Google DeepMind researchers say AI development is on track to surpass human-level AGI and achieve ASI, consistently beating expert groups
deepmind google
| Source: Techmeme | Original article
Google DeepMind researchers claim AI development is advancing toward artificial superintelligence that will consistently outperform groups of human experts.
Google DeepMind researchers have warned that the trajectory of artificial‑intelligence research is now pointing toward artificial general superintelligence (ASI) – systems that would not only match human‑level AGI but consistently outperform groups of human experts across a range of tasks. In a paper presented by Tim Genewein of the DeepMind Institute, the authors outline several emerging technical pathways that could converge on this capability, arguing that the pace of progress makes an ASI breakthrough increasingly plausible in the near term.
The claim matters because an ASI that can outstrip collective human expertise would reshape competitive dynamics in sectors from finance to drug discovery, and raise the stakes for safety, governance and alignment research. It also amplifies concerns voiced by other AI leaders; for example, Anthropic’s CEO recently reiterated a commitment to “pace” AI development at a Nasdaq opening bell, underscoring industry awareness of the risks associated with rapid capability gains.
What to watch next: DeepMind is expected to release more detailed technical roadmaps and proposed mitigation strategies in the coming weeks. Regulators in the EU and the United States are likely to intensify scrutiny of advanced AI systems, while rival labs may accelerate their own superintelligence programmes. The next set of peer‑reviewed results and any policy responses will be key indicators of how the community balances breakthrough potential against existential risk.
Sources
Back to AIPULSEN