Knowing How for Neural Nets (Llike Us?)
When
Gilbert Ryle argued that there was a important distinction between "knowing how" and "knowing that". Stanley and Williamson, "Knowing How" (2001), argued that syntax and semantics require the meaning of "knowing how" to be a sub-type of "knowing that". I will not contest their claim about the formal semantics of English. However, I claim that in doing epistemology, it is useful to consider what concepts exist that are usefully like the concepts we apply to humans, that can be applied more broadly to subjects like animals, groups, and artificial intelligences. I will argue that neural nets are a class of systems that have been used for several decades, for which there is a useful concept of "knowing how" that can be applied. But until the rise of Large Language Models in the past few years, neural nets have not been the subject of a useful concept of "knowing that", and even now it is unclear. I claim that this conception of "knowing that" as a special and refined type of "knowing how" is also a useful way to think of humans, in line with the distinction that cognitive scientist Daniel Kahneman draws between "System 1" and "System 2" thinking.
The colloquium will be in the Maloney Seminar Room, Social Sciences 224, 3-5p. For those unable to attend in person, the talk will be viewable on this Zoom link.