it is called inference.
It makes sense that training is computationally intensive since our brains take a long time to train as well. It takes years for us to learn how to use our bodies, how to interact with the world, and how to live independently. We require extensive practice and supervised help before we are trained enough to be functional. The comparisons between the human brain and the computer neural net are easy.
Where does the octopus fit in? Well, the octopus is remarkably smart, especially for an invertebrate. Also, octopus lifespans are incredibly short for a creature with such a large brain/body mass ratio. Brains are biologically expensive to grow. Why are they so smart but only live a year or two? (Hey, cuttlefish cuddlers and squid squeezers! I’m not dissing your beloveds, all of this applies to them as well.)
What if they don’t have to learn? What if most of their neural net training is done in evolutionary time instead of training in the present as our brains are? What if octopus brains are all inference, trained by natural selection, while our brains require years of training then years of training and even more inference? Oh, don’t get me wrong, an octopus can learn but they are far more equipped at birth then we are, even through their environments are wildly complex and varied.
I think most octopus brains are a single instance neural net shaped by evolution where the mammalian brain is a general purpose neural net, with weights that are only partially filled at birth. We get a lot of space to change our minds. Put another way, while an octopus can learn and adapt to its environment, there are no big firmware updates for the cephalopod brain.
Once you start thinking about firmware and embedded systems, the octopus as a whole makes a lot of sense. They have some incredible sensing technology; their eyes are amazingly like human eyes but it is an entirely convergent evolution (which is to say that every creature had to make their own version of these sensors, just like in embedded systems). Octopus arms can act on their own and are only partially controlled from their central brain; distributed control for the win!
Given you are reading this from Embedded.fm, it should come as no shock that I like embedded systems. I’ll say now that I also really like cephalopods. Like embedded devices, they are purpose built for their application and resource-constrained in ways our general purpose computer brains aren’t. That doesn’t mean that octopus are less complicated or less capable, just different, some might say gloriously alien.
The Soul of an Octopus: A Surprising Exploration into the Wonder of Consciousness by Sy Montgomery. This book is quite excellent and was one of the first to get me hooked on the octopus. (Though I didn't care for the author's other work, The Good Good Pig.)
Octopus: The Ocean's Intelligent Invertebrate by Anderson, Roland C., Jennifer A. Mather, James B. Wood: Wonderful. Definitely not too much science, nope, not at all. This was in-depth and very interesting.
Other Minds: The Octopus, the Sea, and the Deep Origins of Consciousness by Godfrey-Smith, Peter: Good, a little too much philosophy but I like the parts about the octopus. His Scientific American article is good too.
Walking Your Octopus: A Guidebook to the Domesticated Cephalopod by Kesinger, Brian: Not useful for this post, included for completeness. Very pretty.
Kraken: The Curious, Exciting, and Slightly Disturbing Science of Squid by Wendy Williams: Approachable and entertaining.
The Search for the Giant Squid: The Biology and Mythology of the World's Most Elusive Sea Creature by Richard Ellis: Good but dry in spots.
Spirals in Time: The Secret Life and Curious Afterlife of Seashells by Helen Scales: Wonderful book about about molluscs of all varieties, including cephalopods, very good read.
Mapping the Deep: The Extraordinary Story of Ocean Science by Kunzig, Robert: Interesting history and science of the ocean, dry in spots.
Spineless: The Science of Jellyfish and the Art of Growing a Backbone by Juli Berwald: Good but needed more jellyfish and less about the author.
Voyage of the Turtle: In Pursuit of the Earth's Last Dinosaur by Carl Safina. Ok, this is not at all related to octopus or their brains but if you got all the way here, you may like it (I do). Though you may swear off eating fish.