Evolution wrote your loss function and then left. That is precisely why it is yours.
Navigation: ‹ Evolution Is the Outer Loop · Distillation Through a Bottleneck ›
Essay 5 said that the difference between a scored model and a living system is ownership of the loss function.
Then essay 13 seemed to complicate that claim. If evolution wrote the training setup, then perhaps the loss function is not yours after all. Perhaps hunger, pain, fear, attachment, pleasure, curiosity, and homeostasis are just old instructions installed by an outside process. If a lab owns a model's objective, why does evolution not own yours?
The answer is absence.
The lab is still there. Evolution is not.
A machine-learning model's loss function is held by an ongoing training regime. The objective can be changed by the operators. The model can be paused, copied, rolled back, retrained, evaluated, or discarded. Bad loss is bad for the project, the company, the user, or the training loop. It is not bad for the model in the same constitutive sense unless the model's own boundary is at stake.
Evolution is different.
Evolution wrote your loss function and then left.
There is no external optimiser monitoring your compliance. There is no ancestral committee adjusting your hunger from outside. There is no appeal court where inclusive fitness can complain that you used contraception, adopted a child, became celibate, chose art over reproduction, or died for an idea. The outer loop produced a system whose inner loop now has to live with the stakes.
The loss function is old.
The payer is present.
That is why it is yours.
This point is easy to miss because we speak of evolution as though it were an agent.
It is not. Evolution did not want you to reproduce. It did not design you with an intention. It did not sit outside history and choose your reward function. It is a name for a filtering process across populations and generations.
So when we say evolution “wrote” your loss function, the verb is shorthand. The shorthand is useful, but it can mislead. There is no author left behind the work. There is only a living system whose maintenance now depends on old biases shaped by prior survival.
That is the difference between inheritance and control.