Could progress in one research field speed up other fields enough to make growth snowball instead of leveling off?

This explores how a breakthrough in one research field can make other fields more productive, and whether those knock-on effects can stack up into accelerating growth. The corpus covers this question mainly through one model, so treat what follows as a single strong argument plus several reasons for caution, not a settled consensus. The central piece is a network growth model When do AI feedback loops trigger explosive growth?. It treats research sectors as connected nodes, so progress in one lowers the cost of progress in its neighbors. Normally each field hits diminishing returns, because the easy ideas get used up. The model's key claim is that spillovers alone aren't enough. They compound when paired with a second loop, in which higher economic output pays for more research. When the two run together they can outweigh diminishing returns, and the calibrated simulations reach a singularity within about six years under fairly modest automation assumptions.

The model depends on a few large assumptions, and other notes in the collection push hard on them. A critique of the claim that automated AI research could compress four or five years of progress into one Could automated AI research compress years of progress into months? finds that the speedup rests on unproven premises. One is that skill on small, checkable tasks carries over to the research that actually matters. That carry-over between kinds of work is itself a form of spillover, and nobody has shown it yet. A related note on measuring research efficiency Do fixed-budget efficiency gains translate to real research progress? makes a similar point from the measurement side: higher benchmark scores on a fixed budget don't show that the cost of each real discovery is falling.

The most surprising lateral evidence suggests AI may shrink the very thing spillovers need. Spillovers depend on breadth: many fields connected to each other, so that progress can travel between them. But a large study of AI-augmented science Does AI help individual scientists while narrowing scientific focus? finds that individual researchers publish three times as many papers, while science as a whole covers 4.63% fewer topics and has 22% less collaboration, as work clusters on data-rich problems. If the network gets narrower and less connected, there are fewer paths for spillovers to travel. That pressure runs directly against the explosive-growth model.

There are also small-scale glimpses of how spillover-like building could work in practice. In one experiment, thirteen AI research agents with no central coordinator shared a versioned Git record Can decentralized agents coordinate research without a central planner?, so later sessions could build on earlier results without redoing them. That is a mechanism for making knowledge accumulate across workers. On the economic loop, firm-level evidence Do firms substitute labor for AI at different rates? shows that AI gains build on themselves inside firms that already have AI capability, rather than spreading evenly. Anthropic's scenario modeling Does AI growth inevitably shift wealth away from workers? suggests that even when growth speeds up, the gains go mostly to capital owners. The takeaway: compounding spillovers are a real theoretical route to explosive growth, but what the corpus observes looks more like concentration (fewer topics, leading firms, capital owners) than wide cross-sector spread.