Bottou and Schölkopf reframe LLMs via Borges, not sci-fi

Léon Bottou and Bernhard Schölkopf have posted a paper arguing that most AI discussion today is shaped by the imagery of science fiction, and that we should ask whether that imagery actually describes what Large Language Models are.

In the short arXiv essay titled 'Borges and AI', the two senior machine-learning researchers write that both AI proponents and opponents 'grasp AI through the imagery popularised by science fiction.' The abstract puts the point as questions: 'Will the machine become sentient and rebel against its creators? Will we experience a paperclip apocalypse? Before answering such questions, we should first ask whether this mental imagery provides a good description of the phenomenon at hand.'

The proposed alternative is literary. The authors want to read LLMs through Jorge Luis Borges, whom they describe as 'a master of 20th century literature, forerunner of magical realism, and precursor to postmodern literature.' This exercise, they claim, 'leads to a new perspective that illuminates the relation between language modelling and artificial intelligence.'

The paper was submitted in late September 2023. The abstract itself does no technical work: it names no Borges stories and reports no model behaviours. It is a framing argument, and the frame is the whole proposal.