No. 018

The same month that Science's Editor-in-Chief called AI in publishing "slower, worse, and more expensive", the White House committed $5B+ to a national AI-for-science mission and ARIA backed a directory of 123 problems for AI researchers to solve. Meanwhile, a PNNL team argues the entire "AI scientist" wave has been scaling individual reasoners when the actual bottleneck is team coordination. Institutions are not arguing about whether AI belongs in science, hey are arguing about which layer it should occupy.

Institutional Coordination

• Trump Administration Announces More Than $5 Billion for the Genesis Mission, a National Mission on AI for Science

  White House, July 22 2026

  The largest federal AI-for-science commitment to date: $5B+ across 15 agencies, 278 DOE-selected projects from 5,000+ applications, spanning health, energy, infrastructure, and manufacturing, the biggest response to any DOE funding call in the department's history.

• Challengescape: a living landscape of AI-for-science challenges

  Encode (Pillar VC / ARIA / UK Government), July 20 2026

  A directory of 123 real AI-for-science problems across 10 domains, submitted by domain experts with mentor support, so researchers without years of domain-specific taste can pick up worthwhile work; state-backed problem coordination rather than another agent or tool.

Peer Review and Publishing

• Peer Reviewers Should be Allowed to Upload Manuscripts to AI

  ResearchGate, July 2026

  Argues for allowing peer reviewers to upload manuscripts to AI, against the current publisher consensus (93% of medical journals prohibit it on confidentiality grounds), on the reasoning that restricting reviewer AI use is a bigger cost than the confidentiality risk. Direct counterpoint to the Thorp editorial in edition 017.

• The Future of the Scientific Article in the Age of AI

  SciELO in Perspective (Ernesto Spinak), July 24 2026

  Argues the static PDF article is obsolete and the unit of publication should shift to open research notebooks with auditable code and data, positioning reproducibility infrastructure as the structural defense against AI-generated fraud.

Research Practice

• Networked Intelligence: Active Shared Context Graphs for Human-AI Team Science

  arXiv (Choudhury, Czajka, Monteiro et al., PNNL), July 14 2026

  This paper argues the bottleneck is team coordination, not solo reasoning, and introduces Mycelium, a shared workspace that routes insights across researchers and AI agents, tested on a multi-omics campaign where shared context changed experimental design.

• Don't let AI steal all the joy: what scientists won't give up to chatbots

  Nature Career Feature, July 21 2026

  Nature surveyed researchers on which tasks they refuse to hand to AI: writing (because it reveals gaps in understanding), primary data collection, and in-person observation ranked highest, mapping the boundary where automation stops being help and starts being loss.