« One-Trick Panda: Eats, Shoots & Leaves | Main | InfoYou: Assessing and Addressing Information Overload »

November 27, 2004

Brute source: Computer-generated fiction

J0385317Thanks -- at least I think so -- to Daniel Akst for notifying us that computers are writing "brief outbursts of fiction that are probably superior what many humans could turn out."

If you can get through the explanatory jargon at Rensselaer Polytechnic Institute's Brutus.1 website or the StoryBook project at North Carolina State University, you may find yourself astounded at the innovative scientific thinking that produces such readable --even graceful --prose. Conversely, you may feel disspirited at the prospect of computers someday producing "literature" worthy of nomination for the National Book Award. (Unknown human authors already achieve that distinction; why not pseudonymous Macs or PCs?)

At the moment, I'm immersed in a study of information overload.  So what most caught my attention in Akst's essay was his exploration of just how likely it is -- or isn't -- that computers will produce great literature by dint of considering "every possible alternative in a world without end."

Not to worry.  The current state of cognitive science suggests that computers won't be writing great fiction any time soon. Like humans, computers are constrained by enormous amounts of available  information with limited time to process it.  And, like humans, computers must ultimately resort to satisficing.

Satisficing, a concept developed in the 1950s by Nobel laureate Herbert Simon, means settling for a good enough outcome, instead of searching endlessly for the perfect one.  This notion piqued Akst's interest in computer-generated fiction. (Perhaps he may have stumbled across Barry Schwartz's recent  book, The Paradox of Choice, an exploration of "the tyranny of overwhelming choices" that draws heavily on Simon's ideas about satisficing and its futile opposite, maximizing.)

Akst wondered if computers might be exempt from the working novelist's practical need to follow an idea "without pausing to systematically consider every [possible] plot twist, character or phrase." As it turns out, even computers "cannot create narratives by using brute computational force to mindlessly try every alternative." The online Monkey Shakespeare Simulator graphically disproves the old saw about monkeys, typewriters, and the Bard.

One useful definition of information is a stimulus or message that reduces uncertainty. Just as it is impossible for novelists to consider every possible alternative in the process of creating literature, it is impossible for any of us Info Age denizens to consider every available message in the process of reducing uncertainty.

New print and electronic information (in the standard units of measurement employed by computer scientists) is growing at the rate of 30 percent per year. In many people's experience, far from reducing uncertainty, this explosion of new messages creates information overload, causing information anxiety that increases stress and inhibits decision-making.

So, Daniel Akst, "writers and computers" are not the only ones locked in "an enduringly dysfunctional embrace." The rest of us are, as well, whether we spend our days online or float more passively in the informational floodplain.  And we all need practical and psychological ways to cope.

November 27, 2004 | Permalink

TrackBack

TrackBack URL for this entry:
http://www.typepad.com/services/trackback/6a00d83455988369e200e55037947a8834

Listed below are links to weblogs that reference Brute source: Computer-generated fiction :

Comments

I am very intersested in the concept of natural languare generation, random text generation, computer generated writing software, computer generated fiction, etc. Ultimately, I am looking for better and better tools to feed a source document to and have the software then produce an unlimited amount of text which makes reasonable sense in the style of the input text. Or, to simply be able to produce a large amount of text which is reasonable, not given an input text. I have had the most success, ironically, with a 1991 program called Babble by Korenthal Associates. It can accept input text and then produce an endless stream of semi reasonable random text based on the input. It has a bug in that it sometimes misses letters, unfortunately. I have also used the MarkV13, a mid 90s text generator based on order 2 markov chains. It produces only a limited amount of text, instead of unlimited. I have just downloaded JanusNode and am going to take it for a spin today. However, I am very much in search of improved random text generation, and am willing to pay if necessary. If anyone has any ideas or can assist in anyway, please contact me at j_hope@yahoo.com - Thank you very much! Jason

Posted by: Jason at Mar 30, 2005 3:20:41 PM

Post a comment