Ted Chiang writes only a few stories per year at best, but many of them win awards. I mention him to most anyone and the response ranges from admiration to out and out idolization. I’ve never heard anyone speak of disliking his stories. Often people openly idolize him. I will read any story of his. He’s really quite good.
The Lifecycle of Software Objects is his longest work yet, published as a standalone hardcover novella by Subterranean Press. It’s a gorgeous edition. Simple but evocative cover. The occasional interior illustration similar to the cover. The maps
that precede each chapter are gorgeous. I should also mention that the type design and layout are among the cleanest I’ve ever seen. Very easy on the eyes for reading. They’ll likely be sold out of the book by the end of the month, and I have no idea if there are plans to print another run. If you are at all interested, go buy a copy now.
Chiang’s stories tend to be idea stories infused with realistic personalities, though not driven by them. The Lifecycle of Software Objects promotes the idea that artificial intelligence can only be created through the use of genetic programming algorithms with lots and lots of training. For those not familiar with the term, genetic programming is a kind of computer design that incorporates feedback loops into how it works. Standard programming figures out how something should be done and then programs the computer to do it. Genetic algorithms have the software start off at a reasonable point. After that the software will repeat what it does making small changes each time, getting feedback whether it’s successful or not. For example, the following video shows a robot learning how to flip pancakes with this kind of software.
In the novella, two employees of a software company that makes A.I.s for virtual worlds (e.g., Second Life) struggle with raising
the programs they’ve adopted
, called digients. At the beginning, they aren’t very bright. They get smarter through interaction with people in the virtual worlds, but at a pace that isn’t much faster than actual people. They aren’t exactly people though. They react differently. They have different motivations and constraints. And digients can’t really be let to run in their own virtual world by themselves. That’s essentially a closed system where they can’t really learn from each other. They need new experiences that can only come from interacting with real people.
So what does that life look like? From creation to adolescence (of a sort), the two employees raise the digients, dealing both with some expected pitfalls as well as interesting philosophical questions. For instance, when is an artificial intelligence ready to move out on its own?
One thing that’s really clear to me is that this story really benefits from Chiang’s involvement working in the software industry. The narrative is infused with little details (and key plot points) that would be extremely hard to write realistically without having 20 years experience in the field. Much like like A.I., stories written about A.I. work much much better when they incorporate actual software development experience. I don’t think I’ve ever seen fiction about the software development business be so spot on.
I’m still not quite sure where I’d rank this compared to other Ted Chiang stories. That’s partially because I have enough of a background in programming that the A.I. ideas as well as the philosophy isn’t particularly new to me. Chiang still makes me think, though. Irrespective of it’s place in the Chiang oeuvre, The Lifecycle of Software Objects deserves year’s best consideration.
Other blogged reviews:
Title: The Lifecycle of Software Objects
Author: Ted Chiang
Cover creator: Christian Pierce (artist) / Jacob McMurray (designer)
Imprint / publisher: Subterranean Press
Format: Hardcover
Length: 150 p.
Publication date: August 2010
ISBN-13: 978-1-59606-317-4
I’m intrigued, I have to admit!