AI for Just Work: Constructing Diverse Imaginations of AI beyond "Replacing Humans"

Weina Jin, Nicholas Vincent, Ghassan Hamarneh

TL;DR

This position paper critiques the dominant AI narrative centered on human replacement and productivity maximization, which the authors characterize as an ideological monoculture. To counter this, they introduce a conceptual framework called "AI for just work" that embeds ethical values directly into foundational technical assumptions, demonstrating its practical utility through a case study in medical image synthesis.

What Problem Does This Paper Address?

The AI research community excels at answering how to build advanced models but rarely pauses to reflect on why they are being developed. This lack of critical reflection has allowed a mainstream imagination of AI (MIA) to dominate. This mainstream view operates on several questionable assumptions: that AI can solve all complex social and scientific problems, that automating human occupations inherently liberates workers, and that productivity gains automatically translate to societal prosperity.

The authors argue that this monoculture causes real-world harm. It devalues human labor, intensifies worker exploitation (often hiding the massive human effort required to clean data and train models), and exacerbates economic inequality through "so-so automation"—automation that displaces workers without actually improving overall productivity. Furthermore, it frames AI harms as mere "malicious use" or accidental side effects rather than structural design flaws.

Figure 1: Roadmap of this paper that highlights the structure and key contents of each section.

Methods

As a position paper, the methodology is conceptual rather than algorithmic. The authors outline a structured process to construct alternative "imaginations" of AI by replacing unspoken, market-driven assumptions with explicit ethical axioms:

  1. Foundational Assumptions (A1–A3): Defining AI models as limited abstractions strictly dependent on human experience and material resources; ensuring workers and work are never devalued; and explicitly prioritizing ethical values over raw growth.
  2. Deducing AI Properties (P1–P4): Translating these assumptions into concrete technical requirements:
  3. P1: Grounding problem formulation and evaluation in real-world tasks.
  4. P2: Making thorough limitation analysis a default component of AI evaluation.
  5. P3: Designing AI and data collection to be explicitly rejectable (informed refusal).
  6. P4: Designing systems that actively enhance worker rights and power.

To demonstrate how this framework alters technical workflows, the authors apply it to the task of Medical Image Synthesis (MISyn)—the generation of artificial biomedical images.

Datasets

Not discussed in the paper. As a position paper, it does not introduce or evaluate empirical datasets.

Key Insights and Ideas

  • The "Blind Spot" Worldview: The paper draws on philosophy of science to argue that the AI community suffers from a "computational blind spot." It mistakenly treats abstract computational models as more real or dominant than the concrete human experiences and labor that anchor them.
Figure 2: Visualization of two contrasting worldviews of science and technology including computational models. (A) The normal worldview: it acknowledges and values the fact that computational models are built upon human experience (embodied as labor, knowledge, dataset, infrastructure, etc.), and rely on it to validate and apply models in the real world; human experience and models are like pillars and floor in the figure: they are mutually dependent on each other to better understand our living world; computational models are useful reductive abstractions of, but cannot fully represent, the complex real-world phenomena they model (the well-known aphorism that “all models are wrong, but some are useful” [25]). (B) The blind spot worldview: it wrongly assumes that computational models are dominant over human experience, “belies AI’s fundamental dependence on physical nature and socially organized, collective human knowledge” (“the amnesia of experience”), “substitut[es] computation for genuine embodied intelligence,” and “treat[s] abstract computational models as if they were concretely real” (“confuse the map with the territory”). Unspecified quotations in the figure and caption are from Frank et al. [47].
  • The Amnesia of Experience: To make AI appear completely "automatic," the mainstream narrative erases the essential human labor behind the scenes—such as data curation, labeling, and the human buffers required to safely deploy models in real-world environments.
  • Refusal as a Core Design Feature: True ethical alignment cannot just be a post-harm patch. The authors argue that AI systems must inherently support the right of stakeholders to refuse data collection or automated processing, shifting the power dynamic back toward users and workers.
  • The Illusion of "AI for Good": Simply targeting an AI project at a benevolent domain like healthcare does not guarantee an ethical outcome. If the technical practice still prioritizes abstract model-space metrics over real-world clinical utility, it risks becoming "ethics-washing."

Key Results

The paper provides a qualitative mapping (detailed in the paper's Table 1) contrasting current MIA-driven practices with proposed ethical practices for Medical Image Synthesis:

  • Risk Prevention: Shifts from a lack of awareness regarding misinformation risks to establishing strict technical standards and usage declarations that explicitly label synthetic images.
  • Problem Formulation: Shifts from developing models driven purely by technical novelty (e.g., text-to-image generation for its own sake) to grounding the technical design in specific, verified clinical needs (e.g., simulating rare wound types where real data is scarce).
  • Evaluation Scope: Replaces purely algorithmic evaluation in abstract model spaces with a five-phased evaluation paradigm designed to test how the AI and clinical workers perform together in real-world clinical contexts.

Limitations and Future Work

The authors acknowledge that their abstracted statements of the mainstream AI narrative (S1–S4) cannot capture every nuance or variation across the entire AI community. Additionally, their arguments intentionally prioritize the perspectives of vulnerable and marginalized groups, meaning the framework is bound to a specific philosophical and ethical standpoint.

For future work, the authors call for empirical studies to track and monitor structural changes in the community's collective mindset, alongside the development of bottom-up, participatory methodologies to co-design alternative AI frameworks with non-technical stakeholders.

Summary updated

Abstract

"why" we develop AI. Lacking critical reflections on the general visions and purposes of AI may make the community vulnerable to manipulation. In this position paper, we explore the "why" question of AI. We denote answers to the "why" question the imaginations of AI, which depict our general visions, frames, and mindsets for the prospects of AI. We identify that the prevailing vision in the AI community is largely a monoculture that emphasizes objectives such as replacing humans and improving productivity. Our critical examination of this mainstream imagination highlights its underpinning and potentially unjust assumptions. We then call to diversify our collective imaginations of AI, embedding ethical assumptions from the outset in the imaginations of AI. To facilitate the community's pursuit of diverse imaginations, we demonstrate one process for constructing a new imagination of "AI for just work," and showcase its application in the medical image synthesis task to make it more ethical. We hope this work will help the AI community to open critical dialogues with civil society on the visions and purposes of AI, and inspire more technical works and advocacy in pursuit of diverse and ethical imaginations to restore the value of AI for the public good.
BibTeX
@misc{jin2025ai,
  author = {Weina Jin and Nicholas Vincent and Ghassan Hamarneh},
  title = {{AI for Just Work: Constructing Diverse Imaginations of AI beyond "Replacing Humans"}},
  year = {2025},
  eprint = {2503.08720},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2503.08720},
}
Figure