Research guide
What Is AGI? An Engineering Definition of Artificial General Intelligence
AGI stands for Artificial General Intelligence. COG examines AGI through complex-project delivery: can an AI system sustain work that would otherwise require a human team, under clearly stated conditions? This guide explains that engineering perspective and links to the original research.
This guide introduces COG's research perspective. Consult the original papers for complete arguments, conditions and references.
Goals · Execution · Verification · Delivery
How does COG define engineering AGI?
In COG's working definition, engineering AGI is a system-level capability state. Given objectives, available resources, safety boundaries and initial context, and where acceptance boundaries can be established independently, an AI system can reliably complete most typical complex projects that would otherwise require sustained collaboration by ordinary human teams, without continuous human steering, while producing verifiable deliverables. This is a working definition for engineering research and assessment, not a claim that all AGI research uses one standard.
An Engineering Definition of AGI: Complex-Project Closed-Loop Capability as an Assessment Criterion
How do closable projects differ from open-frontier innovation?
Closable projects can involve unknown solution paths, exploration and design innovation. Their defining feature is that delivery can be judged against independently established goals, constraints and acceptance conditions. Engineering AGI does not require sustained contributions beyond the human frontier. Moving toward ASI additionally raises questions about direction discovery, structure generation and value judgment in open problems.
An Engineering Definition of AGI: Complex-Project Closed-Loop Capability as an Assessment Criterion
How is AGI different from performance on individual tasks?
An answer, a code-generation attempt or a benchmark score primarily describes local performance under particular conditions. Complex projects also involve dependencies, persistent constraints, error recovery, acceptance and reasonable changes. Engineering AGI therefore concerns the ability of a complete system to sustain a project loop, rather than the quality of a single model call.
An Engineering Definition of AGI: Complex-Project Closed-Loop Capability as an Assessment Criterion
Does autonomous delivery exclude all human participation?
No. Humans may still set objectives, allocate resources, define safety boundaries and provide final authorization. The distinction is whether the system continues to depend on humans to decompose routine work, correct its direction, restore context or govern the process. Deliverables still require independent verification.
An Engineering Definition of AGI: Complex-Project Closed-Loop Capability as an Assessment Criterion
How can engineering progress toward AGI be examined?
The three-axis model considers scale, paradigm and engineering as complementary directions of progress. AI engineering studies how finite capabilities can support reliable delivery, while MRAC uses audit-and-repair trajectories as process evidence. The effectiveness of an individual mechanism, or audit convergence alone, does not establish that AGI has been achieved.
Original papers and DOIs
An Engineering Definition of AGI: Complex-Project Closed-Loop Capability as an Assessment Criterion
Artificial general intelligence (AGI) has long lacked a stable and operational definition. Existing definitionsvariously emphasize human-like behavior, subjective mind, broad cognitive ability, human-levelperformance, economic value, or soc…
DOI: 10.5281/zenodo.22803294The Three-Axis Model of AGI Progress
A model of AGI progress using the scale axis, paradigm axis, and engineering axis, with the threshold of stable complex-task delivery as the key milestone.
AI Engineering: A Disciplinary Program for the Reliable Delivery of Intelligent Tasks
Large language models and their Agent systems can already generate code, plan tasks, invoke tools, and modify real engineering states, yet growth in capability has not been matched by corresponding improvements in the reliable delivery of c…
DOI: 10.5281/zenodo.21931156Multi-Round Audit Convergence: Coverage-State Assessment and Recursive Control for Complex AI Engineering Tasks
Large language models and their agents can already handle some real-world software engineering tasks, yet a single successful execution is still insufficient to demonstrate that a specific result has reached a level at which it can be deliv…
DOI: 10.5281/zenodo.21959920