Two publications this week, separated by every conceivable demographic, arrived at the exact same anxiety from opposite directions. ARTnews declared that systems art is the 21st century's dominant trend: artists translating abstract infrastructure into sensible, human-scaled experience. On the same day, researchers published BOHM, a zero-cost attribution framework for compound AI systems, which routes tasks through hierarchies of specialized components and then asks: who gets credit for what?
The Attribution Problem Is an Art Problem
Hans Haacke spent decades making visible the invisible systems, corporate funding, real estate networks, institutional power, that structure experience. The BOHM paper is trying to do the same thing for AI pipelines: make legible the hierarchical flow of decisions through components that otherwise operate as black boxes. Both practices share a foundational premise. You cannot critique, govern, or even properly see a system until you can trace what moves through it. Emily Watlington's ARTnews piece argues this impulse is not a style but a methodology, which is exactly what the arXiv researchers claim about attribution: it is not a feature, it is an epistemic requirement.
Infrastructure as Art Medium and Research Object
The parallel runs deeper when you consider a separate paper on the moral geography of agentic AI, which asks whose values get encoded when AI systems are deployed for social good. Haacke asked the same question about museum trustees in 1971. The art world's $2.1B auction week, as Artnet reports, suggests that systems critiquing capital remain comfortably inside capital's embrace. The BOHM framework, notably, is designed to be zero-cost, meaning its insights do not require computational overhead. Systems art, at its best, was also zero-cost: it used existing infrastructure as its medium. The loop closes.