# Ai2 — The hard parts of AI-assisted science

- Company: Ai2 (allenai.org)
- Announced: 2026-09-01
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://allenai.org/blog/swedish-autodiscovery-recap
- Record: https://forck.live/items/18535-the-hard-parts-of-ai-assisted-science
- Subject: Olmo / Molmo / Asta

Ai2 hosted an event on August 27, 2026, discussing challenges in building AI systems that scientists can steer as research evolves. Five recurring challenges were identified: leaving room for human judgment, keeping AI steerable, deciding what to delegate, avoiding amplification of bad science, and tightening the loop between AI and experiments.

## Evidence

Verbatim from https://allenai.org/blog/swedish-autodiscovery-recap:

> Majumder described scientific taste as the judgment that helps a researcher distinguish an interesting result from one that's trivial, implausible, already understood, or unlikely to lead anywhere. Today's AI-for-science systems don't reliably make those distinctions on their own.

---

Record: https://forck.live/items/18535-the-hard-parts-of-ai-assisted-science
Catalogue: https://forck.live/llms.txt
Current issue: https://forck.live/feed.md