From AGI to ASI: What Might Follow Human-Level AI?

Looking beyond human-level AI

Artificial general intelligence (AGI) is the starting point for this report’s question: how might machine intelligence continue to develop after reaching a human-level capability? The authors focus on a possible transition toward artificial superintelligence (ASI), which they describe intuitively as a system more intelligent and cognitively capable than large organizations of humans.

The report places machine intelligence on a continuum and notes that Universal AI, at its endpoint, is theoretically well understood. Its main focus is the less certain transition from human-level AGI to ASI. The abstract describes this as an investigation of possible development, not evidence that a particular outcome is assured.

Four possible pathways

The authors discuss four potential routes: scaling AGI, AI paradigm shifts, ASI emerging from large-scale multi-agent collectives, and recursive improvement. These are named as possible pathways in the abstract. Because the available source is the abstract rather than the full report, it does not provide enough detail to explain precisely how each route might work or what evidence supports it.

The pathways are possibilities, not findings that any one route will succeed. The abstract does not state that a route is already underway or specify which is most likely. It gives readers a set of directions the report considers, while leaving their prospects unresolved in the evidence available here.

Frictions and bottlenecks

The report also discusses possible frictions and bottlenecks along these pathways. The authors say their effects could be negligible or substantial, and identify determining their impact as a source of concrete open research questions. The abstract does not quantify the obstacles or describe specific examples, so it cannot support a more detailed account of what might slow progress.

This uncertainty applies to the broader trajectory too. The authors say it cannot be ruled out that AI progress might continue to accelerate over the next years. That wording leaves acceleration as a possibility, not a confirmed trend or a definite prediction. The abstract does not give a date or probability for ASI; it also does not settle whether the proposed pathways will occur.

A sequence, not one turning point?

The authors question whether societal change would take the form of one transformative step when AI first reaches human-level capability. Another possibility, they suggest, is a series of transformative changes as AI-enabled progress and breakthroughs occur across areas of science and technology.

This is a possibility raised by the report, not a demonstrated account of what will happen. The abstract does not establish that change will follow this pattern, that acceleration will occur, or that any one pathway will dominate. Its central message is that substantial uncertainty remains about how AI progress may unfold.

The authors argue that preparing for this prospect requires an interdisciplinary effort with global scope. The abstract presents this as a broad call for preparation, not a detailed policy plan or a tested intervention.

What the source can—and cannot—show

This article is based on the report’s abstract, not its full text. The abstract supports the central question, the four named pathways, the discussion of possible frictions, and the emphasis on uncertainty. It does not provide the report’s full definitions, arguments, evidence, or qualifications. In particular, the pathway labels should not be mistaken here for detailed descriptions of mechanisms.

The report is listed on arXiv. Availability there alone does not establish whether a manuscript has undergone peer review. Based on the abstract, the work is best read as a discussion of possible developments and open questions, rather than as a settled forecast about the timing or likelihood of ASI.