he Organisation for Economic Co-operation and Development (OECD) has released the first standardized framework to measure AI capabilities against human benchmarks. The AI Capability Indicators represent the most comprehensive attempt yet to cut through the hype surrounding artificial intelligence, offering businesses, policymakers, and educators a clear roadmap of what AI can—and cannot—do today.
Developed over five years by 50+ experts in computer science and psychology, the framework categorizes AI performance across nine key domains: Language, Social Interaction, Problem Solving, Creativity, Metacognition, Knowledge, Vision, Manipulation, and Robotic Intelligence. Each is graded from Level 1 (basic tasks) to Level 5 (human equivalence). The results? A sobering reality check: most AI systems cluster at Levels 2–3, far from the superhuman AGI often portrayed in media.
Key Findings: Where AI Stands Today
- Language Models (e.g., ChatGPT):
- Level 3 in language comprehension but struggle with analytical reasoning, often «confidently stating nonsense».
- Social Interaction:
- Barely Level 2—AI can mimic emotions but lacks genuine understanding of social dynamics.
- Vision & Robotics:
- Level 3 in object recognition but falls short in adaptable, learning-oriented tasks.
Implications for Business and Policy
The framework empowers leaders to:
- Demystify vendor claims: Ask for specific capability levels before investing in AI solutions 1.
- Identify automation vs. augmentation opportunities: Level 3+ tasks (e.g., structured customer service) can be automated, while Level 2 areas require human oversight.
- Prepare for hybrid workforces: Education, for instance, may see AI handling standardized instruction while teachers focus on mentorship and creativity.
The Road Ahead
The OECD’s work highlights critical gaps in social intelligence and creativity, where progress is slowest. Meanwhile, breakthroughs like GPT-4.5 (OpenAI’s latest model) and DeepSeek R1 (a cost-efficient, open-source alternative) continue pushing boundaries in reasoning and multimodality.
