Notes on notes

I loved Jasmine Sun’s notes on AI and writing, par­tially because I’m a sucker for this format, but mostly because her notes — musings — provocations — are all just very good.

One of the final items on her list is this question — 

How much of the world does lan­guage contain?

—which is actu­ally a huge one regarding the AI industry’s hopes and promises, at least insofar as they depend on lan­guage models.

For my part, I think the answer is “much less than you’d think”. That’s for sure true if you take “the world” to mean the real world, human and non­human like … but even if “the world” is only the realm of human expe­ri­ence and activity, I think the frac­tion is still fairly small.

You might reply, “Okay, but the important parts of the world are encoded, and encodable, in lan­guage.” That’s only con­ceiv­ably cor­rect under the second, shal­lower def­i­n­i­tion of “the world”, yet even on those terms: nah, I don’t think so.

The proof is that writers of all kinds are, today and every day, finding new ways to say things, and dis­cov­ering new things worth saying that have never been said before. There is no sense of scraping around the edges for like, the final one per­cent of interesting, useful statements. The opposite! Real expe­ri­ence looms vast across every sen­sory channel; all the words spilled through all the cen­turies seem not to have made a dent.