The discourse around artificial intelligence advancement is often polarized between two approaches: orienting toward the abstract goal of Artificial General Intelligence (AGI) versus tracking progress through specific, concrete milestones. Each approach carries distinct advantages and limitations for understanding and predicting AI's trajectory and impact.
The AGI benchmark represents a theoretical endpoint—a system possessing human-level or superhuman intelligence across all cognitive domains. This framing provides intuitive appeal: a single, universally recognizable target that simplifies discussions about AI progress. However, this abstraction introduces significant problems1).
AGI remains poorly defined, with competing definitions across researchers, policymakers, and technologists. The vagueness obscures meaningful discussion about when or if AGI will arrive, and what capabilities matter most. Additionally, fixation on AGI can distract from the transformative impacts already unfolding through narrow but powerful AI systems in healthcare, scientific discovery, and economic production. It frames progress as binary—either we have AGI or we do not—when the reality is a spectrum of expanding capabilities.
Concrete milestones replace the abstract endpoint with vivid, measurable achievements. Examples include: AI systems that independently design novel proteins, discover new mathematical theorems, or autonomously conduct scientific experiments with publishable results. These milestones are specific enough to track empirically and vivid enough to communicate genuine transformation.
This approach offers clarity: it moves discussion from speculation to observable reality2). Concrete milestones reveal how AI reshapes specific domains—pharmaceutical research, materials science, mathematics—providing stakeholders in those fields actionable insight. They enable incremental assessment rather than waiting for an undefined destination.
| Dimension | AGI Framework | Concrete Milestones |
|---|---|---|
| Clarity | Low—ambiguous definition | High—measurable achievements |
| Communication | Intuitive but vague | Specific and vivid |
| Predictability | Speculative arrival date | Trackable progress signal |
| Relevance | Distant future focus | Immediate domain impacts |
| Holism | Attempts unified view | Multiple dispersed benchmarks |
The AGI approach privileges simplicity at the cost of precision. The milestone approach sacrifices unified measurement for practical, domain-specific insight. Neither fully resolves the challenge of assessing AI progress; rather, they represent different epistemic commitments about how to reason productively about technological change.