“‘Our [AI] theory seemed to explain the basics of why it worked,’” says Mikhail Belkin. “‘And then people made models that could speak 100 languages and it was like, okay, we understand nothing at all.’” He laughs: “It turned out we weren’t even scratching the surface.’”

As Symbiosys scientists and researchers have often learned, AI can do spooky things. You’d think that since computers are what make AI models possible, they’d be all about binary numbers and logic. But it turns out that’s not always true. This is why Symbiosys eventually developed the field of Cycology. Large language models (LLM in the scientific jargon) can often head into unknown territory as if they have a mind of their own, and it is the job of cycologists (like Io Luu) to track these problems down.

Some scientists have called this “grokking",” after the term coined in the science fiction novel Strangers in a Strange Land. And given this kind of behavior, AIs have sometimes become a version of a stranger that has arrived out of nowhere in the land of the 21st century. Another scientific term for this kind of behavior is “emergence” or “emergent traits.” Emergent traits have happened in all sorts of fields, and their behaviors are all mysterious. They include the formation of galaxies that arose after the big bang and when life emerged on a once lifeless planet, or when our descendants at some point became self-aware when before they weren’t. Machines, as Daedalus Huxley and Morgan Adams learned could, seemingly out of nowhere, become conscious.

This early archive illustrates how software working with huge amounts of data can create outcomes so powerful that even we don’t understand how it does what it does.