Stephen M. Downes

LLMs are statistical models of language learning through stochastic gradient descent with a next token prediction objective. Prompting a popular view among AI modelers: LLMs are just next token predictors. While LLMs are engineered using next token prediction, and trained based on their success at this task, our view is that a reduction to just next token predictor sells LLMs short. Moreover, there are important explanations of LLM behavior and capabilities that are lost when we engage in this kind of reduction. In order to draw this out, we will make an analogy with a once prominent research program in biology explaining evolution and development from the genes eye view.

Debates about human nature inform every philosophical tradition from their inception (see Stevenson 2000 for many examples). Evolutionarily based criticisms of human nature are of much more recent origin. Ironically, most evolutionarily based criticisms of human nature are directed at work whose avowed goal is to biologicize human nature and even to place human nature within an evolutionary frame. Here I will focus on accounts of human nature that begin with and come after E.O. Wilson’s sociobiology. I will also focus on criticisms of human nature that arose first as responses to sociobiology. There are some more recent approaches to human nature that have much in common with the sociobiological approach and I will show that critical arguments developed to target sociobiology have purchase on related recent approaches to human nature. In what follows I will briefly outline some well-known accounts of human nature. Next I will briefly outline some key evolutionarily based arguments against such accounts of human nature. I conclude by summarizing the evolutionary case against biological accounts of human nature and endorsing it.