Friday, October 19, 2018

Life by Trial and Error

A new look at assumptions
in the Resource-Patterns Model of Life


Can a living thing survive simply by trial and error? In the 1970s this question started my thinking which has grown into the subject of this blog. If “yes”, if life does succeed by trial and error, I could see back then how to start modeling the process of life. I would start by writing computer simulations in which little agents could roam around in a computer-simulated environment. The little agents would represent Living Things in the real world. The brains of these little agents would be computer programs which I would write, programs which tested strategies for survival using trial and error.

I loved computer programming so I started writing those programs. I also started scanning scientific literature to see if anyone else was working in the same track. This project gripped me and, even though I could spend only part time on it, it shaped my quest for further education up until 2013. Then in retirement it became my full-time project.

Only last month did I realize what I had done 40 years ago. In my eagerness to start modeling I had jumped right over the question in the opening sentence above. I had assumed “yes”:

Life can find ways to succeed simply through trial and error.

So now I had better take stock. In this post I will expand upon the consequences of that assumption.

Reasoning to Further Assumptions

Before I jumped in I believed that I could probably succeed in modeling life by trial and error. There is an obvious strategy: A Living Thing (LT) must try a variety of acts, remember the success or failure of each act, and use this memory in choosing future acts. See the figure below.


Information Processing in a Living Thing

But notice that this strategy for information processing within a LT can work only if the environment surrounding a LT offers a possibility of survival. The rewards offered by the environment to a LT which can learn must outweigh the costs such a LT must incur from errors as it experiments with how to behave in this environment. That is, the environment must contain sufficient resources, and the resources must be distributed in a way which may be learned by at least some of the LTs trying to survive there. The resources must be distributed in patterns which may be exploited by the LTs.

So we must have:
  • Living Things with memory;
  • environmental Resource Patterns (RPs) which may be learned.

But, as we advance toward creating a working model it becomes clear that we can specify more attributes of our Living Things. In addition to memory, our model of LTs must have:
  • senses to pick up clues from the environment;
  • ways of acting to harvest from the environment;
  • an internal store of essential resources sufficient to carry a LT through a time of learning which must include some failures;
  • ongoing consumption of resources which have been imbibed, since a LT needs fuel to continue living;
  • a bias to favor choosing acts which will probably lead to success in discovering and exploiting new RPs;
  • a bias to do some act — even any random act — before too much time has passed, to avoid starvation which must result from prolonged idleness.
It was through reasoning like this that I arrived at the basic assumptions for this model of life. I have listed these assumptions in my presentations of the model (for example see Section 1.4).

One feature of this model stands out when it is compared with other models: In this model Resource Patterns are of paramount importance. This observation helped me decide the name which I have used for this model, being the Resource-Patterns Model of Life (RPM). But I may change the name after some more reflection. Perhaps the name should reflect the prior underlying assumption which I have just noticed, the assumption of life by trial and error.

Comparison with Darwinian Evolution

This Resource-Patterns Model of Life shares some important similarities with Darwinian evolution. In both theories there is trial and error. In each theory there is (1) a mechanism for generating unpredictable variants and (2) an environment which passes judgement on those variants.

But the two theories differ in the range of variants which may be tested within the theory. In Darwinian evolution these variants are limited, as I understand it, to biological traits or species. Whereas in RPM we may also test variants in:
  • single acts of behavior by a LT;
  • adoption of bias by a group of LTs;
  • transfer of life from one celestial body to another.
RPM, we see, is a more general theory than Darwinian evolution. Some of this generality comes from RPM’s application to any size of LT.

RPM also provides a platform to model a set of LTs which cooperate to form one higher-level LT. In many circumstances this higher-level LT will be capable of exploiting a RP which none of the smaller, constituent LTs would have been able to exploit alone. In this case the learning which goes on among the constituents will pertain to how they interact with one another. RPM then becomes a platform for modeling development of language, social instincts, and exploitation of one group by another.

In RPM the failure of a choice does not necessarily lead to death or failure-to-reproduce of the LT making that bad choice. A failure of choice leads, rather, to memory of the error, so such a choice may be avoided in similar future circumstances.

RPM may be seen to encompass Darwinian evolution by saying that the inheritance of attributes in Darwinian evolution is a way that a species (seen as a single LT) remembers what it has learned. Darwinian evolution is one possible mechanism of learning bodily design. Whereas RPM opens study of a broader set of ways to learn.


Basis in Thermodynamics

Living Things must eat if they are to survive. This will seem obvious to most readers without further scientific justification. But, for readers who want deeper science, the necessity to eat can be explained by the second law of thermodynamics. The second law asserts that in every process of energy exchange some useful energy is lost as heat. This means that no machine or LT can carry on forever with its initial store of energy.

A car must occasionally be given gas. A LT must occasionally be given more food energy. But notice the difference between machines and LTs. While cars have us LTs to fill their gas tanks, we LTs have only ourselves to get more energy. How do we LTs manage to get new supplies of food and energy for ourselves? This puzzle, presented to me by one of my mentors during 1973–74, made me think that maybe trial and error could suffice in a system which could remember. As such the second law of thermodynamics underlies my whole RPM project, and the second law is one level deeper than my assumption described above that life can work by trial and error.


Assumptions Might Be Wrong

In quick review, the assumptions in RPM include:
  • Every LT must eat (from the second law of thermodynamics).
  • A LT might succeed if it has capacity to learn by trial and error.
  • The environment must contain appropriate RPs.
  • A LT must have physical abilities to sense and act (in addition to the information-processing capacity to learn).

I admit that one or more of these assumptions might prove wrong one day. Notably, advances in quantum physics might overturn my views of energy and order.

As an engineer I am willing to believe the second law of thermodynamics; it works after all in our human experience to date. But as a philosopher I remain skeptical about the final verdict on the second law. The second law seems vulnerable because it stands upon concepts like energy, matter, time, and information — concepts which may be scrambled by a new and deeper cosmology.

In RPM I build upon the assumptions outlined above to reach a number of socially important conclusions — as you may see elsewhere in this blog (I suggest you start with the Statement of Purpose page). While I allow that the entire structure of RPM is vulnerable, I believe nonetheless that RPM should prove valuable for many of our present purposes.