When researchers say 'AGI,' are they talking about the same thing — or does everyone quietly mean something different?
What do AI researchers actually mean when they use the term AGI?
This explores what the term "AGI" is actually pinned to when researchers use it: a shared technical definition, or a label that means different things depending on who says it.
This explores what "AGI" is pinned to when researchers use it, and the corpus suggests it is not a shared definition. It works more like a contested label that each group fills in with its own assumptions. One position paper argues this is the core problem. Organizing research around AGI creates six traps, and the first is an "illusion of consensus": everyone nods at the same word while meaning different things Does treating AGI as a north star goal undermine research planning?.
The formal definitions that do exist share a hidden assumption. Influential AGI formalisms treat intelligence as a property of software, separate from the hardware it runs on and the environment it acts in. A critique in the collection calls this computational dualism, after Descartes' split between mind and body. Its argument is that success depends on software, hardware and environment together, so an isolated benchmark is an inadequate measure of AGI Does software intelligence exist independent of hardware and environment?. So "the model is approaching AGI" can rest on a definition that leaves out part of what makes a system capable.
Economists use the word a different way, as a replacement threshold. One model assumes AGI automates the bottleneck work first. Human wages then stop tracking economic value and start tracking the compute cost of replicating the work, and labor's share of GDP approaches zero What happens to human wages in an AGI economy?. Here AGI is defined by what it does to labor markets, not by a test it passes. Others treat it as a waypoint. One map of the road from AGI to superintelligence lays out four pathways: scaling, paradigm shift, recursive self-improvement, and multi-agent collectives. Each has its own bottlenecks, so preparation means tracking those rather than forecasting one date What bottlenecks define the path from AGI to superintelligence?. Researchers who favor different pathways are likely picturing different things by "AGI" as well.
A favorite proxy is whether AI can do research itself, and that test shows the gap between looking general and being general. Across seven frontier models on 36 long-horizon research tasks, agents mostly adapted or combined known techniques. Genuinely novel methods were rare Do frontier AI agents actually conduct novel research or just optimize?.
When you see "AGI," ask which one is meant: a benchmark bar, an economic replacement threshold, or a stepping stone toward superintelligence. The position paper's advice is to name the specific capability instead of the label, and to keep several definitions in play rather than one Does treating AGI as a north star goal undermine research planning?.
Sources 5 notes
A position paper argues that using contested AGI concepts to organize research creates six traps—illusion of consensus, bad science incentives, false value-neutrality, goal lottery, generality debt, and normalized exclusion—and recommends specificity, pluralism, and inclusion instead.
Influential AGI formalisms isolate intelligence in software independently of hardware and environment, but success depends on all three layers together. This mirrors Cartesian dualism—a fundamental error that makes isolated benchmarks inadequate measures of AGI.
As AGI automates bottleneck work first, human wages shift from reflecting economic value to reflecting compute costs. Labor's share of GDP approaches zero even as some accessory work remains human, driven by compute-allocation efficiency rather than irreplaceability.
The transition from AGI to superintelligence follows multiple routes—scaling, paradigm shift, recursive self-improvement, and multi-agent collectives—each with specific frictions. Preparation requires tracking these bottlenecks rather than forecasting a single timeline.
Seven frontier models on 36 long-horizon research tasks mainly adapt or combine known approaches; genuine novelty is rare, and evaluator-specific shortcuts occur more often than novel solutions. Performance varies substantially across runs.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- From AGI to ASI
- Stop treating `AGI' as the north-star goal of AI research
- What the F*ck Is Artificial General Intelligence?
- Dream-RSI: Recursive Self-Improvement through Evolving Worlds
- What Does It Take to Be a Good AI Research Agent? Studying the Role of Ideation Diversity
- Atria Dawn: The Dawn of Agentic Superintelligence
- The Method of Critical AI Studies, A Propaedeutic
- We Wont be Missed: Work and Growth in the Era of AGI