Vintage Artificial Intelligence: Before It Got Awkward

Leave it to the Internet Archive to remind us that AI had a whole childhood before it became the most overhyped thing on the planet. We're so deep in the current gold rush — the benchmarks, the existential doom posts, the $200-a-month subscriptions — that it's easy to forget there was a time when "artificial intelligence" meant a LISP script that could beat a human at tic-tac-toe and a conference paper that actually explained the damn approach. This collection is a middle finger to the era of "we can't share the weights, but trust us, it's AGI." Back then, you could kick the tires. Now you're lucky to get an API key and a vague system card. The deeper problem this exposes is AI's institutional amnesia. The field is uniquely bad at remembering its own history — every few years, some lab "discovers" something that was published in 1987 and the whole industry gasps like it's brand new. Attention is all you need, but apparently archiving is not. The Archive stepping in to preserve these artifacts matters because Big AI certainly won't do it. OpenAI isn't going to preserve the lineage of the earlier work their models stand on; they're too busy filing trademark takedowns and polishing their safety theater. The messy, awkward, glorious early days are exactly what the corporate narrative wants erased, because they complicate the fairy tale of lone geniuses and sacred scaling laws. And let's sit with that word "awkward" for a second. The title is doing a lot of heavy lifting. Vintage AI was awkward in the way all earnest young technologies are — clunky interfaces, naive assumptions, demos that crashed at the worst possible moment. But it was also *honest* awkward. The current stuff is awkward in a different register: awkward when it makes up a case law citation, awkward when it tells you it has feelings, awkward when it's used to replace workers at scale and everyone pretends