“What distinguishes leaders from laggards and greatness from mediocrity is the ability to uniquely imagine what could be,” American author Robert Fritz once boldly asserted. This quote applies well to the fact that the U.S. edge in artificial intelligence systems is dwindling by the day. The stakes are increasing, however, with artificial general intelligence on the horizon.
Although there are a plethora of definitions for AGI and no real wide-ranging consensus, the industry commonly defines it as an AI system that can perform 100% of human-level cognitive tasks. In this respect, we exclude physical robotics or cutting-edge scientific breakthroughs but only envision a system versatile enough to achieve everything an average human can do in front of his desktop at the office. This AGI race is no longer within the realm of science fiction and has been amply documented or theorized over the last months — by Leopold Aschenbrenner in a document called “Situational Awareness,” Dario Amodei in “Machines of Loving Grace,” Fin Moorhouse and Will MacAskill in “Preparing for the Intelligence Explosion,” or in the more recent, and highly and rightly criticized, “AI 2027” paper by AI Futures project.
Recommended Stories
- International assistance is important, but it must serve American interests
- Revamp US assistance to counter China
- Auto tariffs are steering jobs back to the US
The risks for U.S. leadership are significant. As we have seen with numerous benchmarks — the Turing Test, exhibiting human intelligence equivalent behavior, was officially passed by AI once and for all last week — the Trump administration needs to get serious about the approaching AGI horizon and care about its definition. No one wants to see XYZ lab in Silicon Valley coming up with its own definition of AGI and suddenly claiming the prize.
The Chinese are leading contenders too, but I do not believe the Deep Seek/cheap models/open-source triad really matter here for AGI. These will be valuable following AGI’s arrival. And that’s where the United States finds itself in a predicament. It currently holds the most formidable, unassailable position for AI diffusion: it leads in semiconductor design, data centers, chips, commercial models and deployment. Yes, there are some weak spots, despite the smart decisions taken by the new Administration (deregulation, energy, permitting, project Stargate and the massive infrastructure effort).
As recently observed by Navin Girishankar and Matt Pearl at the Center for Strategic and International Studies, we lack a cohesive digital infrastructure when it comes to networks, grids, and communications infrastructures. Our telecommunications equipment makers are global laggards, and if the administration is so proactive for data centers and chip manufacturing, it should also take care of digital infrastructure. Besides, we do not have a clear view as to how this could play out with tariffs, and even more so in a world where most AI moves to the edge in a few years — portable devices, urban environments, smart vehicles, drones, robots. But achieving AGI, then super artificial intelligence, will not follow the same playbook as the technology diffusion game. Technology diffusion matters the most for economic growth, and I am certain we will win here. Nonetheless, reaching AGI will be much more of a geopolitical game.
Despite our legitimate effort to invest in computing power — chips and data centers — the so-called scaling laws can only take us so far when it comes to AGI. Scaling laws refer to the idea that more compute-heavy systems yield larger and better models, presumably up to AGI. This is not what top AI researchers contemplate though. Computing is roughly only 50% of the equation. The U.S., including its commercial labs, neglect two other components.
The first is algorithmic research — not just model efficiency, the point demonstrated by DeepSeek, but radically innovative models, such as neuromorphic AIs or the return of symbolic AIs, such as brain-inspired systems. We are mistaken in believing that the brute force of computing is all that matters in AI — in fact, leadership in the industry can always be threatened by a group of brilliant people in a garage in Silicon Valley or Bengaluru.
Second, AGI will require systems to acquire an accurate representation of the physical world. This is an essential condition for decent robotics but also for the development of a real AI “brain” smarter than the current stupid systems. The self-driving vehicle industry is a good proxy here, but many other industries will need a mapping of the real world, whether acquiring live data or training systems in virtual environments.
One reason the AGI race is imbalanced is that it pits Silicon Valley against China. The Chinese Communist Party can really structure the whole pipeline of efforts, even though it also relies on a bevy of companies. Thus far, the U.S. has let Silicon Valley rule the game. The U.S. free market system believes that entrepreneurs and innovators are the ones who will eventually win the race against the authoritarian states.
WHY TRUMP’S TARIFFS ARE HARMFUL
Nevertheless, it is time for the Trump administration to go all in. Silicon Valley has no strategic experience in counterespionage, security, and military applications, albeit slowly changing with Anduril and Palantir. The federal government needs to boldly step up to structure the nationwide AGI effort. It should take the burden of security and safety from the commercial labs. The Defense Department can share crucial physical information, for example, to accelerate AI research and robotics and to allow companies to manufacture cutting-edge products for national defense.
More importantly, there are aspects of AGI that the commercial laws are not focused on. We should fund public research at the National Science Foundation laboratories, the Defense Advanced Research Projects Agency, and the AI Safety Institute. We should also pursue moonshot projects that will complement commercial models. Only through this public-private, coordinated effort will we triumph when it comes to new powerful AI systems.
Sebastien Laye is an economist and AI entrepreneur.