In Neal Stephenson’s novel Seveneves, there is a group of people called Cycs who have undertaken to collectively memorize a carefully preserved five-thousand-year-old Encyclopedia Britannica. One of the characters is named Sonar Taxlaw, after the 17th volume, which she has full mastery of at her 16 years; she also has a loose acquaintance with the rest.
It beggars belief that a novel written so recently would have chosen the Britannica over a paper copy of Wikipedia, but what would the Wikipedia-based equivalent look like?
The English Wikipedia is about 5 million articles at this point and about 50 gigabytes of UTF-8 text, for a mean of some 10 kilobytes per article. The “Vital Articles” lists are carefully chosen subsets of 10, 100, 1000, and 10 000 (actually 9841) articles, which tend to be longer than the mean (which is dragged down substantially by stub articles, which are vanishingly unlikely to be Vital), at around 22 printed pages or 320 kilobytes each.
If we take Stephenson at his word that a 500-or-so-page Britannica volume can be adequately mastered by one bright person, then we have about 23 Wikipedia articles per person; so the Wikipedia Vital 100 could be summarized by four or five people, the Vital 1000 (close to Britannica in size) by about 44 people, and the Vital 10000 by about 440 people.
However, it is often the case that useful knowledge arises only from connections between topics. Suppose that instead we want to ensure that each pair of articles is present in the mind of at least one person, so that at least syntheses that need to draw on any two different articles have a chance to arise.
Here’s one scheme for how to do that: divide the articles, perhaps randomly, into groups of 11, and make a matrix with a row and a column for each group of articles: 91 groups to cover the Vital 1000, say, gives us a 91×91 matrix. Then, assign a person to each above-diagonal cell in this matrix, (* (/ 90 2) 91) = 4095 people in this example. Have that person learn all 22 articles from their row and their column.
This seems somewhat wasteful: each person contains (* (/ 22 2) 21) = 231 pairs of articles, but (* (/ 10 2) 11) = 55 of them are shared with everybody else on their row, and another 55 with everybody else on their column. So only (- 231 55 55) = 121 of their pairs, just over half, are unique. So out of the total (* 4095 231) = 945945 pairings of articles in people’s heads, you only have (* (/ 1000 2) 999) = 499500 unique pairs. Furthermore, a person who can master 23 articles could actually contain (* (/ 22 2) 23) = 253 pairs, which could potentially reduce the necessary number of people further to (/ 499500 253.0) = 1975. But it’s not clear how to reach or approach that bound.
Regardless, the number of people needed to cover all the article pairs in this way is proportional to the square of the number of articles, or rather, the amount of information. So to cover all the pairs in the Vital 100 with the simple assignment scheme above, you would need only ten people.
Printing out all of English Wikipedia on paper at full size would use about 3.4 million pages; linearly reduced 4:1, at which point it’s 3-point text that’s barely still readable without a magnifying glass, would be about 212 thousand pages, or 212 reams of paper.
Wikipedia:Size notes, “In 2015, Michael Mandiberg published the English Wikipedia in 7473 volumes of 700 pages each via Lulu, an online e-books and print self-publishing platform, distributor, and retailer.” Probably nobody has yet printed out all 7473 volumes, a total of 5.2 million pages of paper at a cost of US$500k, but he did print out 106 volumes for an art exhibit in New York. The Print Wikipedia blog post notes that the table of contents occupies 91 volumes.
Presumably the same 4:1 reduction trick would bring this down to 325 000 pages, a mere 465 volumes or 325 reams at a cost of US$31,250. At the standard 80 g/m², an A4 sheet weighs 5 g, so this printout would weigh 812 kg. Staples sells 75 g/m² US Letter paper in a case of 10 reams for US$56, so the paper alone would cost US$1820. However, their acid-free paper costs US$22 per ream and weighs 89 g/m², which would raise the price to US$7150. Presumably you could get the weight down by using thinner paper, as was normally used for encyclopedias and dictionaries, perhaps at a higher dollar cost. Wikipedia says “onionskin” typically weighs 25–39 g/m²; 30g would lower the weight to (* 325 500 (/ 30 16.0) .001) = 305 kg. Amazon lists a “9-pound” “FIDELITY” brand onionskin bond paper for US$25 per ream, weighing 3 pounds per ream, which works out to about 34 g/m². It’s 100% wood pulp, but despite that, it’s buffered, pH neutral, and 100-year rated. (1000 years is readily achievable.). If it’s 50μm thick, then the shelf length in A4 size would be (* 325 500 50 .001 .001) = 8.125 m.
(Working out from the shitty basis weight standard, the basis ream for bond paper is 500 sheets of 17" × 22", which is how we know it’s 34 g/m².)
Here in Argentina, the terminology seems to be “papel alcalino”, and 75g “papel alcalino” costs only about US$10 per 500-sheet A4 ream. This would bring the paper cost of the whole Wikipedia project to US$3250.