Position paper proposes argumentation as foundation for Evaluative AI
Evaluative AI (EAI) is a recently proposed way to support human decision-making. Rather than having the system output a single recommendation, an EAI tool presents competing hypotheses side by side, each backed by evidence for and against it, leaving the final call to the human. In this position paper, the authors advocate computational argumentation as a particularly suitable paradigm for giving EAI a formal, computable foundation, specifically for building EAI systems that are explainable and contestable: a person can see why a hypothesis is favored or opposed and can challenge the reasoning rather than just accept an output. The paper does not present an implementation, an experiment, or a fully specified argumentation formalism. It is framed as the starting point of a long-term research agenda aimed at distributed, human-centered EAI systems, though no timeline or concrete milestones for that agenda are given.
Key facts
- Evaluative AI supports human decision-making by presenting competing hypotheses with evidence for and against each, rather than a single recommendation.
- The paper advocates computational argumentation as a particularly suitable paradigm for a formal, computable foundation for EAI.
- The stated goal is EAI that is explainable and contestable, not just accurate.
- It is a position paper: no experiment, implementation, or specific argumentation formalism is presented.
- The authors frame the work as setting the ground for a long-term research agenda toward distributed, human-centered EAI systems, without a timeline or milestones.
Why it matters
Most AI decision-support tools hand over a single answer and expect trust in the black box that produced it. Evaluative AI takes a different stance: show the competing hypotheses and the evidence on each side, and let the human weigh them. This paper's contribution is proposing a formal backbone for that idea. Computational argumentation, a field built around representing and reasoning with conflicting claims and evidence, is offered as the mechanism that could make EAI systems explainable (the reasoning behind a hypothesis is visible) and contestable (a person can push back on that reasoning), rather than just another opaque scoring system with a friendlier interface.
Who it affects
Directly, researchers working on explainable AI and human-centered decision support, since the paper is pitched as an agenda-setting call rather than a finished tool. Indirectly, anyone who would eventually use an EAI system to make a decision, since the paper's whole premise is changing what such a system shows them: hypotheses and evidence instead of a verdict.
How to use it
There is nothing to deploy yet. The paper does not describe an implementation, a released system, or a specific argumentation formalism to adopt; it is a position paper laying out why argumentation is the right foundation to build toward, available as an arXiv preprint for researchers who want to engage with or build on the proposal.
How solid is it
This is explicitly a position paper, not an empirical study. No experiments, no implementation, and no fully worked-out argumentation formalism are presented; the case rests on argument, not on demonstrated results. The abstract also does not name the authors or their institutions, so the claim carries no attached track record to weigh it against.
Risks and caveats
The proposal is programmatic rather than demonstrated: it advocates a direction and sets an agenda without showing that computational argumentation actually delivers explainable, contestable EAI in practice. The 'long-term research agenda' toward distributed, human-centered EAI systems is named without a timeline or milestones, so there is no way yet to judge how far off, or how achievable, that agenda is.