Taming Uncertainty

I read "Taming Uncertainty" (R, Hertwig et al., MIT Press. 2019). It is a book about decision-making and proposes a more humanistic, cognitive process-based approach to decision-making, which has often been discussed in terms of probability theory.

Among the 18 papers, I was first attracted by the title, "Computational Evolution and Ecologically Rational Decision Making (Peter D. Kvam, Arend Hintze, Timothy J. Pleskac, and David Pietraszewski).

It is a method to study the evolutionary process through the life, death, and reproduction of virtual artificial life forms, and discusses how the decision-making process has developed and will develop while dealing with uncertainty.

I read the paper while wondering probably how the logic of the algorithm that generates the genetic code is put together, which is the novelty of the research. What is interesting is that the AI simulates the genetic code using a string of genetic symbols, perhaps because it deals with evolutionary studies. It seems to be very common in evolutionary cognitive studies to use genetic codes.

The point of this paper is that it takes a Computational Evolution approach to simulate how decision-making evolves using artificial intelligence. The advantage of using computational methods is that they can obtain the results of many hypothetical cases in a much shorter time than before.

Specifically, the AI simulates using a probabilistic/statistical model called "Markov Brain" and conducts experiments to determine whether a plant is toxic or non-toxic, and under such simple conditions, even if there are more than 100 pieces of information, the best decision can be made using one or two pieces of information, but under more complex conditions, it is difficult to determine how to make the best use of the information. However, under more complex conditions, the key to the evolution of decision-making is how to obtain information from the surroundings and integrate it appropriately.

Here are the results of a trial run of the tools proposed by the authors.
In this tool, the more the gene code in the "Output by the agents" Best column matches the gene code in the "Target string" in the upper left corner, the more appropriate the decision is. The '#' is the agent, i.e., the number of times the decision was made (the number of times the genetic code was generated).
The premise of this paper is that decision-making is changed by "mutations" in the evolutionary triad (genetic mutation, selection, and reproduction). If the mutation results in winning the race for survival (i.e., matching the genetic code of the target string), then the tool works on the logic of "reproduction" and propagation to the next generation.