Why the AI–Culture Relationship Matters for AI Research

The relationship matters because AI systems encode the cultural perspectives of the people and institutions that build them, and those encoded assumptions then spread at scale through content creation and recommendation tools. The Cultures in AI/AI in Culture workshop (a NeurIPS 2022 workshop, held virtually on December 9) frames this as a two-way problem: culture shapes AI, and AI reshapes culture. Understanding either direction is a prerequisite for diagnosing the biases that AI development may be globalizing without explicit intent.

Culture shapes AI in concrete, traceable ways

The workshop description identifies specific mechanisms through which human values and biases enter AI systems:

• Annotation practices — how labelers interpret and categorize data
• Training data pipelines — what gets included, excluded, or over-represented
• Model design choices — the assumptions built into architecture and objectives
• Larger organizational incentives — what institutions reward and prioritize

Because AI development and deployment "is shaped by the particular perspectives of the people and institutions it emerges from," a system's behavior is never culturally neutral. The workshop's stated concern is that without understanding the cultures shaping AI-driven technology, researchers "cannot fully respond to or diagnose the biases we are unconsciously globalizing through AI development."

AI shapes culture through creation and curation

The influence runs in the other direction as well. AI-driven technologies actively shape human experience through art, music, and language production tools that curate content and shape consumption patterns. As AI researchers become involved in designing systems with enormous cultural impact — AI-generated art is the example the workshop highlights — the stakes of getting the cultural dimension right increase.

Why foundational AI/ML research is the gap

The workshop's premise is that early conversations in the field have begun to question AI's impact on global cultures and the Western-centric cultural biases embedded in AI development, but that foundational AI/ML research "lacks a deeper understanding of what culture is and the ways in which it relates to AI research and development." The proposed remedy is cross-disciplinary: pairing NeurIPS and ML community experts with experts on human biases and cultural theories from HCI and the social sciences, so that researchers from outside machine learning can contribute perspectives and methods the ML community does not typically draw on.

What this means in practice

For a researcher or practitioner, the workshop's framing suggests a few checkable questions about any AI system:

1. Who built it, and what perspectives did they bring? The description treats this as a design input, not an afterthought.
2. Where do cultural assumptions enter the pipeline? Trace annotation, data collection, and design decisions rather than treating bias as a property of outputs alone.
3. What cultural effect does deployment have? Content generation and recommendation are named as the primary channels through which AI shapes consumption and experience.

The workshop's own answer to "why does this matter" is that producing culturally sensitive, responsive, and competent AI systems requires thinking about both the cultures shaping AI production and the impact those systems have on global cultures — and that this requires a richer definition of culture than ML research currently operates with.