I was watching a PBS special on how plants, through their roots, compete with other species and sometimes support their own relatives. The narrator and some of the scientists hinted at plant cognition and asked the viewer how plants think. Observing analogous dynamics in software, it’s clear that there is no brain or thinking required.
These types of behaviors can have local (cellular) controls and do not require central planning. Coordination emerges from the cells’ common responses. All plant roots produce chemicals and these chemicals can promote or inhibit growth, trigger other chemicals, etc.
This leads me to point out another widespread misunderstanding of human behavior. A lot of what we do is conditioned or innate, and does not use the brain. Our bodies have nervous tissue throughout, and muscle stimulation originates all over the nervous system. The brain gets too much credit for the complex system of cells that in many cases are doing their own thinking in their own simplistic way.
Ray Kurzweil has been prophetic. I saw him once in a diner but I’ve never met him. I feel like I know him, though, and have benefitted so much from his published work.
However, I’m confused by this update on his new job with Google. Maybe he is simplifying for the interview, but his focus on language seemed off target. I also believe in Marvin Minsky’s “Society of Mind“, and see language as a small piece of the big picture.
I think the information architecture (how concepts are defined and connected) is much more important, and distinct from the language used to represent it. In other words, the reality is distinct from the words used to describe it. If you are a programmer, you might draw an analogy to the MVC framework, where the Model is the information architecture and the View is the language. If you are bilingual, you feel this.
If Kurzweil and his team focus on language, I hope they do so merely as an interface. They can look through the lens of language to build the AI, and users will use language to interact. But language seems like just an interface to the actual interesting work to be done building the singularity.
I’ve been watching Science Channel shows about black holes lately. I know nobody knows the physics beyond the event horizon, including time and space… but doesn’t it seem like an elegant possibility would be that inside a black hole, everything collapses and explodes in a big bang that is entirely contained within the black hole. New time and space, and all the mass and energy in the black hole continue within the sphere of the event horizon. An explosion of the singularity at the center would look from the inside like a big bang, and the curvature of space time might even have edge effects that draw matter back toward the event horizon, causing the accelerating expansion we see in red shift.
It feels like the more we learn about the universe, the more we discover that we have yet to understand.
We marvel at the power and potential of digital computing, mechanical tools, and computer interfaces, but is it any wonder compared to the analogous systems that have evolved naturally with biological rather than electronic mechanisms? Each of us is an independent system–with our own processor, frame, muscle structures for output, and sensory organs for input. Humans thankfully evolve because each of us is different, and different from the bodies that came before us. Further, our likelihood for passing on characteristics to future generations is related to our viability and the functions of our biological systems. We are biological machines.
But biology as we know it is limited by physical constraints on our senses, memory, and life-spans. It may even be the case that biological imagination and creativity are limited by the inherent constraints of neurological chemistry, however, I don’t imagine this is the case 🙂 I cannot see 3000 miles away without a camera and transmission, and I cannot remember the URL of the 473rd web page I ever viewed without electronic logs.
It seems clear that humans are developing electronic and mechanical tools to move beyond the constraints of our biological selves. We use electronics to extend our senses, empower our expressiveness, assist our memory, automate our processing, and improve our life-span. It seems inevitable that the evolutions of biological and electronic systems will begin to merge in order to take advantage of the best characteristics of each. To reach such a state, the interface between these systems needs to be improved. We are working on it already, and it is a ways off, but simply a matter of time. We are truly fortunate that the basic input and output signals of our biological nervous system are electrical.
Cointegration typically uses the price information for two related securities, and provides relative value signals. As with traditional technical trading strategies, changes to the fundamentals create a risk of bad relative value signals. With 2-security cointegration, this risk is doubled because changes to the fundamentals of either company can skew the relative value signal. However, this problem can be cut in half by creating baskets (portfolios) of securities and running the cointegration analysis with each security against the basket. This effectively generates signals which are skewed only by changes to the fundamentals of the individual security. Additionally, the required correlation matrix for the permuted set of security combinations can be replaced by a single vector of correlations – greatly improving calculation efficiency and extending the analysis processing potential.
To improve upon normalizing data to % changes, factors typically associated with beta may also provide better signalling data. For example, as the size of a company grows over the course of a few years, its price volitility may fall. Similarly, as market cap grows, the price change correlations may increase relative to larger cap baskets and decrease relative to smaller cap baskets.
By backtesting, optimal trigger strengths and bet sizes can be measured, however, given the correlation coefficients, volitilities, number of positions, and risk preferences, probabalistically optimal bet sizes may provide better results.