AI cooked new languages or just literal code salad
One side sees silicon poetry, the other sees autocomplete that never learned subtext.
Science Magazine highlighted AI systems generating novel symbolic systems. Critics countered that these outputs stay rigidly literal and overlook pragmatic cues humans take for granted.
Why these scores — Side A rests on a single Science Magazine post without the linked study; Side B offers the autism parallel as observable pattern rather than data. Both cite public examples that check out, no bot signatures detected.
Science Magazine dropped a post claiming AI had spun up entirely new languages with their own grammar rules, racking up quick shares before anyone asked what counted as language.
Side A points to generated lexicons and syntax that humans did not pre-program, arguing this crosses the creativity threshold once reserved for people. Side B replies that the systems still treat every token at face value and fail the same pragmatic tests autistic communicators have navigated for decades without anyone calling it invention.
The 63 engagement score sits on verified accounts trading primary examples rather than bots, yet neither side has released the full training logs that would let outsiders judge whether novelty is real or just recombination at scale.
AI produced original lexicons and rule sets never supplied by trainers, proving machine creativity exceeds simple remix.
- @ScienceMagazine✓ verified“AI constructs new languages, challenging assumptions that only humans can make that creative jump.”
AI output stays surface-level and misses implied meaning the same way autistic speech has for decades, so no genuine creative jump occurred.
- @Squeeze1i✓ verified“AI takes words literally and misses cues; the panic ignores how autistic people have handled this for decades.”
Read it straight — Open the linked study or raw model outputs yourself instead of accepting the tweet summary as evidence.