If “intelligence” is the thing that makes us most human, then the sociologist Harold Garfinkel can tell us a lot more about it than computer scientists like Geoff Hinton or Joseph Weizenbaum.
I’m still having a bit of trouble in understanding what the “documentary method” means. It sounds a lot like what the philosophy H Paul Grice calls the “cooperative principle”, which is the idea that we pick up on what he calls “conversational implicatures” that go beyond what is literally said, by trying to interpret why the person thought the words they said were a helpful contribution to the purpose of our conversation, but it goes on both on the speaker and hearer side, as speakers are picking out what to say on the basis of how they think the hearer will be able to interpret it. Although Grice himself talks about this only in the context of language, it does seem to me to be an important observation about all interactive behavior.
Clearly ELIZA doesn’t actually do anything like this, but I don’t see why a modern LLM couldn’t.
I haven't read Grice but yes, I think they are talking about roughly similar things. I know, for instance, that John Searle talks about the "Background" that is the basis for all interactions in The Construction of Social Reality and I think he got that from his colleague Dreyfus who was definitely reading the ethnomethodolgists. I suspect the differences between ethnomethodologists and the analytic philosophers of intentionality comes down to how much emphasis they put on this Background in terms of how it structures social action. Garfinkel argued that if all interaction is grounded on the documentary method (which is itself a product of embodied socialization), then this means there can be no "objective" social science since social science sits on top of commonly-held, largely invisible, social intuitions that structure both the problems and methods of social science. It was a controversial view that was not popular even among sociologists.
Can an LLM do it? I mean, the more LLMs can remember the interaction, the more they will look as if they are engaging in the documentary method, that's for sure. I think ethnomethodologists would say that there is something ineffable about the documentary method because you can only learn it through embodied socialization but again, I would never say "never."
I’m still having a bit of trouble in understanding what the “documentary method” means. It sounds a lot like what the philosophy H Paul Grice calls the “cooperative principle”, which is the idea that we pick up on what he calls “conversational implicatures” that go beyond what is literally said, by trying to interpret why the person thought the words they said were a helpful contribution to the purpose of our conversation, but it goes on both on the speaker and hearer side, as speakers are picking out what to say on the basis of how they think the hearer will be able to interpret it. Although Grice himself talks about this only in the context of language, it does seem to me to be an important observation about all interactive behavior.
Clearly ELIZA doesn’t actually do anything like this, but I don’t see why a modern LLM couldn’t.
I haven't read Grice but yes, I think they are talking about roughly similar things. I know, for instance, that John Searle talks about the "Background" that is the basis for all interactions in The Construction of Social Reality and I think he got that from his colleague Dreyfus who was definitely reading the ethnomethodolgists. I suspect the differences between ethnomethodologists and the analytic philosophers of intentionality comes down to how much emphasis they put on this Background in terms of how it structures social action. Garfinkel argued that if all interaction is grounded on the documentary method (which is itself a product of embodied socialization), then this means there can be no "objective" social science since social science sits on top of commonly-held, largely invisible, social intuitions that structure both the problems and methods of social science. It was a controversial view that was not popular even among sociologists.
If you wanted to get down into the weeds, I would read this response to Searle from Emmanuel Schlegloff defending the scholarly enterprise of "conversation analysis" and explaining how it's different from Searle's framework of speech acts: https://www.conversationanalysis.org/wp-content/uploads/2019/06/02_Schegloff_To_Searle_on_conversation_-_a_note_in_return.pdf. It's very dense.
Can an LLM do it? I mean, the more LLMs can remember the interaction, the more they will look as if they are engaging in the documentary method, that's for sure. I think ethnomethodologists would say that there is something ineffable about the documentary method because you can only learn it through embodied socialization but again, I would never say "never."