how do you build the everything machine?

Our tech oligarchs want an "AGI." No one really knows what that actually means.

One of the most often repeated jokes about nuclear fusion is that it’s the clean, safe, and almost infinitely abundant energy source that’s always 20 years away. Which is a harsh but fair assessment. The complexities of recreating fusion on Earth have been either estimated or downplayed for decades, and we’ve only recently started making genuine strides towards it, although the shoestring budget isn’t helping. This is why so many skeptics doubt we’ll ever make it happen in the first place.

You can also say the same thing about the most sought after breakthrough in Silicon Valley currently animating the obsession with things like large language models: AGI, or Artificial General Intelligence. Since the early 2000s, it’s been a matter of faith that at some point in the foreseeable future, we will be able to create an artificial mind just as flexible, adaptable, and inquisitive as ours, destined to outpace us within a matter of years after its invention, and possibly even rule the world.

However, there’s a very big and important difference between the two. We know that fusion exists and happens regularly. Not only is it why we exist and are alive, it’s by far one of the most common processes in the known universe. We can induce it as a very significant net loss. We also know why we’re struggling to recreate it in our machines.

Stars use immense gravitational pull to mechanically crush atoms until the Coulomb barrier is broken. We can’t create that in tokamaks, so we have to heat up plasma to between 100 and 150 million degrees for extended periods of time, and while we can pretty trivially do the first part, we don’t have the materials or tools for the second. Hence the constant delays in building a reactor that can produce even the slightest but more energy than it consumes. It’s not that we don’t know if it’s possible, or what the problem is. It’s just a really difficult problem to solve.

AGI is a very different animal because if you ask a dozen computer scientists what is artificial general intelligence, you’ll get two dozen different answers. There isn’t any concrete or accepted definition of AGI in the discipline, and promises that an LLM or some startup will build an AGI by a certain date, usually in the 2030s, are the comp sci equivalent of saying that a supplement “boosts the immune system.” It’s vague, functionally meaningless, but sounds impressive.

Unfortunately, computing is bound by the rules of math and general vibes or vague and obscure ideas won’t cut it. Just like two plus two doesn’t equal “I don’t know, I wanna say maybe like something between negative six and 887,” a function call or a statistical analysis has to do something concrete, or it can’t exist at all.

Without that concrete definition, computer scientists are looking at a scenario out of Slavic folk tales in which a villain tells a protagonist to “go somewhere, I don’t know where, and fetch me something, I don’t know what” so the protagonists get lost and hopefully die on an impossible quest with no defined goal.

Increasingly, even to the most devout adherents of the AGI mythos tasked with trying to make it happen, the whole idea seems doomed and they’re not sure exactly how to break this realization to their bosses who bet a not insignificant part of the economy on this highly anticipated leap. There are very real worries that admitting that AGI is a fool’s errand, at least developing AGI the way they picture it and with timelines already promised to investors, could cause a significant economic correction, if not worse.

Just imagine losing your job because a philosopher in the late 1990s decided that he suddenly figured out how to make computers smarter than humans, a small group of tech billionaires on a cyberpunk sci-fi kick believed him, and plowed a few trillion into trying to make it all happen. Sounds absurd, but sadly, this is very much a reality if the current AI bubble pops, as seems more and more likely with every month.