Artificial Infinitude #3: Finitude and the Two Categories of AI

If you haven’t read Artificial Infinitude #1 and Artificial Infinitude #2 yet, please do—the following will make more sense.

During the extreme reification of AI that has occurred since November 2022, terminological and conceptual distinctions that used to exist within the field have eroded. One such distinction—the one relevant to this post—is between Narrow AI and General AI. On campuses and in offices, you’ll hear AI referred to as if it’s one thing, almost as if it is everything digital. But the AI that is an approximation of God—that is ultimately most relevant to finitude—is General AI or Artificial General Intelligence (AGI). AGI is categorically different from Narrow AI, which is the AI we’re all living with now.

Narrow AI

In addition to increasing our longevity, we already use artificial intelligence to do things like automatic writing and otherwise improve productivity and potency. Increased productivity and potency is expressed countless ways, from visual analysis tools that can spot tumors on CT scans to Large Language Model-based tools that can write blog posts, Wikipedia articles, and short stories, to massive security systems that simultaneously monitor thousands of camera feeds and alert their masters when there is a potential threat.

AI aids life extension in countless ways too, from accelerating drug discovery to powering the core products of companies such as HereAfter AI, which creates a virtual version of a person using photos, documents, and interviews so that their loved ones can continue to “interact” with “them” after they die. In 2020 the Korea Times hosted a virtual-reality “reunion” between a mother and her daughter, who died of leukemia four years prior and had been reconstructed out of data.

Importantly, however, both productivity enhancement and life extension only cast AI as a tool within an already-existing category of tools. Sure, AI increases our productivity, but so do airplanes, which allow us to attend meetings in Boston in the morning and San Francisco in the afternoon. AI can extend life in new and even eerily human ways, but penicillin extends life too. In the important ways—by which I mean the ways AI lives within and in relation to our consciousness, the ways in which it is or is not like a God—Narrow AI is more similar to the printing press and the internal combustion engine than it is to General AI. It is no replacement for God.

General AI

In contrast—across a categorical divide—AGI is a machine that is not narrowly task-focused but can behave at or above a human level on the same broad spectrum humans can. It seems apparent that to be general, AGI must include the physical realm—if an artificial entity requires a virtual environment to execute its tasks, I don’t think it’s general—though that doesn’t really get talked about among those who claim we’re on the precipice of AGI.1 But hasn’t true omnipotence always meant impacting the physical world at will? If a given AI sounds and acts like a human if speaking through a computer but cannot drive a car and unclog a toilet and rain down sulfur on Gomorrah, is it really General? No matter what the Altmen say, AGI has not been achieved and may never be. There are many reasons to think AGI is not just a matter of scaling existing technology, even if some disagree.

Tad Friend does a pretty good job nutshelling the relationship between immortality and AGI as viewed by the Altmen in a recent New Yorker article on the longevity-obsessed superrich. Here he talks about Ray Kurzweil, who I mentioned in Artificial Infinitude #2:

Kurzweil, now the “principal researcher and A.I. visionary” at Google, is the leading biotech futurist. Decades ago, he predicted that aging would be “dramatically slowed” by 2029, the same year that computers would achieve consciousness. Personalized immune therapies and organ replacements would propel us nearly to longevity escape velocity. The rest would be done by nanobots—cell-size paramedics that zip through our bloodstream repairing aging tissues—and the ability to upload our emulated brains. After our intelligence merges with A.I. in the cloud, we will become mostly digital entities a million times more intelligent than mere humans. This will happen in 2045, a moment called the Singularity.

In the Artificial Infinitude series I am trying to eke out territory in the individual’s consciousness and perception by juxtaposing concepts of AI with life and death in realms both conscious and subconscious. It is in this framing that AGI is crucially distinctive from Narrow AI (and not just an extension or endpoint of it). Not because Narrow AI has nothing do do with our desire for infinitude, our eternal yearning, our ceaseless march to save time and thus distance death, but because Narrow AI still exists relative to humanity whereas AGI signals a break from humanity.

Whether or not AGI will exist as a real thing in the world; what’s important is that it is the acknowledged goal of those who are building artificial intelligence. The idea of it guides real behavior, engenders devotion, dictates the direction of institutions that dictate the direction of societies. Like God, its primary draw is that its achievement represents victory over our own finitude.

It would be an artificial win, though. “For ‘tis the sport to have the engineer / Hoist with his own petard: and ‘t shall go hard,” as Hamlet says. Especially when the petard is smarter than the engineer, as I’ll examine in the final installment of Artificial Infinitude.