NASA’s new dark energy space telescope can also detect killer asteroids
technologyreview.com·2h ago
AI Research
arXiv CS.AI·Jun 12·5 min read
Definitional alignment before capability alignment: a Design-Science framework for adjudicating claims about AGI
TL;DR
Disagreements about the status of artificial general intelligence (AGI) stem from the lack of a clear definition. A new framework called DAF-AGI was developed to assess definitions of AGI using five criteria and a governance audit.
✦ Why It Matters
Engineers and researchers can use the DAF-AGI framework to critically assess AGI claims and definitions.
Key Takeaways
- Utilize DAF-AGI to evaluate AGI definitions critically.
- Understand the importance of definitional sovereignty in AI governance.
- Recognize the limitations of performance-based definitions for AGI.
- Engage in public accountability when contesting technological categories.
Full Summary
The debate over whether artificial general intelligence (AGI) has been achieved is complicated by the absence of a universally accepted definition. To address this, DAF-AGI, a framework based on Design Science Research Methodology, was created to evaluate AGI definitions through five ordinal criteria and a governance audit that examines authorship, interests, and verification processes.
The framework was applied to various measurement families and tested against a claim that current generative systems outperform educated adults in cognitive tasks. Results showed that this claim could only be validated through performance-based definitions, while other approaches, such as capability-ontology and psychometric methods, did not support it.
The study emphasizes the need for definitional sovereignty, allowing institutions to contest and revise technological categories with public accountability. This framework aims to enhance clarity in AGI discussions and improve governance in AI development.