What Evolution Learns and Other Essays

The near book-length title essay in this collection looks at the deep processes of evolution and shows that it has (in a certain sense) progressively become faster and more efficient.

Evolution is a spontaneous adaptive process that finds solutions to environmental challenges through slow processes of brute-force trial and error. Brains also discover adaptive responses to challenges but do so much more intelligently. Creatures equipped with brains surpass mere trial and error because they learn as they go along and acquire the ability to generalize from past experiences.

But are these two forms of adaptive problem-solving completely different? Newly emerging insights suggest otherwise. Richard A. Watson and other researchers in the field of computational evolutionary biology have found a deep similarity between long-term evolutionary processes and the neural-network learning processes that occur in brains and artificial neural network machine-learning systems. Their work suggests that evolution, too, has moved past mere trial and error and can now make intelligent guesses.

Evolution proceeds not only through mutations in genes but also through shifts in the systems that regulate gene expression. Watson's analysis shows that these "gene regulatory networks" transform over evolutionary time in much the same way that the neural networks in our brains transform as we learn through life experience. In the language of artificial neural network machine-learning systems, evolving gene regulatory networks undergo a form of unsupervised learning. When faced with novel environmental challenges-such as anthropogenic climate change-populations of organisms can effectively generalize from past adaptations to "propose" novel variations that have an elevated likelihood of working out. Evolvability has itself evolved.

Both neural networks and gene regulatory networks operate by constructing models of the world. The second essay, based on the fertile, surprising, and sometimes head-spinning ideas of physicist Yoshitsugo Oono, attempts to answer a deep question: How is it that the world can be modeled at all? The third essay builds on elements of the first two to propose a deflationary account of such metaphysical concepts as consciousness, sentience and self-awareness. The final essay shows how our intuitive models of the world can easily go astray, as demonstrated by the counterintuitive results of double-blind studies.