Following nature, and an SI experiment

 kw: book reviews, nonfiction, biomimetics, simulated intelligence, evocative images

You may have heard of the Lotus Effect. Nanotexturing on lotus leaves prevents or greatly reduces the ability of water to wet them, such that it beads up and easily runs off. This also makes the leaves self-cleaning. This principle is being used to make self-cleaning windows—imagine if you hire someone to clean house who says, "I don't do windows," and you can answer, "No problem!"

This is one of the examples in the book Biomimetics: How Lessons from Nature can Transform Technology by Brian Clegg. Another is Velcro^®, which was inspired by cockleburs.

When you get right down to it, these two examples are the two primary biomimetic products that have proven to be economical and widely useful. Others have not. For centuries, flying machines were attempted, based on the flapping flight of birds. There are toy ornithopters, but nothing large enough to carry people has become practicable. Frankly, riding a large ornithopter would probably be unpleasant: Surge, surge, surge. Vomitorium country! The first true powered aircraft, by the Wright brothers in 1903, had fixed wings. However, it did use wing-warping for steering, which is based on the flexing of the pinions on birds' wings. Soon, wing-warping was replaced by ailerons (flaps), which are easier to control, and a rudder was added for further control.

If there were no flying animals, no birds, bees (nor flying insects of any kind), or bats, would humans have yearned to fly with such intensity? The idea of flight was based on biology that we see daily, but its implementation has moved far from what biology can produce. Thus, the author makes the point repeatedly in this little book (~150 duodecimo-sized pages), that biomimetic engineering and technology may use a biologically/evolutionarily developed mechanism for inspiration, but must move rather far afield, conceptually, to be realized as a practical product.

For example, robots in fiction, beginning with the Golem of 16th Century folklore, and moving through the robots in R.U.R. by Karel Čapek and the "positronic robots" of Isaac Asimov, all were more or less human-like. Yet practical robots such as those used by the millions in manufacturing, are seldom more than articulated arms of various sizes and other specialized shapes. Human-appearing robots such as Asimo are still pretty much sideshows, although Boston Dynamics is having some success producing humanoid and canoid (doglike) robots. It turns out that, for most uses, a wheeled vehicle is more useful than one with legs and feet. And while a legged robot can traverse terrain that defeats wheels, it appears that flying drones will take over that niche. It's questionable whether a legged robot uses less energy than a drone, except where there would be the need to carry a heavy load (such as rescuing a person). And, as we see in a later chapter of Biomimetics, a self-driving car is a robot, one that has to see and perhaps feel to do its job.

Neural networks, the technology behind "AI" these days, are inspired by the way neurons interconnect in animal brains, but are only glancingly similar. Such a network relies on raising or reducing the strength of each of millions or billions or trillions of connections, either literally wired connections or, much more commonly, simulated in software. "Machine learning" is nearly all accomplished nowadays with software-emulated neural networks. For comparison, the "neural network" we call the cerebral cortex has more than ten thousand trillion connections, and they have more complex action than just "on/off" with various strengths.

Just by the way, I object to the term "artificial intelligence". While the mechanisms are indeed artificial—that is, produce by artifice—they are not intelligent, having no insight or understanding of what they perform. I much prefer the term "simulated intelligence", or SI. These mechanisms simulate activity that is similar to activities of animals, including humans. But we are no closer to producing AGI, or artificial general intelligence, than the Rabbi of Prague was, when he put a slip of paper bearing the name of The Lord into the mouth of a pottery humanoid figure. Well, enough on that for the moment.

This is the first book about biomimetics that I've seen (there are many) that doesn't go all goo-goo-eyed about the subject. Slavishly following nature doesn't produce useful results. For one thing, evolution doesn't fully optimize any of its "designs" (to anthropomorphize a bit). Consider the human body, sometimes called the "pinnacle of creation": an underdesigned back that is prone to slipped disks and muscle spasms; eyes that get myopic when we read or spend too much time doing close work; muscles that require constant exercise to retain their strength and flexibility… On that last point, most other primates don't need constant exercise, not just because they are more active than sedentary westerners; even when an ape is sedentary in a zoo setting (and usually bored out of its mind!), its muscles retain their tone. Somewhere along the way, humans lost the maintenance mojo, so that "use it or lose it" applies to us more than to nearly all other animals!

What a refreshing and informative book! Much recommended.

Now, to add a bonus subject: The image at the head of this article was produced by Dall-E3 from the prompt "An image that captures the essence of biomimetics." It was the best of eight very diverse images based on that prompt. Another by DE3 is shown here.

One great thing about the generative art programs is that most of them give you four images at each try. So, I "hit" DE3 twice and picked two from the eight. Having several other options, I ran the same prompt by the newest, Google's Gemini. Just below are two images, of eight produced, that I particularly liked.

Note that Gemini tends toward photographic realism, and has to be told to be fanciful. It caught the "bio" part, ignoring the "mimetic" part. Also, Gemini images are 1536x1536, and it tells me that "soon" we'll be able to ask for any size up to 2048x2048. I reduced these to half size.

Then, the Playground AI environment has three "engines". This pair of images came via the Stable Diffusion XL engine:

The second image used the Mysterious filter. One can request different aspect ratios, so these are 1024x576, a 16:9 ratio. SDXL also aims at photorealism unless told otherwise.

Now, with the Playground v2 engine:

No filter was used for either of these. All the images from which these were selected had a whitish theme. Lastly, using the Playground v2.5 engine:

This engine is deliberately more "creative" and produces more colorful images than the others. The second image used the Masterpiece filter (Mysterious is only available for SDXL). Whatever goes on with the Masterpiece filter, the result is the least "biomimetic" of the bunch.

Simulated Intelligence can do wonderful things. I suspect it is more free than a human artist would be, because the artist would have a lot of background knowledge that the generative art programs lack.