Google’s Wikipedia-killer, Google Knol, has passed some anniversaries; in January 2009, it reached its 6-month anniversary. 6 months is a fairly good time-period to look back—long enough that the hype has died down, and for big things to have been done (8500 hours is a long time online).
How had it done by then? Faster than Citizendium, even, it was dead. As a long-time Wikipedia editor, one might question my objectivity. So we’ll see what a relatively trusted third-party, Ars Technica, says:
What happened to Knol? Announced by Google in late 2007 and launched in July 2008, the site was meant to bring more credible (read: not written by anonymous Wikipedians) “knowledge units” to the web, and it would allow the authors to cash in on their work. But it’s 2009, and Knol appears to be notable largely for its non-notability…Now the bad news: no one’s reading the site, and it’s awash in poor content.
As for the quality of the content, Google’s attempt at monetizing (both for itself and for its authors) the Knol entries has had a perverse effect. While it has attracted plenty of detailed commentary from learned professionals, it’s drowning in plenty more that is basically spam, plagiarism, or a stub, thrown up in the apparent hope of making some quick cash. (Though because of point number one, that’s not happening, either.)
Take “Barack Obama”, for instance. A search for his name brings up 809 entries; since most Knol users appear to write their own entries rather than add to others (for which no compensation is forthcoming), the proliferation of entries is inevitable. And it’s not at all clear that the best ones are rising to the top.
At six months, Wikipedia only had a few hundred contributors and ~6,000 articles; but Knol had ~100,000. Surely this demonstrates that Knol has met with some sort of success?
Well, Knol ‘knols’ do not correspond to Wikipedia articles. Knols repeat themselves. Worse, they’re often spam, copyvios, repetitive. The true count of knols that are original to Knol and unique is probably very low—quite possibly down to ~6,000.
But getting bogged down in article count omits a major metric, and perhaps the most important metric: rate of growth. Wikipedia at 6 months could claim 6,000 articles, but what’s important is that just 5 months (a little less than twice as) later, the article count had more than doubled to >13,000. This was the knee of the curve that would lead to Wikipedia’s >3 million articles (for up-to-date statistics, see Special:Statistics).
Another profound point is that if Knol is only equal to Wikipedia, then it is a worse model than Wikipedia.
We should expect more of Knol than of Wikipedia at similar stages! Knol has, by virtue of its position in time, numerous advantages over Nupedia/Wikipedia. We should expect a lot more.
8 years of hardware advances; ~5 iterations of Moore’s law.
8 years of wiki development, demonstrating dead ends, the good ideas, & what remains to be improved. Imagine if Knol had to start with the state of the art in 2001. It would be truly gruesome. (Anyone looked at the very old Wikipedias in Nostalgia, or old Usemod wikis like Ward’s Wiki? They’re hideous and unusable! They make me quite grateful for 2009 MediaWiki with all its modern conveniences.)
The backing of a commercial juggernaut, with its many benefits:
- world-class hosting & technical support
- which they did not have to develop
- long-term backing (Bomis turned out to support Wikipedia for a good 5 years, but there was no guarantee)
- world-class software development resources1
- Massive publicity. At launch Knol had infinitely more publicity than did Wikipedia; much of this buzz was due simply to who the backer was.
A public educated to read wikis, and to use them.
This is a greater advantage than it seems. How many people could Wikipedia hope to draw on at day 1—that cared even a little about Free content, that knew what a wiki was, that wouldn’t dismiss it as hopeless, and had an editing familiarity with wikis? Darned few. We had to constantly evangelize and educate people about wikis, and by dint of unremitting effort create the English Wikipedia and make it interesting and valuable enough that people would contribute who didn’t fulfill any of those criteria. En was the existence proof that large-scale wikis were possible and valuable. Knol, on the other hand, can draw immediately on that pool of people Wikipedia created.
A model targeted directly at people unhappy with Wikipedia. Are you an expert tired of ‘anti-expertism’ on Wikipedia? Why try to get along with those bumpkins when you could have your own article completely to yourself on Google Knol (and get paid for’t)? Wikipedia appealed to those unhappy with Nupedia. Nupedia when Wikipedia launched was a lot smaller than Wikipedia was when Knol launched. I think this pool of possible contributors was thus also much larger for Knol than it was for Wikipedia.
With Wikipedia by this point, the basic concept of collaboration had been proved. With Knol, we see only the divisiveness of the payments system, and a few isolated authors striving on their own.
The important thing here is that people work on different things. No article writer on Wikipedia is discouraged by the fact that, say, Tim Starling is being paid for his system administration and PHP hacking—they don’t want to do it and are happy to exchange donations for the practical results of Tim’s labor viz. high uptime and faster servers. But if, say, every FA author got 1000 dollars, then perhaps writers of ordinary articles will get discouraged, or the bitter battles over what will be an FA or not will burn out the best editors.
Intuition suggests, and some results support, the hypothesis that paying only some contributors can “crowd out” other unpaid contributions (through shifting the reason for contribution from intrinsic to extrinsic motivation). See for example:
“Pay Enough Or Don’t Pay At All”, Gneezy & Rustichini 2000 found that “offering money did not always produce an improvement: subjects who were offered monetary incentives performed more poorly than those who were offered no compensation”; the theoretical model & experimental lab results have been extended to some interesting real world results:
“Crowding Effects: How Money Influences Open Source Projects and its Contributors”, covering Linux results
UP researcher Alexey Komissarouk on how paying more on Mechanical Turk led to more lying on the assigned tasks, even by the most highly rated Mechanical Turkers
There may be no happy medium between paying only for things which won’t get done otherwise (tedious system administration), and paying for most everything; I think Knol is in that unhappy middle.
How did Knol fare since I diagnosed its high levels of failure and predicted general doom in January 2009? Google continued to update the site, but then stopped any updates in December 2009, and the site stagnated over the 94 weeks since. Intermittent technical problems have lead to the accusation that Knol has been abandoned—presumably for hotter initiatives like Google Wave, which was publicly announced, released, and abandoned during the interval that Knol has stagnated; or Google+, which looks like it will have more staying power than its predecessor Google Buzz.
User activity is similarly pretty low. On 13 October 2011, I took a look at Knol’s “What’s New” section on its front page.
|3,000||83||6 minutes ago|
|900||23||45 minutes ago|
|90||20||One hour ago|
|80,000||59||One hour ago|
|10,000||58||5 hours ago|
|23,000||74||5 hours ago|
|7,000||90||8 hours ago|
|1,000||29||9 hours ago|
|19,000||135||9 hours ago|
|300||7||10 hours ago|
Over half a day for the entire site, 10 knols were changed in some fashion; edits probably are not huge updates given the relatively large number of revisions like 135 or 90. A quick-and-dirty estimate of the userbase: 1 edit per hour, each by a unique user, who edits on average once a week implies a userbase of .
Traffic-wise, one of them has a respectable 80k views (an accounting knol), but this seems to be lifetime views. Wikipedia, of course, has more like 10 edits a second, and even a dry article like Accountancy will pick up ~62k views a month. (I couldn’t find a direct equivalent for that knol—Wikibooks has so many accounting-related texts, and
stats.grok.se doesn’t cover Wikibooks.) Another amusing comparison is seeing how often Knol is submitted to Reddit versus Wikipedia submission. In August 2009, TechCrunch reported on Knol’s declining traffic as measured by Quantcast; in October 2011, Quantcast’s estimates shows Knol cycling from a high of <300k monthly visitors to a low of ~170k, with ~260k at the endpoint in July, although it appears to have never again reached its January 2009 peak of 300k monthly visitors. (For comparison purposes, one site claims that the number of global Internet users increased from 1.596 billion in March 2009 to 2.110 billion in June 2011; Wikipedia continues to remain popular, #6 worldwide according to Alexa.)
So even in absolute numbers, Knol has stagnated in most every respect. Some of the links above are from months or years ago—no one cares enough about Knol even to discuss it. (Somewhat like Citizendium.) Knol has all the characteristics of a doomed Google service, which leads to an interesting question: when will Google kill off Knol?
To predict, we should consider past examples: services can linger for a very long time, entirely neglected, or die very quickly. Which group does Knol fall into?
Yahoo: Geocities is very like Knol in being a collection of public content created by individuals and hosted on a corporate site, and was ignored by Yahoo for almost a decade before its shockingly abrupt shutdown, Delicious.com did little better while Flickr is still with Yahoo (and still neglected).
- Google Answers: public content; closure announced November 2006, shut down December 2006. All the archived question-answers and pages are still hosted by Google.
- Google Base: public content appearing in Google searches; visibility eliminated October 2009, APIs deprecated December 2010. Unclear where the content went—seems to’ve been incorporated into other Google services.
- GOOG-411: private service; closure not previously announced, November 2010. The content is private and folded into Google’s audio analysis algorithms.
- Google Lively: introduced July 2008, closed December 2008. Apparently little or no user-generated content, and the content/codebase is unavailable.
- Google Mashup Editor: closure announced January 2009, closed August 2009. User-generated, their content apparently made available indefinitely on
code.google.com(unclear how useful that would be to the users)
- Google Page Creator: closure announced 2007, user-signups disabled September 2008, closed 2009. User-generated, all content shifted to hosting on Google Sites and presumably still available.
- Google SearchWiki: opened November 2008, closed March 2010. User-generated but mostly private, the content (such as it is) is available to users in private.
- Google Video Marketplace: announced 2006, closed August 2008. All content unavailable, but the content appears to have been completely commercial, and none of the content seems to be unique, so the disappearance isn’t a big concern.
- Google X: something of a joke or mistake; irrelevant.
- Google Code Search: closure announced October 2011, closed January 2012. Presumably entirely unavailable, but as a search utility, it was private use only.
- iGoogle social features & Google Buzz: closure announced October 2011, closed November/December 2011. User-generated content, but the posts will remain publicly visible and users can download their material.
- Jaiku: closure announced October 2011, closed January 2012. User-generated content; apparently Google will not continue to host the content like Buzz or Page Creator, but the codebase was open-sourced in 2009 and Google will let users download their own material.
- Academic API access: closure announced October 2011, closed January 2012. Private, for academics only; no user-generated or public aspects. (Presumably most academics using it have published their results already.)
Looking through the most relevant examples, Google tends to shut down services fairly fast and is surprisingly good about preserving data, in line with its data liberation views. Other examples are relevant: Google eliminated file storage for its mailing lists, it gave the lists something like half a year of lead-time, and its Usenet archives have been poorly maintained since 2001 without being eliminated.
That it might do so is not a great challenge: disk space is cheap—fractions of pennies per gigabyte—and Google could shift Knol to all-static content, eliminating any software maintenance overhead. Unlike Google Notebook, knols are inherently public works, meant for public distribution and consumption, so it may be better to compare Knol against Google Groups rather than Google Notebook. Knol has not been shut down yet 2 years after its lack of noticeable success became clear, suggesting that it won’t be shut down any time soon. Weighing Google’s usual announce lead-time, the public nature of the content, its survival until now, and so on subjectively, these probabilities feel about right:
- Google Knol will be shutdown in 2011: 10%
- Google Knol will be shutdown in 2012: 25%
- Google Knol will be shutdown in 2013: 25%
- Google Knol will be shutdown in 2014: 20%
- Google Knol will be shutdown in 2015: 15%
- Google Knol will be shutdown in 2016: 5%2
- Google Knol’s shutdown will leave the knols’ content publicly accessible at some URL: 90%
(In March-May 2013 I conducted an extensive study of Google shutdowns to come up with some predictive model of shutdowns; while obviously Knol had been shutdown already, I still used the model to calculate a 5-year survival estimate for Knol from the day the shutdown happened, which turned out to be a suitably-low 43%.)
On 22 November 2011, Google put the speculation to rest when it announced:
We launched Knol in 2007 to help improve web content by enabling experts to collaborate on in-depth articles. In order to continue this work, we’ve been working with Solvitor and Crowd Favorite to create Annotum, an open-source scholarly authoring and publishing platform based on WordPress. Knol will work as usual until April 30, 2012, and you can download your knols to a file and/or migrate them to WordPress.com. From May 1 through October 1, 2012, knols will no longer be viewable, but can be downloaded and exported. After that time, Knol content will no longer be accessible.
This immediately resolved the above 7 predictions. I was correct to most expect Knol’s shutdown to come in 2012 or 2013 (and not earlier or later), however, my final prediction signally failed—the announcement is clear that the knol content will not be hosted indefinitely but will be deleted3 (‘accessible’ being an euphemism). Why did I blow it? My reasoning was based on one of my standard tools while predicting: the Outside View. Google had kept public content hosted in the past when it shut down projects (see the previous list), so it would probably continue to do so. The Outside View technique fails, however, when the same fundamentals that give it predictive power do finally change and a new regime is entered. In retrospect, there may have been such a regime change within Google when Larry Page became CEO in April 2011 and began pushing for a Google-wide focus on social networking, with a ruthless approach to projects which were not succeeding or connected to social network. This was all laid out in Wired’s “Inside Google+—How the Search Giant Plans to Go Social”, which I read at the time but examined more for how Google+ might differ from Facebook than how it might affect allocation of Google resources.
Not to denigrate the efforts of Magnus Manske and Tim and all the other MediaWiki developers over the years, but one simply expects more of Google’s supposedly world-class full-time developers using the famous Google infrastructure.↩︎
I can’t really see Knol surviving completely intact until 2017; I’m sure enough that it’ll be changed somehow before getting near its 10th anniversary.↩︎
Thankfully, Google being Google, they are being fairly generous in making it easier to save one’s stuff. Exporting still requires action on the author’s part, which will frequently not be taken (eg. if the author has died), and so hopefully the Archive Team will be able to do a clean sweep.↩︎