Discussion summary by Magdalena:

On the 21st of February we discussed the paper by Richard A. Watson and Eörs Szathmáry “How can evolution learn?”. In this opinion piece the authors pose a strong claim that “learning and evolution share common underlying principles both conceptually and formally”. They provide some examples when evolutionary processes could be explained by tools from learning theory.

The discussion was especially interesting as Alfredo Rago who is currently working in the group of Watson and is interested (among others) in the question whether “using evolutionary models inspired by learning processes can help us explain the incredible diversity of nature”. He could explain better the point of view of the authors and his own ideas.

However, we identified some issues with the paper and the claims of the authors, e.g.

  • The paper uses “learning theory” in the context of machine learning and artificial intelligence, which may lead to misunderstandings,
  • The authors are not very detailed when discussing papers that supposedly support their claim. Some of them study only special cases of evolutionary processes and as such cannot provide a good support for the authors’ claims,
  • The goals and processes of machine learning can be set in advance but that does not happen in evolution (see the section on evo-ego)
  • While proposing new views and opinions is important in science and while studying similarities and differences between learning theory and evolutionary theories can lead to new insights, too strong claims and simplifications may lead to more harm than good

All in all, the authors did not convince us and we think (machine) learning and evolution may be analogous but they are not equivalent.