New Cognitive Framework for Measuring AGI Progress
Researchers have proposed a framework to measure progress toward artificial general intelligence (AGI), addressing the lack of clear metrics. Drawing from psychology, neuroscience, and cognitive science, they introduce a Cognitive Taxonomy that breaks down general intelligence into 10 key faculties. The framework includes a rigorous evaluation protocol where systems are tested on targeted cognitive tasks to generate a 'cognitive profile' of strengths and weaknesses. The goal is to provide a practical roadmap for empirical AGI evaluation and support responsible governance.
Key facts
- No clear framework existed for measuring AGI progress.
- Framework uses a Cognitive Taxonomy with 10 cognitive faculties.
- Evaluation involves a suite of targeted, held-out cognitive tasks.
- Output is a 'cognitive profile' of system capabilities.
- Aims to reduce subjective claims and improve governance.
- Draws from psychology, neuroscience, and cognitive science.
- Proposed as a starting point for rigorous empirical evaluation.
- Published on arXiv under computer science and artificial intelligence.
Entities
Institutions
- arXiv