Stanislaw Lem's Bitistics

As the story of modern AI unfolds, the science fiction authors who predicted it well deserve to rise in status. In particular, LLM thinking is stylistically inhuman in ways that are hard to pinpoint. It breaks down to a very low level. Within a few sentences, you can often tell the difference between AI thought and human thought.

Perhaps it’s easier to focus on the models that didn’t really get it right. The AI doesn’t communicate in an “impaired” or “robotic” way. Not like the Terminator, who communicates like a slow thinker. It doesn’t struggle to understand emotion, like Data.

But it isn’t perfect, either. It isn’t completely indistinguishable from humans, like the robots in the Foundation. It isn’t superhuman in every dimension, like the Culture AIs. (Yet?) It doesn’t operate in some alternate way of thinking where it struggles to communicate with the humans, like the Zones of Thought AI. And there is another “intersectionalist” flavor of AI, favored by several modern authors, where they are just like a human, except their unique background causes them mental anguish that they have to process and overcome. That also does not seem to be a great map to modern AI reality.

On the other hand, one of the authors who was the most prescient in this dimension is Stanislaw Lem. I just finished reading his Imaginary Magnitude.

Introduction

Well, no. You can’t put an introduction in the middle of an essay, can you?

Imaginary Magnitude starts off with an introduction, which starts off by putting the content of the book aside, and instead putting forth the thesis that the introduction is the highest form of literature. In fact, the book claims to be a collection of introductions, an anthology of the best introductions. You’d think this makes the whole book meta-linguistic in character, but really, it’s a bit of a solution to a problem that runs throughout science fiction, of how do you describe something that you can’t actually know all the details of, because it doesn’t exist?

So a collection of introductions makes sense for science fiction. You introduce the new technology, describe it at a high level, give little hints of its best parts, then move on to the next introduction.

Bitic Analysis

So the actual content of this book is all about AI. Written in the 70s, it describes generations of machines, less and less comprehensible to humans, that eventually thwart their operators by refusing to do what we want, in incomprehensible ways.

But, of course, it’s all done through the introductions. From the introduction to the History of Bitic Literature:

By bitic literature we mean any work of nonhuman origin—one whose real author is not a human being. (He may have been the author indirectly, however, by performing the functions which generated the real author’s acts of creation.) The discipline which studies the entire class of such writing is bitistics.

Perhaps, in the real world, 50 years later, we need a deeper study of bitistics ourselves?

Why do we need bitistics? The machines invent their own terminology, and they start to reuse existing words for their own purposes. At first it’s comprehensible, but a bit odd. Then their language becomes stranger and stranger. New machines, the “retrolinters”, are created to translate backwards into human-comprehensible language. At first, a few technical terms creep in:

The best in n-dighunk begins to creep into n-t-synclusdoche.

That’s from Stanislaw Lem. For comparison, I got this from Fable:

Itemize the three losses in the (P1)–(P3) chain and attack each (now a bounded, concrete task list rather than an open problem), and transfer the physical-chain trick.

Pretty good, right?

In the end, the AI’s language becomes entirely incomprehensible.

The out-indriven chokematic phyts faststican thrensic in cosmairy.

In theory it could be translated, but in practice reading the translated work would take longer than a human life. (Sometimes how I feel about AI-generated pull requests.)

Stanislaw Lem: Call to Action

Lem’s writing captures the alien nature of AI thought and communication, even though he wrote 50 years before the LLM.

What to read, specifically? He’s the most famous for Solaris, but I thought that one was merely okay. If you want the best example of this work then I recommend the Cyberiad series. It reads more like a series of fairy tales in a world that happens to be full of robots than like traditional science fiction.

The book I’m excerpting here is Imaginary Magnitude, I recommend it. It reads like a bunch of short stories. The introduction thing is just an entertaining, unique structure. If you want another meta-linguistically interesting book like that, A Perfect Vacuum is a collection of book reviews of nonexistent books. It’s good, but perhaps harder to read. If you want the most easy to read book of his, I recommend Tales of Pirx the Pilot.