Well of course. But I was thinking about how much of your standard comment section practically anywhere would count as knowledge.
Yeah, most of it is d/misinformation or useless drivel, for sure.
I think the quality of the information is beside the point. Even with the lowest possible threshold for what counts as human generated knowledge, the problem with the factoid isn’t that threshold: it’s the power of doubling. Doubling 40 times in 40 years? Sure. Doubling 1000 times, ever? Out of the question.
This is unrealistically low given his peak production. Any one long novel would run to more than 600 typewritten pages, and he routinely did 6-8 nonfiction books a year. Asimov played tricks with his count to run it up but his genuine work was prodigious enough.
Million word per year writers are a dime a dozen. Loads of pulp writers did it before electric typewriters. Robert Silverberg one year wrote three soft core porn novels a month just working mornings, while devoting his afternoons to nonfiction.
I don’t now if Fuller called fiction “knowledge” and his lack of definition makes this argument moot, but there must be some difference between creating a million words that are entirely new and a million words that Asimov pulled out of his eidetic memory. Libraries are full of fiction, nonfiction, and pure data so any estimation based on them are less about “knowledge” than “marks on paper.”
1 to 3 pages per typewriter; an average of 6 to 8 pages a day, and unlike most authors who have a period of writing and then editing/promotion/et cetera, Asimov basically worked that way for a period of five decades.
There is information in the thermodynamic sense of the world (which is any organized data regardless of whether it is original or duplicative), and then there is knowledge in a sense of contributing original works or a useful survey and organization of existing but uncatalogued knowledge. Although Asimov is known today mostly as a fiction author (and largely science fiction, although he was also a noted writer of mystery fiction) his most prodigious work is actually collecting, analyzing, and reviewing existing bodies of knowledge; his Asimov’s Guide to Shakespeare is, if not a definitive study, accessible and widely read, and his Understanding Physics is an excellent primer for the layperson seeking to understand the fundamentals of (mostly classical) physics if not to master its application and sits on my shelf next to the more extensive Feynman’s Lectures on Physics.
Generative ‘AI’ regurgitates knowledge (or all to often contaminates it with nonsense), and at its best application can draw out patterns from data to sparse or to voluminous for a person to process and review. When an AI can actually produce wholly original works and novel concepts that aren’t just manipulations of cromulent-sounding word salad, I’ll be suitably impressed.
As noted above it was actually 1-3 pages per typewriter. Today an author that worked in such a parallel fashion would just keep the documents in separate files on one computer but for his day and writing process that made sense, I guess. Much of Asimov’s fictional novels were actually ‘fix-ups’ of short stories and serialized novellas stitched together into novel length, and it shows when he went to author an actual novel because he has some difficulty with such overarching plots and pacing. That isn’t a slight—being a good short fiction author is arguably a more difficult skill than writing in novel form—but the later Foundation sequels and prequels really struggle to maintain their story, and Nemesis was just aimless with a premise transparently filched from Stanislaw Lem’s Solaris but without the philosophical musings.
Stranger
Much to nitpick here. Fewer than half his early fiction books were “fix-ups”. By the time he got back into novels in 1982, the publishing world had entered the era of bigger is better, and from Foundation’s Edge onward, his books were twice as long as any early work. He also claimed to have written 90,000 letters and postcards, some of which were posthumously published, creating information if not necessarily knowledge.
A New York Times article in 1969 for Opus 100 said:
He types his 90 words a minute until about 5 o’clock [starting about 10am], sometimes having a coffee break, and always trying not to overeat at lunch. Usually he goes back to the shop after dinner and sometimes remains there until 10 o’clock, when he takes outgoing mail to a box in front of nearby Warren Junior High School.
He obviously couldn’t be continuously composing, since that would be 130 pages per day (ppd). But an average of 8 ppd (on four typewriters) is 730,000 words per year. That’s a good clip by ordinary standards, but nothing special even if spread across many subjects and boosted by numerous short children’s books and various junk books. The real number is somewhere in between.
Asimov was a simply terrible writer of fiction and it’s an embarrassment that the f&sf genre continues to overrate him. He wrote hundreds of guess-the-answers mysteries, but he was never a noted writer of mystery fiction either. He was a truly excellent author of popular synopsizing nonfiction - none of his autobiographies mention any research - and especially short essays, for which he should really be lauded, except that so much of his work is dated today and no longer read.
He was helped producing his volume of work by having an eidetic memory. (I heard this directly from him.) He didn’t have to look up facts, he could draw them from memory. Saved a lot of time, no doubt.
I apologize for starting this mess. I was just making a tiny joke about Asimov being more productive than an AI. At least no one picked up on my other example.
Which novels, besides the first three Foundation novels, are stitch ups? Novels serialized in the magazines before publication don’t count. I Robot is not a novel, so I wouldn’t count that.
I’ve seen, as I’ve noted in another thread a while back, some supposedly technical papers clearly generated by AI. I’d be hard pressed to call them knowledge. But I don’t know how far it would have to go to turn into knowledge. Would a paper as good as the very worst paper submitted to a conference count? (Some are pretty bad.)
“Fix-up” is the standard name for novels which are several short stories by an author which the author puts together. They rework the stories so that the novel is more of a continuous story. This became common from the late 1940s on among science fiction and fantasy authors. There are some examples going back to at least 1851 in other genres. Before the late 1940s, most of the works in science fiction and fantasy were published in magazines. Sometimes the authors thought of the short stories as a continuous story before trying to put them together into a fix-up and sometimes they realized later that the short stories could fit into a continuous story:
“Simply terrible” is an exaggeration. He wasn’t that great certainly. By the end of the twentieth century, the science fiction and fantasy genres have enough other good writers that the readers could quit thinking of the earlier works as being that great.
May I politely request that debates over Asimov’s output and quality go to a different thread?
I doubt this is making any meaningful distinction between data, information, and actual knowledge. It’s possible that the amount of data is doubling that quickly, but there’s no reason that any doubling process must continue to double forever.
Moreover, who is doing all of this doubling? Is everyone on average doubling how much they know every 12 hours, or is it a few massive super-brains multiplying their knowledge by 100x every 12 hours? Both of these seem highly unlikely, so these figures clearly are more tilted toward an assumption of what’s knowable rather than what is known.
What I’m curious about is if there’s any set of assumptions about what “knowledge” or “doubling” could mean that would make the “every 12 hours” part tenable over, say, a period of two years.
The area of the Earth is about 5 \times 10^8 km^2 = 5\times 10^{14} m^2
Planck Length = 1.6 \times 10^{-35} m
Planck Area = 2.6×10^{-70} m^2
So,
Area of Earth in Plank Areas = 1.9 \times 10^{84}
If you started with one bit of information, and you doubled every 12 hours, it would take 139 days before the Earth became a black hole.
Now that’s the kind of quality knowledge that could stand to get doubled!
A few questions:
- Why are you looking at the area of the Earth instead of the volume?
- How are you converting between bits and Plank lengths?
Area because of a rather unexpected result. The Beckenstein Bound. (Also cited earlier in this thread.) The maximum information density of a volume is dependant upon its area, not its volume.
Deeper than my pay grade. But quite interesting in the manner in which thermodynamics, information theory, and entropy are melded with arguments about the nature of black holes. The argument essentially goes that you can only hold at most one bit within one Planck lengthed region, thus area.
In the end, it turns out that black holes exactly meet the maximum defined by the Beckenstein Bound. So once you meet the bound, you have a black hole.
Well, given that the purpose of a technical paper is to present the results of original or replicated research, provide new insights, or present a metastudy of the comparison of previous papers, and generative ‘AI’ large language models are essentially just putting together words in a statistically determined manner to respond to a prompt with no ability to perform any kind of research or concoct original ideas, I wouldn’t classify them as producing ‘knowledge’, and indeed they often produce ‘negative knowledge’ in the sense of confidently wrong conclusions. Of course, I’m not sure how much many human-written papers also provide in terms of real knowledge these days; ideas or research that could at most justify two or three pages circa 1980 seem to take an eight to twelve page layout, puffed up with figures that don’t actually demonstrate anything useful.
Stranger
This is a point I dropped in to make. There is a whole field of information theory, which I have not formally studied, but it seems to relate to this. Data is bits and bytes. It may or may not have meaning. Information is data that we have assigned meaning to or derived meaning from. Then there is knowledge, which is probably a philosophical question. For example, the height of a column of mercury is data. It represents a temperature of 80°F; that’s information. It’s warm outside; that’s knowledge.
Then there is wisdom, which is judging what to do with your knowledge. It certainly doesn’t appear that wisdom of humankind as a species is doubling at any measurable rate.