Do Machines Really Understand?
Pietro Perconti, Alessio Plebe · 2025
This chapter analyzes the lively ongoing debate on the actual understanding of language by machines. It may seem ironic that the most heated debate about whether machines could understand in the way we humans understand took place in an era, around the 1980s, when there was no AI realization that could truly aspire to that title. It is not so strange when it comes to philosophers, on the contrary, it is their habit to passionately discuss currently non-existent objects. Let us not forget that the spark to detonate the discussion was lit by Turing, in years when computers had yet to be born. Toward the end of the last century, the debate had lost its enthusiasm, a lot had already been written, and the frustrating lack of progress by AI towards a system that showed some glimmer of understanding of the world contributed to a certain scarcity. Now, suddenly we find ourselves even having crossed the finish line of the Turing Test, so it is more than reasonable to see the debate about whether machines really understand or not reignite. In this section, we try to gather the different positions, and we can immediately notice how the most vehement are of reaction, denying any possibility of genuine understanding for AI. They partly relate to the reactions already seen to the new AI landscape offered by deep learning. But now the stakes are much higher, it is really about understanding natural language, often considered a human prerogative, unreachable by computers.