Needless to say, that's how John Long's Darwin's Devices: What Evolving Robots Can Teach Us About the History of Life and the Future of Technology got into my hands. This can be seen as a rather odd book, as Long is a biologist (he's Chair of the Department of Biology at Vassar), although he's also a professor of “cognitive science”, which I suppose does get one into at least the neighborhood of robots. He was a PhD candidate who got enticed into studying the backbones of marlins – a fish that is incapable of surviving in captivity, so can't simply be brought into the lab. In Hawaii, he was able to obtain the backbones of recently processed marlins, and was able to study in detail the fine structure and various motions achievable via that bit of organic architecture, but this wasn't something that could be functionally tested as the backbone was missing the marlin.
One of the most interesting factors of Darwin's Devices is it's very much a “science book”, no so much a “popular presentation” of the subject matter, but a tracking through the process of “doing science”, including discovering one's big errors, bad assumptions, experimental challenges, and limits caused by both funding and available technology. Early on here, the author frames what's coming with:
While modeling offers a lot of flexibility as far as how/what you're looking into he notes “... we always have to make, even in the most accurate models, many simplifying assumptions. The trick is to make the right ones.”. His initial “capstone” to his doctoral research involved a computer model of the dynamics of the marlin backbone in action. One can model in either the computer or in a physical device, but, as he's reminded “every computer model is doomed to succeed”, and his had the unfortunate factor of violating the laws of physics (in this case, the 2nd Law of Thermodynamics), something that did not faze the computational environment at all … he subsequently outlines why it's often more useful to go with a physical model: “If an engineer's design violates the laws of physics, the machine won't go on forever: instead, it just won't go.”At this point the best model of a marlin backbone is not a marlin backbone. Because we couldn't study it any further in the living fish, we were left with three choices. One: quit and do another project. As depressing as that sounds, sometimes it is the only practical alternative. In the hopes of finding a species that works really well for answering a ton of different questions (which would make it a “model organism”), switching species is a common response. Two: try to build a new instrument or experimental procedure to answer the question. For the stubborn and electromechanically minded, this is often a way to work out your frustrations and keep busy while you come to grips with the fact that you really, truly are stuck. Three: build a model of your fish. For those of us who need to keep writing papers so that we can earn tenure and win research grants, this is the way to go – we model.
At this point Long goes into a discussion of the surprisingly wide array of backbones and related structures (notochords, etc.), various of which appear to have evolved independently in a number of different phylogenetic lines. This sets up the choices made for the first physical model, the Tadro (shortened form of “tadpole robots”), which “are based on the tadpole-shaped larvae of sea squirt chordates”, each having “for its axial skeleton a notochord of differing stiffness”, the stiffness controlling the swimming performance of the model, and which is genetically coded, allowing that variable to evolve from one generation to the next.
Here the author goes into a bunch of technical detail about natural selection, and how traits will change in a population across generations … even getting into some delightfully obscure (to me) mathematical short-hand such as “delta x-bar equals delta p”, which indicates how genes relate to phenotype, and logical formations such as ceteris paribus, a Latin phrase meaning “all else being equal”, which is the method by which “we isolate one variable and understand how it influences the whole system”.
The Tadro model went through various stages until they had Tadro3, which was a simplified system (basically a small computer in a bowl) which, like its tadpole-ish larva predecessors, responded to light, and whose tail stiffness could be varied (the stiffness standing in for vertebrae). The “success”, evolutionarily speaking, was the Tadros navigating to a light source, which was its “food”. Through a number of equations, the ability to do so defined the “fitness”, and so determined what particulars the next “generation” would exhibit.
Here the book wanders into a look at robotics and “intelligence”, noting that the evolution of these robots involved “embodied intelligence”, each generation got more efficient via optimizing chordate stiffness, not getting any “smarter” except in a body sense (the entire program that ran these is reproduced here, and it's only about 50 lines of code). There's a reasonably detailed look at the competing intelligence theories of Alan Turning and John Searle, and how these different stances can create dramatically divergent ways of considering what's happening with the robots. This then leads into a thread about the work of a number of neuroscientists, whose research points to yet another whole “world” in which the Tadros operate (and the author does admit – even celebrates – the confusion inherent in these different cognitive contexts).
There is a LOT of material being backgrounded in these sections – with discussions of if a “brain” (what one MIT professor calls a “cognition box”) is really necessary, when a palette of “reflexes” might be as functional, or even more so. Various versions of these frames are charted out as both organic and electronic diagrams, and reduced into some more Greek-abbreviated mathematical formulas. It's all fascinating (and not oppressive) in context of the read, but a bit complex to summarize in this review.
While not evolving per se, the Tadro3 gets supplanted by the Tadro4, which is equipped to model predator avoidance. It ends up with two light sensors (to better determine direction), and “an infrared proximity detector” which is designed to some extent mimic the “lateral line” of sensing cells on fish. One of the other interesting “sciency” things here is that one of the factors that they'd set up to determine “fitness” in the Tadro3 turned out to be messing up the data. They had decided that “body wobble” was a negative, but discovered that penalizing for wobble ended up degrading the feeding efficiency … as it was “functionally dependent of swimming speed”, and the faster moving units were exhibiting more wobble, but could maneuver better. There are various tables and charts looking at how they processed this info, but it stands out as big “oops”, and a cautionary tale of how one's initial assumptions when setting up models need to be very carefully considered!
Another significant change in the Tadro4 was the addition of “vertebrae”. They took the gel-based notochord of the Tadro3, made it a consistent stiffness and length, and added bead-like vertebrae … as the other elements are constant, the flexibility of the “backbone” was only determined by the width of the “intervertebral joints”, which was variable in relation to the number of vertebrae placed on it (more vertebrae, less joint space, stiffer spine). They also made two versions of the Tadro4, an evolving “prey” unit, and a non-evolving “predator”. The Tadro4 was modeled on a different type of critter, an early (400 million years old) jawless vertebrate fish, Drepanaspis. Having multiple sensors allowed the team to test for the relation of sensory systems and vertebrae, with the hypothesis that having the sensory system (to determine the presence of predators) would spur the development of (propulsion-enhancing) vertebrae.
It pains me to do so, but at this juncture I'm going to throw my hands up and say “too much stuff – can't summarize it!” … the author bounces around between some very technical evolutionary theorizing, overviews of the experiments his team did, and charting out “adaptive landscapes” (which, short of scanning and including those graphics in here, are kind of hard to describe). He also shifts from the development of the tail-mobile Tadros, and into an “ET” (Evolutionary Trekker) called Madeleine, which has four flippers … and is named for its vague similarity in shape to that small French pastry. This takes side trips off into considering Plesiosaurs, and aquatic vs. terrestrial tetrapods (where there are 1,679,616 possible different mobility options … needless to say, only a tiny fraction of those being tested with models).
It's at this point that a lot of the action shifts from the college lab to the R&D centers of various robotics companies … and ultimately off into the acronym-laden world of DARPA and military applications of robotics. However, it's hardly just our folks looking at this … he quotes an expert in the field as saying that at least fifty-six countries are developing robotic weapons. He quotes an associate as saying that military robots should be “unmanned, expendable, and cause maximum damage”, and gives an example of something called the MicroHunter which is a palm-sized torpedo-like vehicle, with just one moving part – the propeller. These were tested against a SEAL diver, and the SEAL was only able to stop these from hitting the target 50% of the time (they had otherwise been getting 100% marks) – and that was with just four in play. The book ends up with a “philosophical” look at how to manage this sort of technology, but with a “SkyNet” dystopian vibe hanging over it all.
As noted, Darwin's Devices is still in print, and the on-line big boys seem to have it a full cover price. However, having gotten into the Dollar Stores, “good” copies are available from the new/used guys for a penny plus shipping, and “new” copies can be had for under a buck (plus shipping). Again, this isn't exactly one of those “popular science” books, as it's more focused on the experimental/research/theory aspects than most of those would be … which is one of the reasons I'm looking to pass this along to my robotics-obsessed (she's currently off developing an aerial mapping drone on a summer internship!) engineering student daughter … but it might be a bit overwhelming for some (I'll admit that I got a bit lost at a few points here). It is, however, a fascinating look at a line of research, with all the complexities involved in that, with an over-all arc which charts out the (somewhat disturbing) development of this sort of robotic system. A definite recommendation for all science/engineering geeks out there (others' “mileage may vary” on how you'd like this).