Is Devolution possible?

[alvations]

C.f.

It's actually quite hard to come with a fitness function for de-evolution. If evolution is to reach the optimal fitness with the range of (-∞, x], then de-evolution has a negative infinity end.

Defining what's the end is a little awkward too. If x' devolution optimal is a proportion of x, where the proportion is between [0, 100], then the x' is bounded by [0, x] and technically not a real "devolution".

Dollo's law of irreversibility assumes/proposes that devolution is not possible.

So the simplest hack is to emulate an evolution process and create some sort of inverse/negative fitness function and drive epitome state (final state of evolution) towards some arbitrary fitness space, e.g. [-x, 0].

Questions:

• Even if final devolution fitness is set to 0, is "convergence" towards optimal fitness always possible? 🤔

[alvations]

Maybe setting a max generation and resets would be more sane to keep track of devolution. Arbitrary fitness doesn't confirm convergence.

One other hack is to rewrite a new fitness function and force the fitness to change to MinimizeFitness instead of the standard maximization strategy.

But the question here is, given the same evolution mechanism and the some sort of inverse or manipulation of the fitness function can devolution happen?

The genesis geneset was never kept in the final evolved form so it's unclear what's the fitness at genesis that devolution should drive towards. And the random initializations mean that there are actually many genesis points to look out from the final evolution.