By Terry Winograd and Fernando Flores.
First part is intro into non-rationalist approaches to cognition. They mostly talk about the theories of Maturana (who comes from biology) and Heidegger (who comes from philosophy) and try to tie them together. The main idea is that language comes about through structural coupling — we want to talk in such a way that others understand and believe us. There is no ‘objective’ definition of water; it’s contextual; but that’s not to say there is no grounded definition: we use the word contextually in a way such that others continue to believe our commitment to common understanding. e.g. “Is there water in the refrigerator?” is an ambiguous question without the context: are we looking for something to drink or trying to find out what’s wrong/if it’s leaking? If we’re in the former situation and reply “Yes, in the cells of the eggplant” that’s an obnoxious answer and too much of this will result in no longer believing we are committed to answering questions.
Maturana did really interesting work in vision, showing that there’s no ‘absolute’ processing of light but rather only contextual processing, at a biological level.
Heidegger introduces this idea of ‘thrown-ness’ in which when we are using a hammer we are not thinking about using a hammer or the representation of a hammer; we simply are doing. We sometimes think and reflect, use representations or construct them, but not always (and maybe not mostly.) We are more often “in” it.
They also talk about breakdowns, in which our expectations are not met and we must reevaluate.
Part Two gets into AI and the reason rule-based systems (they talk about ELIZA, the block moving program, and the medical expert-systems, as well as Winston’s analogy engine!) are not a good reflection of human cognition, which is situated and highly contextual and flexible.
Part Three gets into designing computer systems. They critique the management theory stuff by Simon that if we could just program in all the options we could make the right decisions. I liked their analogy of your car breaking down: you could fix it, or a buy a new car. But maybe you can’t afford either. Maybe you take the bus and realize you don’t need a car, or convince your company to intro a shuttle service. You could kill yourself — what constitutes all the options is ambiguous at best and most likely not real. Sometimes you can dissolve the problem instead of solving it.
But I wasn’t super into this section. They talk about how calculators and word processors are great; they are good tools, they have thrown-ness. But this didn’t feels super related to part one. It was mostly about design. Maybe I didn’t read it closely enough. It felt a little up in the air in a way that the other parts were very grounded in theory and concrete.
Perhaps I was expecting it to be more about how to get closer to human cognition than how to produce useful computational tools.
I like the idea of thrown-ness, and their definition of language that allows shared meaning without needing absolute definitions of words.
Thrown-ness is super useful for understanding when a tool works. Makes me think of the low floors/high ceilings analogy for good tools.
Context context context.
Poetry in context: what do you want, what have you read, what do you expect?