Define AGI criteria across multiple domains
Test performance on unseen tasks
Measure transfer learning ability
Evaluate general reasoning capability
Assess adaptability to new environments
Check autonomy in goal-directed behavior
Compare performance to human-level benchmarks
Verify robustness under distribution shifts
Test long-horizon planning and execution
Measure sample efficiency
Evaluate multimodal understanding
Assess continual learning without catastrophic forgetting
Test tool use and external resource integration
Verify consistency across diverse task types
Measure reliability under ambiguity and uncertainty
Evaluate self-correction and error recovery
Assess social and cooperative reasoning
Test real-world problem-solving capability
Benchmark against expert human performance
Confirm broad competence without task-specific retraining