Can Evolutionary Algorithms Rival Physical Models? A Deep Dive into DNA as Code

Prologue

I recently finished reading the newly purchased "Proving Darwin" but, due to a busy schedule, I could only write a brief review.

About "Proving Darwin"

The book is disappointing; its first half feels like a prolonged dribble without reaching a conclusion, and the author spends three and a half chapters on ideas that could be summarized in a few sentences.

Computation in Configuration Space

We consider a discrete configuration space C of D dimensions, each with N cells, yielding ND possible states. Each state corresponds to a parameter configuration, analogous to a gene genotype in a genetic algorithm.

Advanced algorithms such as genetic algorithms (GA) and simulated annealing (SA) operate by constructing a controllable potential function V(q∈C, t) where q is a point in the space and t is global time. The evolution step S generates the next point based on V.

For SA: random walk within a neighborhood, selecting points with lower potential; for GA: mutation, segment exchange, reversal, and gene crossover.

Both methods aim to avoid local minima in the vast configuration space, where brute‑force search would be O(N^N) while physical models solve the shortest‑path problem in O(N) time.

Algorithm Comparison

Key challenges for these algorithms are the enormous size of the configuration space, the presence of many local optima, and difficulty in precisely locating the global optimum.

GA and SA mitigate these issues by retaining a population of candidate solutions rather than a single best point, allowing exploration beyond local valleys.

Information Space and Evolution

We extend the model to an information space where each point represents a Turing machine. Genes encode machines that can solve tasks; the potential V reflects task performance.

This framework captures both descriptive and creative aspects of information, suggesting that life, evolution, and even artificial intelligence can be modeled as V‑and‑S processes in an information space.

Philosophical Reflections

The discussion touches on the limits of computable creativity, the role of "garbage" genes as future resources, and the analogy between biological evolution and cosmological multiverse selection.

Ultimately, the essay argues that life is the result of a potential function V and an evolution function S, without invoking supernatural explanations.

Conclusion

The author acknowledges that the presented model does not definitively prove Darwinian evolution but offers a compelling theoretical framework for further exploration.

Editor: Gemini Source: JianShu, Algorithms and the Beauty of Mathematics