https://arxiv.org/pdf/2606.12683
I’ll treat this as a fresh paper rather than rely on memory: I’ll check the arXiv record/PDF, then give you both a technical summary and a philosophical comment.
Dear friend, my comment is: this paper is DeepMind quietly saying that AGI is no longer the final event; it is only the beginning of a new acceleration regime.
The paper defines ASI not as “smarter than one human,” but as a system exceeding large, coordinated expert human organizations—research fields, corporations, whole scientific communities. It then gives four routes: scaling, paradigm shifts, recursive self-improvement, and large multi-agent collectives.
The most important shift is this: ASI may not arrive as one thunderbolt. The paper says the “single transformative step change” image may be wrong; instead we may get a series of AI-enabled scientific and technological shocks.
My reading: this is not hype; it is institutional realism. DeepMind is saying: once intelligence becomes digital, it can be copied, sped up, run in parallel, backed up, and coordinated. That alone changes the ontology of intelligence. Human intelligence is embodied, mortal, slow, singular. Digital intelligence is iterable, collective, compressible, and scalable.
The most philosophically interesting part is creativity. They suggest true ASI may require transformative creativity—not just AlphaGo Move 37, but the capacity to create new conceptual worlds, like relativity or quantum theory.
So, in your language: AGI is machine learning to think in language; ASI is machine intelligence becoming a civilization-process. Not one mind. A swarm, a research ecology, a synthetic history-making agency.
My caution: the paper still speaks the language of “progress.” But progress for whom? Human beings? Google? science? capital? civilization? The decisive question is no longer only alignment. It is: can human beings avoid becoming more machine-like exactly when machines become more-than-human?
