Everything is prediction. Compression is truth. Teleonomy is the new teleology. I’m working on wondermentation. It is of arguable utility to create pithy little epigrams and nonce phrases as markers to different phases of one’s life, but they began to accumulate as graduate school ground down towards a soft landing at Stanford. My studies and research started to get lively towards the end of my undergrad degree with an assistanceship in the Advanced Computing Laboratory. Machine learning and evolutionary computation were my favored areas of interest and I supported my core studies with evolutionary biology, ethology, analytic philosophy and mathematics.
I felt I had crossed a Rubicon late in my senior year at Cornell as I worked on a fundamental challenge in learning patterns directly from data—so-called unsupervised learning and knowledge acquisition. The problem posed as a kind of Manichaean mystery to me, divided between treating every single data point as a singularity and similarly considering them all as a unified whole. Between the two poles was compromise meted out by co-occurrence priorities; events close together in time and space deserved capture as a statistical regularity.
The threshold question was what form that acquisition algorithm could take on that would lead to an efficient coding of the data into a predictive model. The answer was found in an elliptical foray through the fundamentals of mathematics and computing, then straight into the heart of evolutionary thinking. I did not really emerge from it, either. There was a small eureka moment with a gradual fading of interest as summer hit and I was back in Santa Fe after graduating, waiting for my Masters program to kick-off. It stayed with me and I carried a small notebook around, feverishly scribbling notes while once again wandering up those arroyos towards the ruddy canyons above.
Conceptually, we first needed a predictive machine framework. The best choice was a Turing machine capable of universal computation that distills all the commonsense notions of what a computer can do down into a neat mathematical concept. Since we can simulate a Turing machine on cellular automata, the exact form of the machine was not a critical issue—it remained as an abstraction of capacities for basic ratiocination. We then needed a bunch of data. We could simplify that down to a string of 1s and 0s, the crypto-language of computers. We could also write out the structure of a Turing machine as a similar stream since it is just a logical computing system as well. Now we ask a fundamental question: what is the shortest (coded) Turing machine that can generate the data? If there are two machines that have different lengths, then the shorter one is the one that is the most efficient generator of that data in a certain sense. Turning that around, the shortest machine that can drink in a sequence and spit out a conclusion as to whether the sequence is a member of a group of such sequences is the optimal machine for determining the sequence and, in a sense, for guessing the next bit in the sequence if need be. In fact, the smallest machine is the least likely to make errors in that prediction game moving beyond the training sequence, and that is a stunning realization that is only about 50 years old, though there were hints of it in philosophical ideas like Occam’s Razor and parsimony.
Everything is Prediction. Compression is truth.
Prediction is central to living. Those who outpredict you win the game. And the most compact predictive model that best explains the data wins. But there is a problem with the mathematics: there is no logical procedure for finding the simplest machine. You have to guess. But guessing about solutions to difficult problems is exactly what evolution does. Our children are our guesses about survival. Producing random variants of machines and test them, the better ones carry on to become the next generation of solutions.
The puzzle was filling in and those epigrammatic phrases started to feel like markers of understanding. I felt I understood the subtle joy that Korporlik had shown when his cellular automata twinkled down the screen, his grinning face washed by the cold cathode incandescence.
My Master’s thesis solidified around this topic and I began developing evolutionary frameworks that had remarkable properties. One could read a short text and then generate similar texts that looked increasingly realistic though with limited meaning associated with the productions. The system was building syntactic trees from the statistics of letters and words, then varying them to look for shorter, more parsimonious explanatory models. The size of the problems and data sets was growing, as well, as I moved into my PhD thesis, and I began experimenting with a fellow student’s toolkit for implementing parallel swarm solutions using the special chips for graphics processing in video game systems.
By the time I graduated, I was running millions of parallel simulations at once and had managed to develop a model of earthquake prediction that had a forty-five percent chance of success. I published, graduated and moved on to continue my work as a junior research associate back at the Rio Grande Group. It had been a hard decision among many options, but I wanted back to the southwest, back to Santa Fe. Korporlik was still there and had interviewed me for the position. I tried out “Compression is Truth” on him and he seemed genuinely confused as to why I found that concept interesting.
“I suppose there is an element of truth approximation with respect to problems of inductive inference, but the traditional notion of truth is formulated around the satisfiability of deductive statements, analytical and synthetic alike, no?”
Off guard, I scrambled a bit to regain my footing, “But few of those statements are important,” I ventured. “All new knowledge arrives via induction.”
“I don’t think so. Initial observations are treated with induction to build a basic model, but the reorganization of that model is subject to deductive constraints, no?”
“Right, but that is precisely the evolutionary epistemology working to sort between the candidate models and confirm the deductive, eh, ramifications of one paradigm versus another.”
“I suppose so, but then shouldn’t your statement be much longer and more precise? Unless,” his eyebrows shot up and he realized it was a linguistic joke, “unless, yes, that is the joke?”
“Alright, now tell me what you want to work on while here at RGG.”
“I want to carry forward with the existing effort to better develop the hybrid evolutionary learning methods with this informational physics constraining model formation. But I really need to scale up to billions of algorithmic entities, each with a scale of between ten and one hundred billion computing elements. At that scale, I think it is possible that a new level of learning and intelligence might arise.”
“But that is an enormous scale, how can you do this?”
“I don’t yet know. There is the internet itself, but it isn’t even large enough. There is also some promising work in quantum computing that my classmate, Anil Freeman, is working on. He’s at CLN in Boston, now, and working on quantum encryption, but he thinks that quantum computing can obtain the kind of computing densities that I need. Nanomachines are another possibility. Some of this can be done through simulation and estimation right now, though.”
“I think this is worth pursuing and I will make my recommendation to the board.”
I thanked him and was moving back in my battered Subaru within two weeks. I had friends and a few girlfriends through my academic career, though my passion for my research seemed to be perceived as a bit too boring for the girls I had been involved with. By the time my doctorate was done, I was ready to move on and had few ties in Palo Alto. Many of my classmates moved into industry in Silicon Valley and thought I was a bit odd to be interested in joining a research think tank, but at least one of my friends was moving to an academic position and thought it made sense for me to return to the quiet of the desert where there is adequate time to just think deeply about subjects that mattered in some broader sense than serving a commercial interest.
I settled in with Mom initially to save money so I could pay down some of my college debt. It was odd but comforting to be back in the old adobe and I tried to help out with housework and errands. I really didn’t need to be in the office very often, but could work from wherever it was comfortable, though I would often make an appearance just to make sure the senior personnel knew I was busy.
My world mostly consisted of my own thoughts comingled with the massive stacks of research papers I poured over during this period. I developed a research proposal with Korporlik that built on my seismic prediction work and was quickly funded for several years under a young investigator award.
The topic of Harry occasionally arose over dinners and, especially, at holidays. Mom would call him now and again and then tell me as best she could what his life was like. He and Sarah had married and now had two infant girls. They lived in a compound in Nebraska with other members of their church. The very term “compound” bespoke cults and danger to me so I downplayed it by suggesting it was probably just a neighborhood. He refused to come home because his ministry was too valuable and he didn’t want his family infected with our secular values, according to Mom, who was both amused by the concept and a bit sorrowful at being excluded from her son’s life. The protests and arrests continued. He had served six months in a Florida jail for trespassing at Cape Canaveral, though it was unclear what his concern was with NASA. He had served another two months in Colorado. She had found his blog and read through it and been concerned with the scattered thinking she thought she saw in it. Harry had become anti-government because he felt the government was stealing from people with taxes to fund scientific research and healthcare he didn’t sanction. He called the government fascist and called on his followers to help bring down the fascist regime and bring about a new era of Christian governance.
I avoided looking at the blog, fearful of what I might find. I realized, though, that my federal grant for investigating evolutionary simulations might be exactly the kind of work that his group considered immoral theft. The thought amused me at one level, but I was also concerned about Mom’s heartache over Harry’s distancing of himself from us and about what might become of him with the arrests and protests.