Singularity: AI Architects Claim the New Era Has Begun

The word 'singularity' comes out of the closet

For years, talking about technological singularity in Silicon Valley was to expose oneself to ridicule: it was a science fiction topic, good for transhumanist conferences. That's no longer the case. According to an Axios article published in early August, the very architects of artificial intelligence now claim that humanity has just entered a new historical era, one where machines begin to improve themselves without decisive human intervention.

The idea isn't new: it was formulated as early as the 1960s by mathematician I.J. Good under the name 'intelligence explosion.' The principle is mechanical: if a system becomes capable of designing a system slightly better than itself, the loop sustains itself and the acceleration becomes exponential. What's changed is that this theoretical reasoning is now being adopted as an operational timeline by executives of publicly traded companies.

Google at the forefront of AGI

According to Axios, it's primarily the leaders of Google who are delivering the most aggressive message: general artificial intelligence — a system as versatile as a human across almost all cognitive tasks — is close to realization. The group, which merged DeepMind and Google Brain precisely to accelerate in this area, presents Gemini as a step on this trajectory, not as an end goal.

The narrative isn't isolated. For the past eighteen months, the heads of OpenAI and Anthropic have been making dated predictions: systems surpassing the best human researchers in certain fields by the end of the decade, 'countries of geniuses in a data center,' models capable of writing most of the code for their successors. Technical signals exist: models are already used to generate code, optimize chip architectures, and design scientific experiments. The self-improvement loop is no longer purely hypothetical — it's just still largely driven by humans.

What the narrative also serves to do

These statements should be read for what they are: both technical convictions and acts of communication. Announcing the imminence of AGI produces very concrete effects:

• justifying colossal infrastructure expenditures, driven by demand for Nvidia GPUs and by the data centers of Microsoft, Google, and Meta;
• attracting the most sought-after researchers in the field, in a talent market that has become absurdly inflationary;
• influencing regulation, by presenting the deadline as too close to tolerate national brakes;
• occupying the narrative space against Chinese competition, with DeepSeek at the forefront.

None of these interests invalidate the technical analysis. But it would be naive to forget that those who announce the singularity are exactly those who sell it.

Objections remain strong

Part of the scientific community directly contests the narrative. Current language models still fail on elementary reasoning, hallucinate, and progress mainly through a stacking of calculations whose returns are diminishing. The measured gains in real productivity, in companies, remain modest compared to the hundreds of billions invested. And the history of technology is paved with 'imminent takeoffs' that have turned into lasting plateaus.

Yet the debate has shifted. It no longer concerns the possibility of general AI, but its timeline: three years, ten years, thirty years. This is already a considerable shift — and a sufficient reason for governance, model security, and democratic control issues to stop being treated as secondary topics. If the architects are right, we are behind. If they are wrong, we will still have built the safeguards for a technology already ubiquitous.