Slashdot Mirror
Vehicles: Experiments in Synthetic Psychology
Valentino Braitenberg has written one of the cleanest books on robot behavior ever published. It is apparent he wrote exactly what he wanted; no more, no less. The total size of this book is 152 pages, but that seems to be exactly the proper size for the topic he has chosen. Other authors (or editors) would probably say that's not enough pages. It has to be 250, minimum! 400 is better! Not Braitenberg. Vehicles has the raw ideas of a 400-page book. In fact, if you take the proper amount of time to ponder each idea it might even take as long as a 400-page book to get through.
This book contains descriptions of various robots, which Braitenberg calls vehicles since they all use wheels for mobility. They start off simple, then gradually become more complex with each chapter, each new robot being an evolutionary step up from the previous one. In fact, rather than starting with "Chapter 1," Braitenberg starts with "Vehicle 1," and so on. By Vehicle 14 these robots could hardly be said to differ from actual living creatures in the way they behave (though Vehicle 6 describes self-reproducing robots, which is currently beyond our ability to duplicate).
Each new vehicle focuses on an animal behavior: moving, aggression, fear, love and how these can be created in a mechanical vehicle. Braitenberg has a rare mind that can think up original, non-intuitive ideas backed by logic. He also has the ability to present them well. There are a few penalties from Braitenberg's minimalist approach, however. Plain, minimal language can be a bit boring at times, stripping the book of character. Sometimes I like big words and clever turns of phrase that make my mind work, such as the writings of Douglas R. Hofstadter.
How minimal is it? Vehicle 1 contains two pages of text and one page for a diagram. I can just imagine the editor receiving chapter 1 from Braitenberg and saying, "Where's the rest of it?" But it is the perfect length for the simple robot it describes. Vehicle 2 is two pages, plus two pages for two diagrams, and so on. Honestly, for the first four chapters a 12-year-old could read this book and get the same from it as a university professor. His minimalism is admirable, however at times it can feel maddeningly incomplete.
Vehicle 5 (logic) begins by explaining a system of inhibitors that can build a thinking machine. What he is really explaining is the basis for a neural net, however he attempts to do it in five pages. Are five pages enough to explain a neural net? Unfortunately, No. This seemingly simplistic approach actually means he is leaving out vital parts of the explanation that prohibit complete understanding. More description in this chapter would be incredibly helpful. He doesn't talk enough about how the "pulses" given to the neural network gates add up. Is there a cumulative effect going on? After a 1-paragraph explanation he shows 2 examples and describes what they do, but unfortunately he doesn't explain them enough for me to understand the mechanism. Thankfully instances like this are rare, and Vehicle 5 was the only description lacking.
Vehicle 6 describes chance and the role it plays in natural selection. He describes chance as "a source of intelligence that is much more powerful than any engineering mind." Never before have I directly thought of natural selection as being intelligent, but once Braitenberg said it, it sunk in that, Yes, natural selection is intelligent; much more intelligent than any human who ever lived. It is the most skilled engineer ever, making machines of unbelievable complexity and ability. And this "intelligence" has no form, no body. It has always been around since life began and it will always be around until the universe ceases to exist. It is a process; an invisible concept. And yet it is more intelligent than any human.
Artificial Intelligence authors often state the importance of language and symbols, but one can't help but notice that animals seem to do fine without language. And aren't animals intelligent too? He demonstrates that we always assume because an animal reacts a certain way towards an object it must store a symbolic representation of this object. That seems to be reasonable, but Braitenberg demonstrates you can get what appears to be symbolic thought when in fact internally there is no symbol stored -- just electronic paths. It causes one to rethink some well-entrenched ideas about AI. What about meditation? I know when I'm in a meditative state (not thinking/using language) I can perform some actions like sweeping, making food, walking, etc.. So just how important are symbols? Is there a limit to the thoughts that can occur without symbols? I don't think this demolishes the importance of symbols -- likely they are needed to create new ideas -- but they might have their place, one less central than we generally suppose.
At the heart of each vehicle are the pathways that the wires make as they connect sensors to motors. The robots in the first 2 chapters consist of a few sensors, a few motors, and a few wires connecting them. There are no CPUs in any of the robots, except for when the wire connections become so complex, embodying logic, that they effectively become CPUs themselves. The later chapters get into concepts that would not be as easy to replicate in actual robots, and rely a little more on speculation than hard fact. He addresses such difficult topics as getting ideas and having trains of thought. Most of the robots, up to perhaps Vehicle 9 (excluding the evolutionary vehicle) could likely be built in reality. With the recent advent of Lego Mindstorms, the perfect canvas exists to create these types of simple robots, and a programming environment like leJOS Java would make it possible to simulate the wiring described in the book. Maybe someone will eventually recreate the Vehicles in the book using these tools.
The book also includes imaginative artwork of the robots, done in a thought-provoking, abstract style. Unfortunately, rather than interspersing them throughout the book at the appropriate chapter, the editors have placed them all at the end of the book, where they are ineffectual. By the time you get to them, you've either forgotten the thrust of the robot described in the chapter or have mulled over the robot enough already. Having these pictures within each chapter would give the reader something to look at while pondering the meaning of these robots.
So what is this book really about? Well, everyone who reads it probably has his or her own opinion. Braitenberg himself calls it a fantasy with roots in science. I think it is partly about our own origins through evolution, and how something as complex as the human mind might have got started. It's also a bit of a roadmap as to how we might be able to construct our own complex, thinking machines. Braitenberg is laying out no less than the evolution of our brain. For people interested in these topics, he uses his vehicles to construct another metaphor with which to study Darwinian evolution.
Braitenberg includes a section at the end of the book titled "Biological Notes on the Vehicles." These describe the concepts of his robots and how they relate to actual observations in biological creatures. As a scientist, he has done a world of research into brains. I've read his previous book, On the texture of brains: an introduction to neuroanatomy for the cybernetically minded. Though not a popular book, it is evident he is very meticulous in his research. He has dissected and examined fly neurons under microscope for weeks at a time, and from this work, as his mind pondered what he was seeing, came the realizations described in Vehicles. It's quite a treat to read the results of his thoughts without having to do the tedious work yourself! It all adds up to Braitenberg's startling conclusion (which he states at the beginning): The complex behavior we see exhibited by thinking creatures is probably generated by relatively simple mechanisms.
You can purchase Vehicles from bn.com. Slashdot welcomes readers' book reviews -- to see your own review here, read the book review guidelines, then visit the submission page.
"By Vehicle 14 these robots could hardly be said to differ from actual living creatures" As Kraftwerk puts it, "we are the robots".
He tried to kill me with a forklift!
Does anyone really want to compete with a robot for space and a date?
So Vehicle 14 is two pages of text, plus 8192 pages for 8192 diagrams?
I'm going to disagree and say that a process is a process, and intelligence is something different.
Natural Selection is an elegent process and can (for lack of a better word) craft some exquisitely designed things. Trees, eagles, mosquitos, and even humans are all engineering marvels created in the forge of Natural Selection. But there is no intelligence behind it.
If Natural Selection were intelligent then the dinosaurs would not be extinct, nor would the miryad of complex and promising creatures of the Edicarian Fauna. Intelligent design would not waste such potential sources of design diversity.
Even crystals are beautifully "designed". They are pretty to look at, serve useful functions, and can be highly prized as art, or jewelry. But the crystalization process is merely a result of natural chemical forces in action. No intelligence behind that, or natural selection either.
If the reviewer wants to suggest that Braitenberg is implying that "God is in the details," he can. But a process is a process, and chance is not intelligent design.
I heard about this book at the first conference on Artificial Life at Los Alamos in 1988. Maybe I'm out of the loop, but I don't hear Braitenberg's work discussed as much as it deserves to be. The core concept, that great complexity can arise from the interaction of simple systems, is also demonstrated by Cellular Automata, but "Vehicles" has a beauty and simplicity that makes it a classic.
/.ers)
I think breakthroughs in AI will probably come from people who are familiar with physiology (especially biophysics), or some new branch of mathematics. So many theoreticians from cognitive science, computer science, psychology, and psychiatry ignore physiology. I can't blame them, I suppose -- the field is unbelievably complex.
In any case, "Vehicles" should be required reading for anyone aspiring to have a degree in systems, human or otherwise (and that includes
Braitenberg has written a 152 page book describing robots that he has thought about creating, using minimalist language and half-explanations which lack necessary detail. As you stated, the first several chapters are only a few pages long and can be understood by a 12-year-old, and the systems "described" in the later chapters may not be possible to implement.
Brooks, by comparison, has created REAL robots which do REAL work and he (and his graduate students) publish detailed papers which explain their methodology, technique and results in detail.
Compared to Braitenberg's book of "what if..." ideas, I suppose Brooks' approach is novel.
Thousands of geeks were seen scratching their heads, seemingly in deep thought, almost perplexed. What is this supposed to mean anyway??
"And we have seen and do testify that the Father sent the Son to be the Savior of the World"
1 John 4:14
Of course animals think.. You don't think dogs were actually born with the instinct to SIT do you? And even though many animal behaviourists consider this "learned behaivior", they have to think to learn something. Sorry about my spelling...
"And we have seen and do testify that the Father sent the Son to be the Savior of the World"
1 John 4:14
One: there's no detail in this review. It sounds like the author is suggesting behaviourism (cf. Skinner) as a theory of cognition, an idea that was discarded before I was born. Someone give me some details and prove me wrong.
Two: What is being said here that Simon hasn't already said in his essays on complexity of behaviour well before this book was published? In otherwords, even in 1986, is this really new?
Although it is interesting that he describes a neural network here: it is clear to me that the reason the description is so shrouded is because prior to 1989, ANN's were taboo in the literature (Minsky having ripped perceptrons to peices back in the 60s).
> probably deserves more recognition for this
> train of thought than the much more publicized Brooks.
Brooks teaches the Embodied Intelligence course at MIT (which I took two years ago). One of the first things the course covers are Braitenberg's creatures (see the syllabus). So while Brooks may certainly get more air-time than Braitenberg, he certainly gives credit where credit is due. .. but then, remember that Braitenberg focused on astoundingly simple circuits that lead to interesting-appearing behavior, whereas Brooks has used his approach to build working autonomous robots...
Actually the comment about self reproducing robots in this review is incorrect. Please check out:
3 238.stm
Robot learns to reproduce
http://news.bbc.co.uk/1/hi/sci/tech/90
Imagine a philosopher with no practical experience of anything vaguely robotic wrote a book on robotics. This is what they would write. Braitenberg talks about vague concepts like memory, foresight, logic and trains of thought. But these discussions are completely sophomoric jumping from systems with one or two neurons to imaginary systems with the above properties. I don't need a book to point out that an intelligent machine needs foresight, and I don't need a book to point out that a simple neuronal system with persistence might have something to do with memory. Unless you're going to say something about the details between neurons and full blown brains then you're just armchair philosophizing and any sophomore can do that without the help of a book. Maybe if he had written the book in the forties it'd be interesting. But by the eighties every science fiction writer and his dog had written about these subjects with far more detail.
But I do love the pictures by Ladina Ribi and Claudia-Martin-Schubert. They are quite special.
Doesn't it make you feel good to know that our freedoms are protected by politicans, lawyers and journalists.
The idea that the human brain is not logical isn't new. Psychologists and AI theorists have long known that there are several modes used in human reason, and ANalogic reasoning is used 99% of the time. We don't stop and deduce the correct answer, we leap to a similar situation in memory, and adjust the outcome of that situation to this one. Incidentally, this is why we remember events and associate pleasure/pain responses with them. This makes the choice of whether or not to bite your fingers off rather easy... for most of us.
The first "behavior-based robots" along those lines go back to 1948, with the work of W. Grey Walters. Those little wheeled robots did much of what Braitenberg talks about with his earlier models. And, since Walters actually built them, he discovered behaviors that weren't obvious just thinking about it. If you're into this at all, read everything you can find about Walters "Turtles". They were shown, working, in a museum for a year in the 1940s, and modern replicas have been built. Walters was decades ahead of his time.
There was considerable thinking along those lines in the 1950s, most of which didn't go anywhere. I have some old AI books that contain similar speculations, although they're far less readable than "Vehicles".
The basic problem with model-less behavior-based robotics, as Brooks and his followers have discovered, is that the ceiling is low. You can can get some simple insect-like behaviors without much trouble, but then progress stalls. That's why Brooks' best work was back in the 1980s. The robot insects were great; the humanoid torso Cog is an embarassment. This is typical of AI; somebody has a good idea and then thinks that strong AI is right around the corner.
If you have a Lego Mindstorms set, you can build many of the "Vehicles". They're kind of cute, but don't do much.
No, when stepping on a dogs tail, more has to do with attitude, than trained behavior, or instincs... Which is why some dogs bite you, some growl, and some just sit there. Besides dogs, many gorrilas have been taught to use sign language, and this probably requires thinking. IF you would go that far, to say it isn't thinking, but input-output, what is thinking, and do humans really think either?
"And we have seen and do testify that the Father sent the Son to be the Savior of the World"
1 John 4:14
Yeah, that's the problem, nobody really understands, and those who claim to understand all disagree with each other. m-w.com has about 2 pages of definition on the word "think" and about 20 different meanings.
"And we have seen and do testify that the Father sent the Son to be the Savior of the World"
1 John 4:14
This is a book I enjoyed greatly, and that gave me some sort of insight to many problems, most notably debugging software...
In Murphy We Turst
My take on all this stuff is that it's a contrast to Kant. For Kant, the world around us is a bunch of unknowable abstract objects, which we 'know' through our flawed senses. ("Ah, the abstract 'pen' probably exists, but I can only know what my imperfect senses tell me about it.") This is more like the robotic systems that create an abstract construct of the environment and then internally work with that abstracted construct.
As I read Heidegger, he's saying that, yeah, Kant has a point, but it's not very useful in day to day life. When you walk through a door, you don't think about the doorknob, you just turn it, open the door and walk through. It's all what he calls "taken for granted." You don't stand there thinking "Hmm, maybe my perception of the doorknob is flawed, and there is no knob. I can never be sure" (well, some of us have thought thoughts like that, but only after consuming certain molecules).
Essentially, Heidegger's take is much more practical: how do we do the useful everyday stuff? This is a lot more like robotic systems that are based on more reflexive responses.
Yes, Heidegger deals with lots and lots of other stuff ("Language is the house of being" "Death fractures the taken for granted", and the scary stuff about how when you are speaking old German you are more truly in touch with existance!) But the underpinnings of phenomenology is potentially really useful for understanding the "nuts and bolts" of interacting with the world. Oh, and he's the "Velvet Underground" of twentyth century thought (Sartre, Derrida, etc, etc cite him as a critical influence).For kit hardware that implements the ideas of Braitenberg et. al., check out the BYO-bot. I've had one for several years, and they're a great classroom demo.