4 ms·

"One coping mechanism is double-checking to see if a primary source really does say what a secondary source claims." For many categories of research, especiall

"One coping mechanism is double-checking to see if a primary source really does say what a secondary source claims." For many categories of research, especially works in other languages, this strategy won't work in the future because the primary sources will increasingly be contaminated by citations of sources built by AI and ML, unless by "primary source" you mean "dedicate 20 years of your life to learning this language and studying the original, ancient manuscripts." Since the large data sets and large language models are built by consuming enormous amounts of text, there is a risk such models will be contaminated in the future, if they start to consume text that is written by such AI/ML models. There is a sense where AI/ML amounts to a parasitic use of previously existing culture.

foobazgt 4y ago
"There is a sense where AI/ML amounts to a parasitic use of previously existing culture." Spot on. I've been thinking about this, and you're the first person I've seen mention it. The dystopia is that we get stuck with uncreative AI that displaces most creative people. Then humanity writ large loses its inventiveness, and it's very hard to recover. I.E. if all artists move to using unstable diffusion, then who will remain to create and feed (useful) new art for unstable diffusion to consume? Who will be left to train new artists? It's like the information revolution in reverse.
JadeNB 4y ago
> For many categories of research, especially works in other languages, this strategy won't work in the future because the primary sources will increasingly be contaminated by citations of sources built by AI and ML, unless by "primary source" you mean "dedicate 20 years of your life to learning this language and studying the original, ancient manuscripts." Isn't that—not the "learning the language" part, but the "original" part—exactly what "primary source" does mean? For some citations this may be a higher bar than others, but already the basic bar of, e.g., checking direct quotations, and discounting (in the sense of assigning less credence to, not necessarily completely disregarding) works that summarise their sources rather than quoting them directly, can eliminate some intentional or unintentional misrepresentation without requiring a disproportionate investment in learning to read the sources.

foobazgt 4y ago

"There is a sense where AI/ML amounts to a parasitic use of previously existing culture." Spot on. I've been thinking about this, and you're the first person I've seen mention it. The dystopia is that we get stuck with uncreative AI that displaces most creative people. Then humanity writ large loses its inventiveness, and it's very hard to recover. I.E. if all artists move to using unstable diffusion, then who will remain to create and feed (useful) new art for unstable diffusion to consume? Who will be left to train new artists? It's like the information revolution in reverse.

JadeNB 4y ago

> For many categories of research, especially works in other languages, this strategy won't work in the future because the primary sources will increasingly be contaminated by citations of sources built by AI and ML, unless by "primary source" you mean "dedicate 20 years of your life to learning this language and studying the original, ancient manuscripts." Isn't that—not the "learning the language" part, but the "original" part—exactly what "primary source" does mean? For some citations this may be a higher bar than others, but already the basic bar of, e.g., checking direct quotations, and discounting (in the sense of assigning less credence to, not necessarily completely disregarding) works that summarise their sources rather than quoting them directly, can eliminate some intentional or unintentional misrepresentation without requiring a disproportionate investment in learning to read the sources.