Claims that artificial general intelligence (AGI) has already arrived and claims that it remains decades away are often defended using overlapping empirical evidence, yet they reach opposing conclusions because “AGI” lacks a single shared and stable referent.arXiv CS.AI
A new arXiv preprint, “Definitional alignment before capability alignment: a Design-Science framework for adjudicating claims about AGI,” treats this under‑specification as a design and governance problem and offers a structured way to compare and interrogate the definitions that different actors bring to AGI debates.arXiv CS.AI
From Vague Labels to Adjudicable Claims
The authors note that competing AGI claims can be defended “from overlapping evidence” because different operationalizations of AGI return different verdicts on the same system.arXiv CS.AI Rather than attempting to settle AGI as a purely conceptual or empirical matter, they approach it as a problem of adjudication: how to decide, in a transparent and comparable way, which definitions are suitable for answering yes/no questions about particular systems.
Following Design Science Research Methodology, the paper develops DAF‑AGI, a “second‑order conceptual artifact” with two tightly coupled components:arXiv CS.AI
- Five ordinal criteria for assessing the adjudicative fitness of candidate AGI definitions.
- A structured governance audit that examines authorship, interests, certification, external verification, and revision authority for those definitions.
The goal is not to introduce yet another single definition of AGI, but to provide a tool that can be used to compare how existing definitional families perform when they are asked to render verdicts on concrete claims.arXiv CS.AI
How DAF‑AGI Is Demonstrated
The paper demonstrates DAF‑AGI on five prominent measurement families and one deflationary boundary position in a documented corpus:arXiv CS.AI
- Performance‑based approaches.
- Capability‑ontology approaches.
- Psychometric approaches.
- Skill‑acquisition approaches.
- Economic approaches.
- A deflationary position that refuses binary adjudication of AGI.
These are then stress‑tested against a stylized strong arrival claim: that current generative systems constitute AGI because they outperform a well‑educated adult on many cognitive tasks.arXiv CS.AI
On evidence from cited 2024–2025 sources, the paper reports that this strong arrival claim was certifiable only under a performance‑based operationalization. Capability‑ontology, psychometric, and skill‑acquisition approaches did not certify the claim; the economic family remains indeterminate; and the deflationary position refuses binary adjudication.arXiv CS.AI
The authors emphasize that the contribution here is a novel integration and operationalization, not an empirical validation: independent application, inter‑rater testing, and cases developed by authors other than the framework’s creators remain necessary.arXiv CS.AI
Definitional Sovereignty and Algorithmic Sovereignty
Beyond the immediate AGI case study, the paper introduces definitional sovereignty as an enabling component of algorithmic sovereignty.arXiv CS.AI
It characterizes definitional sovereignty as the institutional capacity to contest, certify, and revise imported technological categories under public accountability—that is, the ability of institutions to exercise oversight over the very terms by which technologies are classified and evaluated, rather than treating those terms as fixed or externally imposed.arXiv CS.AI
Within the DAF‑AGI artifact, this idea is operationalized through the governance audit: by systematically examining who authors AGI definitions, whose interests they reflect, how they are certified, how they are externally verified, and who has authority to revise them over time.arXiv CS.AI
A Framework Aimed at Future Application
The authors explicitly position their work as a contribution to how claims about AGI are evaluated, rather than as a final answer to whether AGI has or has not arrived.arXiv CS.AI They highlight that DAF‑AGI is intended to support adjudication of such claims by making explicit the criteria and governance structures embedded in any given definition.
They also note that the framework is designed with potential application to “thresholds in laws and national safety regimes,” even though empirical validation and independent application are still needed.arXiv CS.AI
As with many artifacts in design‑science research, the test of DAF‑AGI will come from how it performs when taken up by other researchers and institutions. For now, it marks an effort to move AGI disputes from loosely structured argument toward explicit, inspectable frameworks for deciding which definitions are fit to bear the weight of high‑stakes claims.arXiv CS.AI