Replies (1)

• @guyjantic@infosec.exchange 2026-09-14 14:09

  @selzero@syzito.xyz @micdan5boringtoots@mstdn.social You ask tough questions. This is good. Ignoring my slipshod use of the phrase in my post, I think my understanding of "generating knowledge" requires a knower, but I don't know if I can defend that position very well. I think that, since LLMs aren't intelligent or sentient, they can't know things. Even a much more sophisticated version of AI (the real kind) might not be capable of knowing if it is not sentient or self-aware or something. Humans can know things. Now, for me, it gets philosophically indeterminate: if a machine with no sentience and no ability to know generates information that did not previously exist (and humans are generally OK with saying summaries or remixes of other information counts, here), has knowledge been generated? I want to say No, not until someone else receives and understands the information. I'm flying a bit blind here, though. This really is philosophy, I think. A hypothetical to test some limits: what if I want to know about all the empirical research done since 2010 about the relationship between subjective age and depression (I'm currently studying these two constructs so this came to mind easily). Let's say level up 100x in my coding skills and code a fairly fancy web scraper in Python. It finds, in a large list of peer-reviewed journals, PDFs of all empirical research articles with data collection from human participants mentioning variants of "subjective age" and "depression" in the abstract, as well as one of several terms indicating study of a relationship, like "correlation," "mean difference," etc. Imagine it is better than what I just wrote. Now what if I let that scraper run for a few hours. It concatenates, in a CSV file or maybe a MySQL database, information from each article, including things like sample size, sample demographics, measures used, effect sizes, and statistical significance. It uses meta-analysis stats to get an initial estimate of cross-study effect size as well as some indications of the variability in the individual-study effect sizes used for this. Did the scraper generate knowledge? It is a few hundred or a couple thousand lines of code doing extremely mechanistic things, but it is doing something human researchers do. It isn't doing it extremely well--even with the web scraper I imagine in my dreams there are many decisions left to be made about the data it compiles, but it's probably doing as good a job at this task as some human researchers. This scraper is as far from true AI as I am from being an international pop star. Is information only knowledge once someone like me reads the CSV file and meaningfully gets it into my brain? Is it knowledge before then? I don't know how to slice that, and I also don't think it's a trivial question. I think it's important, or at least interesting, epistemology. My view right now is that information by itself is not knowledge, but that same information becomes knowledge when it is consumed and understood by a sentient being. This means, I guess, that I think your question doesn't capture the situation completely. The person in the example above doesn't generate the information; they consume the information, and then it becomes knowledge. The generation of the information doesn't count (in my current view) as knowledge until there is a knower. #philosophy #epistemology

  Open ##5156168