Paul Palmer and Matthew Smith's argument that human adaptation to climate change should be incorporated into climate-projection models is entirely reasonable (Nature 512, 365–366; 2014). However, I suspect that doing so could render such models essentially useless.

Climate models are created with the intention of providing predictions that are more reliable, and as such must always wrestle with the bias–variance dilemma. Introducing human responses to climate change will make this issue much more challenging than it already is — and perhaps hopelessly so.

To put it bluntly, one does not need to be an expert in modelling non-linear systems to recognize that the best answer to the question 'How can we get more precise predictions?' is never 'Add lots more variables'.