Hi Mark, Search for "two armed bandit" and "slime mold" for more info. Here's my paraphrase of the research results, and why it is meaningful.
The experiment: place a slime mold in the center of a microscope slide, and some food to the left, and to the right. These are the two "arms". The slime mold will explore both arms, and then pick one, crawl over there, and "exploit" -- chow down on the food that it finds. This is well-known to chronic gamblers -- you stick coins into multiple slot machines, and try to figure out which one pays off the most. This "costs" you coins. Once you've picked the machine with the best payouts, you devote most or all of your attention to that, thus maximizing your payout. Explore vs. exploit is everywhere in nature: a squirrel needs to burn calories to hunt for nuts. Once it has found the best food source, it can exploit that source. It's everywhere: hunting for petroleum, hunting for a good wife. Anyway, back to slime molds. ... In the experiment, the amount of food is varied on the left and right, the distance from the center is varied, or maybe a small bit of decoy food is placed close-by, diverting attention from a much larger pile of food on the other side. From these variants, you can figure out what algorithm the slime mold are using. They are using the best-possible algorithm that does not require memory. There is a better algo, but it requires one bit of memory. Let's dig a little deeper, now, to continue the story. The cells of a slime mold communicate with one another by emitting small polypeptides -- smells, or odors. They determine what to do next by reacting to the intensity of the smell. Bacteria do the same thing: search for "bacterial signalling". Slime molds move at a few centimeters per 5-10 minutes, because the propagation of smells is by diffusion. You have to wait for the diffusion to happen. There is also lots of cross-talk. -- all cells are emitting these polypeptides, and trying to detect them, at the same time. Like loud chatter at a noisy party. You can do better. Like Jelly-fish. Invent the neuron, use point-to-point communication. Neurons work by emitting and smelling the same kinds of small polypeptides -- but now we call them "neurotransmitters". A neurotransmitter detected at one end of a neuron causes the neuron to fire -- an electromechanical soliton -- travelling at the speed of sound, to a spot a few feet away. Milliseconds. Not minutes. At the other end, some neurotransmitters are emitted. What's a neuron, then? It is a startrek teleporter, a stargate, for small polypeptides. Polypeptide walks in at one end, and a millisecond later, pops out at the other end, a few feet away. Instant. No cross-talk. Point-to-point wiring. Real sci-fi stuff. Neurons can implement algos that slime molds cannot even dream of. A jellyfish is an ultra supergenius, compared to slime mold. For example, jellyfish can not only eat, but they can run away from predators. At the same time. Even while they are trying to flee a predator, they're also trying to stuff their mouths. So maybe not that smart: cannot figure out that maybe you don't need to eat while you're running away. There's a way to fix that, though. Add an extra brain structure to modulate this. This is what bilaterians do. For more details: "Forced moves or good tricks in design space? Landmarks in the evolution of neural mechanisms for action selection", Tony J. Prescott (2007) https://www.academia.edu/30717257/Forced_Moves_or_Good_Tricks_in_Design_Space_Landmarks_in_the_Evolution_of_Neural_Mechanisms_for_Action_Selection This email is too long, so I will skip ten or fifteen more steps in the story. Let me just point out that nicotine is a neurotransmitter, and Phillip-Morris is a long-lived entity, vastly larger and more powerful than any one human, functioning 24x7 to bring you, and millions of others, a shot of neurotransmitter. All those axons and dendrites in your head want a fix, and we have a banking, legal, accounting system called "capitalism" that delivers that fix to you. Amazing what bacterial signalling systems can accomplish, when they set their will into motion. I blogged about this a bit on my blog, but perhaps it's too incoherent, too condensed. -- Linas On Sat, Apr 30, 2022 at 8:34 AM Mark Wigzell <[email protected]> wrote: > Hey Linas, maybe you already know about "intelligent" slime molds? My mum > sent me this video: https://www.pbs.org/video/secret-mind-of-slime-oa3w89/ > Using stop frame photography it documents slime mold solving mazes, > transport path optimization and learning. All this in a single cell, albeit > one with multiple nuclei. > > It reminds me of an animation of a graph traversal algorithm...This > protoplasm is computing a traversal through 2D like one of those red/black > tree searches. Seems like its distributed/parallel processing .... > > Stunning I'd say. I think this kind of intelligence / sentience holds up a > mirror on the GAI approach? > -- Patrick: Are they laughing at us? Sponge Bob: No, Patrick, they are laughing next to us. -- You received this message because you are subscribed to the Google Groups "opencog" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion on the web visit https://groups.google.com/d/msgid/opencog/CAHrUA35MU2VvFbqupEYFLEo0T_q_ym0sibXV%2BcsjvVKjK9CvaA%40mail.gmail.com.
