Salvaggio disputes Nature commentary declaring LLMs are AGI

In an essay for Tech Policy Press, researcher Eryk Salvaggio pushes back on a recent commentary in Nature that declares artificial general intelligence has arrived in large language models. The commentary, he writes, applies to machines the same standard used to identify general intelligence in humans, treating sufficient breadth and depth across many domains as enough. Salvaggio calls it "the latest claim that AGI is already here."

His objection is philosophical. Functionalism, the theory that a machine behaving consistently as if it were intelligent is intelligent, has shaped AI thinking since Alan Turing. But the industry, he argues, has flipped philosophical inquiry into a design brief. His blunt line: "An LLM is a system designed by engineers for the explicit task of deceiving functionalists." And grading a system by the same behavioral tests used to build it is, in his phrasing, "circular reasoning to use the same tests to build and to evaluate the same system."

The bar exam is his worked example. "An LLM can pass the bar exam, so some might say it knows the law," he writes. "We should say that it behaves as if it knows the law." The gap between those two sentences carries the argument.

What Salvaggio is worried about is what happens when the as-if collapses in domains that need judgment. "The machine can provide language but cannot provide presence," he writes. Consulting one on such decisions means, in his words, that "we build a world upon simulation, and disconnect our decision-making from the anchor of collective discernment." Two AI researchers in our Who's Who directory shared the piece after it ran.