Drawing from anthropology and science technology studies, the paper diagnoses the conditions that are hostile toward accountability in ML.

Yeah! Like a “repository” for generic social claims they make and what “common” evidence do they need to gather to back them up;)

Hahaha so right, I’ve corrected it!

thank u!

5/ We call for two collaborative research agendas:
1️⃣ A repository of social claim types + evidence.
2️⃣ Tools to map each claim in an ML paper to its supporting evidence.

4/ These resistances uphold a narrow form of epistemic authority - one that is misaligned with modern-day ML’s diverse goals - and deepen structural inequities between techno-elites & other communities.
💡 Social claims should be treated like knowledge claims - explicit, with evidence for scrutiny.

3/ Applying this lens, we diagnose that the gap is sustained by:
🧠 Cognitive resistance - epistemic assumptions that excuse ML researchers from justifying social claims.
🏗️ Structural resistance - norms & incentives (computational capture) prioritizing benchmarks over social relevance.

2/ We identify the symptom: no effective accountability mechanisms for this gap.
🆕 We coin dead zone of accountability - aspects of the ML ecosystem (methods, norms, metrics, power structures) that resist critical scrutiny & meaningful accountability.

Happy new year Seth!

But it seems like the field is about to get an epistemic epiphany for opportunistic reasons

The blogger is targeting AI PhD graduates that prefers industry research positions. Given that confining industry AI scientists’ research effort within product started happening years ago, this should not come as a surprise.