When do AI feedback loops trigger explosive growth?

The paper develops "a general semi-endogenous growth model with an innovation network, where research and automation in one sector increase the productivity of research in other sectors," and from it derives "a clean analytical condition under which growth becomes superexponential ("explosive")." Its central finding is that automating research "can offset diminishing returns to ideas by activating two reinforcing channels: a technological feedback loop across research sectors, and an economic feedback loop in which higher output finances further research." Growth becomes explosive "if the combined strength of technological and economic feedback loops overcomes diminishing returns."

The mechanism rests on separating the two channels rather than treating acceleration as one undifferentiated loop. The technological channel transmits productivity gains across sectors through the innovation network itself — automation in one sector raises research productivity elsewhere. The economic channel is a financing loop: higher output from automated research funds more research. In "a simple simulation calibrated to trends in AI progress," the authors find that "fully automating software research and modest (5%) automation in other sectors generates a singularity within six years," and that "bottlenecks do not overturn the result if task automation advances sufficiently fast" — the two channels compound quickly enough to beat the drag that diminishing returns to ideas ordinarily impose on growth.

This splits and generalizes ground the nearest notes cover differently. Are AI feedback loops strong enough to sustain recursive self-improvement? calibrates a single multiplicative loop against current empirical data (self-assessed productivity uplifts, benchmark win rates) and concludes loops are "not yet self-sustaining" today; this paper instead derives a structural threshold condition from a network model and does not claim to measure where current loop strength stands. Could automated AI research compress years of progress into months? ties the same diminishing-returns obstacle to AI reaching rough parity with "top human experts" in AI R&D; this paper's simulated trigger is narrower — full automation in just the software-research sector plus only 5% automation elsewhere — suggesting the threshold could be approached through partial, uneven automation rather than broad human-level parity. What determines whether AI self-improvement actually compounds? offers a third formalism, a reproduction number comparing recursive gain to research hardening, for the same underlying question of when recursive gains compound into runaway growth.

The excerpt is an abstract-level summary: it gives no values for the sector-productivity spillover parameters, no detail of what "trends in AI progress" the simulation was calibrated to, no specification of which bottlenecks were tested or how "sufficiently fast" automation is defined, and no empirical check on whether the network structure it assumes matches real research sectors. It is a theoretical existence proof plus one calibrated simulation, not a measurement of present-day feedback-loop strength. The fair reading is that explosive growth is a real possibility under a plausible but unverified calibration — not evidence that current automation has already crossed the threshold the model identifies.