In my life, I have come across the John Von Neumann’s Theory of Self-Reproducing Automata more than once. A historic work of one of the greatest minds of the 20th century. While interesting to me, I remember passing it over and dismissing it as too theoretical and unworkable in real life, a mere intellectual curiosity. Subsequently while developing my RASA ideas, I never gave his work any thought at all. As time went on, this changed.
My initial ideas were largely formed based on my preconception built from decades of following the development of robotics. Since my first childhood encounters with robots in the 80’s, with the likes of Armatron and the Omnibot toys, I was disappointed. None of these robots were very fun to play with for long. Also, they didn’t seem terribly useful, and didn’t live up to the promise of the science fiction books I devoured. I recall avidly paying attention to any new developments, and being mostly disappointed. This disappointment carried on well into my adult life, and was probably the reason why I was reluctant to enter the otherwise interesting world of hobby robotics.
While robots have had their successes and certainly have their place, the limitations that robots have with interacting with the real world seem here to stay. Advances have been made, but still to this date, we haven’t been able to design a robot that can outperform even the clumsiest of toddlers when it comes to the complexities the world can throw at it. Robots can outperform a toddler in many specific rote tasks, but even a toddler can beat just about any robot on a task as simple as taking verbal instructions to pick up an object and throw it to me. This isn’t because the toddler has faster reflexes than a robot, or better voice recognition, but rather, because the toddler is smarter and more flexible with how he or she faces the world. Intelligence matters.
People are working hard to create smarter machines, and so far, the progress seems limited. Even in theory, there is no generally accepted model of cognition, so most developments have been improvements in specific narrow tasks, such as face recognition. While the idea of machine intelligence is very interesting to me, I haven’t seen any inspiration in that field. Instead, because of my frustrations with the usefulness of the machines we do have, I have focused on the things that machines can do well. After all, why focus on a system’s weakness when there are so many great strengths to choose from.
After a heart operation that involved angioplasty, I was fascinated by the tool the doctor used. Basically, it was a long and narrow tube that was inserted through my groin and threaded into my heart, where a machine at the tip was to inspect my heart and do necessary repairs. Thankfully, it found no repairs that needed doing. In preparation, I did some reading about minimally invasive surgery, and found it interesting that complex machines could be threaded through a small tube and do useful work by remote control. It left me thinking about the problem of the limit of how complex these machines could become and still be able to be threaded through a small tube.
Years later in 2012, while driving through the mountains in, I remembered my thoughts on those, and pondered transportation systems in mountain ranges. In their own way, I saw that roadways, railways, and tunnels through the mountains were a functionally equivalent to laprascopic surgery. I found it interesting how railways snaked through the mountains to give life to complex towns and cities. In laprascopic surgery, devices that snaked through a body to do complex work saved lives. Under normal circumstances, that’s where my daydreaming would have ended.
Numbed by the hours of monotonous highway driving, yet invigorated by several highly caffeinated drinks, I kept on daydreaming about these ideas while I drove. First I imagined a railway hauling shipping containers, and how in larger centres, the logistics of transporting these shipping containers from port to rail to city had already in large part been automated, much like a computer network. Inspired by the idea of remote control promised by minimally invasive surgery, I imagined a remote control railway, doing work in a city or factory without people, or even perhaps, building a city or factory all on their own. By necessity, this took trains that could lay their own tracks, and from here my imagination went wild. Somewhat inspired by obscure Canadian scifi movies, I imagined cube shaped trains, able to lay their own special track as they needed and able to travel in all three dimensions. With this level of self-assembly achieved, I imagined an ecology of other cubic train cars, each one able to perform specific industrial tasks, replacing all aspects of industry. I imagined their uses in forestry, in mining (both strip and underground), and factory assembly. From there, as I imagined more and more uses for this system, I became excited because I realized that once this system reached a certain level of complexity, it could sustain itself with minimal human intervention. Each new machine built for such a system would bring new uses to the owners of the machines, but would also serve to extend the capabilities of the system as a whole. I knew I was on to something good, because the core of the design didn’t rely on improving machine intelligence or flexibility, but rather, focused on the strengths of machines, which is to do a few things very well. At last, simple robots that could do something useful on a large scale. On arrival at my destination, I began work in earnest on my ideas.
In re-reading Von Neumann’s book and refining my thoughts about the RASA system, the relationship between my work and his became apparent. Because of the grid system inherent in RASA, any abstraction or control systems could be easily visualized as a form of three dimensional cellular automation. I have not given the idea of controls much thought, other than that they will be likely be diverse and need to be appropriate for the task at hand. Some tasks can be fully automated and controlled entirely by the unit riding around in the grid. Other machines will need centralized control with varying degree of human input. All of these need to be visualized and modelled. The similarity with the creations that John Von Neumann built entirely out of imaginary tiles is striking enough to me that I must say that he is truly the grandfather of this design. It will be a long time before this design comes even close to being self-reproducing in the way that Von Neumann envisioned, but when it does, I’m sure that it will resemble his work, if not specifically, but as an organic whole. I think a driving force that will help RASA succeed is that it’s become John Von Neumann’s wholly theoretical self-reproducing automata made real. Combined with an architecture that focuses on the useful strengths of machines, simple construction, easily divided piece-meal development, and an economically based evolutionary fitness function, I think this daydream stands a great chance of success. These beliefs are the fire of my passion to continue my work.