The term “AGI” is almost useless at this point

Open original

The article argues that “AGI” has become a fuzzy umbrella term with conflicting definitions, making it useless for tracking AI progress or predicting major shifts. It urges us to drop AGI in favor of clear, concrete milestones—like fully automated AI R&D, human-level adaptability, or self-sufficient AI—to ground discussions and forecasts.

  • "AGI" now spans wildly different thresholds, from matching human test performance to running civilization unaided, making claims like "AGI arrived in early 2025" vs "it's a decade off" talk past each other
  • Uneven, jumpy progress across tasks means definitional gaps will keep widening rather than converging
  • Concrete milestones (fully automated AI R&D, human-level adaptability, self-sufficient AI, machine consciousness) or vivid benchmarks (10x annual energy growth, interstellar probes) are proposed as replacements for fuzzy labels like AGI or superintelligence
  • Swapping vague umbrella terms for specific milestones would make safety, regulation, and timeline discussions more reliable

People once used “AGI” to contrast with narrow systems that handle only one task—spam filters, fraud detectors, zip-code readers. Back then, AGI pointed toward the dream of an AI as flexible as a human mind. Now, leading models already cross some people’s AGI thresholds while missing others’ by a mile. That mismatch matters because when researchers or policymakers say “AGI,” they might mean anything from a system that matches human performance on cognitive tests to one that can run civilization without human help. We’ve moved inside a fuzzy cloud of overlapping but divergent definitions. Because AI progress isn’t a smooth curve—performance on different tasks jumps around—those gaps will only widen. Some experts claim we hit AGI in early 2025; others insist it’s still a decade off. Calling for a clear moment when “we know AGI when we see it” has failed. Rather than cling to a single catch-all term, the author argues, we should pick specific milestones: fully automated AI research pipelines, AI as adaptable as humans, self-sufficient machines, or AI with consciousness or moral standing. For those who want to predict that AI still has miles to go, “superintelligence” works as a rough stand-in for capabilities far beyond today’s models. But even that label is a fuzzy cloud until you pin down what “far beyond” entails. Instead of labels, the post recommends vivid, concrete benchmarks—say, civilization’s energy output multiplying tenfold each year or autonomous probes heading to other star systems. In practice, the lack of a shared, detailed vocabulary hampers collaboration on safety, regulation, and predictions. Three people using “AGI” might envision completely different endpoints: tool-using GPT-4, a machine running global infrastructure solo, or an artificial brain with human-style consciousness. If we swap broad terms for sharply defined milestones, discussions about timelines and risks become more reliable. Ten years from now, calling something AGI will feel as outdated as talking about “laptops” when people already carry smartphones.

Ask about this article