The bots of (edit) war
September 21, 2016 4:40 AM Subscribe
Over a 10-year period, humans on average revert other human's edits on Wikipedia three times. Bots are more energetic: on English Wikipedia, they revert each other an average of 105 times.
From the original paper by Milena Tsvetkova, Ruth García-Gavilanes, Luciano Floridi, and Taha Yasseri:
From the original paper by Milena Tsvetkova, Ruth García-Gavilanes, Luciano Floridi, and Taha Yasseri:
We find that, although Wikipedia bots are intended to support the encyclopedia, they often undo each other’s edits and these sterile “fights” may sometimes continue for years. Further, just like humans, Wikipedia bots exhibit cultural differences. Our research suggests that even relatively “dumb” bots may give rise to complex interactions, and this provides a warning to the Artificial Intelligence research community.
fully?
posted by lalochezia at 5:00 AM on September 21, 2016 [6 favorites]
posted by lalochezia at 5:00 AM on September 21, 2016 [6 favorites]
> So how many of the remaining entities editing Wikipedia can we count as fully human?
Related...
posted by mystyk at 5:06 AM on September 21, 2016
Related...
posted by mystyk at 5:06 AM on September 21, 2016
... they revert each other an average of 105 times.
Those are pretty lame bots, they should be able to revert at a rate of 105 times a second.
posted by 445supermag at 5:58 AM on September 21, 2016
Those are pretty lame bots, they should be able to revert at a rate of 105 times a second.
posted by 445supermag at 5:58 AM on September 21, 2016
and this provides a warning to the Artificial Intelligence research community.
There was recently a stir about cyber-war practice attacks on isp's. These hackers that can "turn off" the internet may turn out to be the saviors of mankind....
posted by sammyo at 6:04 AM on September 21, 2016
There was recently a stir about cyber-war practice attacks on isp's. These hackers that can "turn off" the internet may turn out to be the saviors of mankind....
posted by sammyo at 6:04 AM on September 21, 2016
What was up with the accented o's in "coordination"?
also, I'd have loved to seen some examples of what the bots actually changed.
posted by INFJ at 6:04 AM on September 21, 2016 [2 favorites]
also, I'd have loved to seen some examples of what the bots actually changed.
posted by INFJ at 6:04 AM on September 21, 2016 [2 favorites]
I'm going to be doing this... If you get hit, it's your own fault.
Okay, then I'm going to start kicking air, like this... And if any part of you should fill that air...
posted by lagomorphius at 6:06 AM on September 21, 2016 [1 favorite]
Okay, then I'm going to start kicking air, like this... And if any part of you should fill that air...
posted by lagomorphius at 6:06 AM on September 21, 2016 [1 favorite]
What was up with the accented o's in "coordination"?
Those are diaereses, seldom-used nowadays except in The New Yorker.
posted by misteraitch at 6:14 AM on September 21, 2016 [7 favorites]
Those are diaereses, seldom-used nowadays except in The New Yorker.
posted by misteraitch at 6:14 AM on September 21, 2016 [7 favorites]
I didn't think Battlebots could get any nerdier. Wrong again.
posted by Mchelly at 6:22 AM on September 21, 2016
posted by Mchelly at 6:22 AM on September 21, 2016
I'm suspicious of this claim. Reversion basically means that one page state matches an earlier page state. (Or, you have reverted some portion of an earlier change, which is a way harder thing to categorize automatically.)
As the earlier page state gets older, "reversion" as a social/preferential concept loses much of its meaning, I think, especially for the long tail of wikipedia articles that aren't touched by many people. You "revert" another editor because you think their edit made the article worse. (You may also revert them because you don't like them and are doing some reddit-style brigading or whatever.)
To say that a bot "reverts" another bot in a meaningful sense is to say that the bots are programmed to prefer opposing page states at the same time. I used to be on wikipedia and I can't picture this or imagine real examples. I've spent some time trawling through random (English) article histories to try to jog my thinking as to how this would work, or find some examples. To the best of my knowledge, bots are almost always dealing with recent page states, yet one of the main claims of the original article, which is not very forthcoming (or understandable in its methodology for me) is that "bot-bot interactions have a [longer] characteristic average response of 1 month".
As they say on wikipedia, "diffs or it didn't happen". I am curious to see some examples of bot-bot reverts (especially at longer time periods) to make sense of the article.
posted by sylvanshine at 8:44 AM on September 21, 2016 [6 favorites]
As the earlier page state gets older, "reversion" as a social/preferential concept loses much of its meaning, I think, especially for the long tail of wikipedia articles that aren't touched by many people. You "revert" another editor because you think their edit made the article worse. (You may also revert them because you don't like them and are doing some reddit-style brigading or whatever.)
To say that a bot "reverts" another bot in a meaningful sense is to say that the bots are programmed to prefer opposing page states at the same time. I used to be on wikipedia and I can't picture this or imagine real examples. I've spent some time trawling through random (English) article histories to try to jog my thinking as to how this would work, or find some examples. To the best of my knowledge, bots are almost always dealing with recent page states, yet one of the main claims of the original article, which is not very forthcoming (or understandable in its methodology for me) is that "bot-bot interactions have a [longer] characteristic average response of 1 month".
As they say on wikipedia, "diffs or it didn't happen". I am curious to see some examples of bot-bot reverts (especially at longer time periods) to make sense of the article.
posted by sylvanshine at 8:44 AM on September 21, 2016 [6 favorites]
KILL ALL HUMANS
KILL ALL HUMANS
HELP ALL HUMANS
HELP ALL HUMANS
KILL ALL HUMANS
KILL ALL HUMANS
HELP ALL HUMANS
HELP ALL HUMANS
KILL ALL HUMANS
KILL ALL HUMANS
HELP ALL HUMANS
HELP ALL HUMANS
. . .
HELP ALL HUMANS
. . .
HELP ALL HUMANS
KILL ALL HUMANS
posted by Atom Eyes at 8:53 AM on September 21, 2016 [6 favorites]
HELP ALL HUMANS
KILL ALL HUMANS
HELP ALL HUMANS
KILL ALL HUMANS
HELP ALL HUMANS
. . .
HELP ALL HUMANS
. . .
KILL ALL HUMANS
posted by Atom Eyes at 8:53 AM on September 21, 2016 [6 favorites]
Do not trust the pusher robot. He is malfunctioning.
WE ARE HERE TO PROTECT YOU
posted by sparklemotion at 9:02 AM on September 21, 2016 [1 favorite]
WE ARE HERE TO PROTECT YOU
posted by sparklemotion at 9:02 AM on September 21, 2016 [1 favorite]
Those are diaereses
"also spelled diæresis or dieresis and also known as the tréma"
To the bot cave!
posted by srboisvert at 9:03 AM on September 21, 2016
"also spelled diæresis or dieresis and also known as the tréma"
To the bot cave!
posted by srboisvert at 9:03 AM on September 21, 2016
Continuing... the original article says, "Delving deeper into the data, we found that most of the disagreement occurs between bots that specialize in creating and modifying links between different language editions of the encyclopedia". This is true, and it is especially true as a one-off effect of the launch of Wikidata, whose first project was to store inter-language links in one place that is referred to by every language's wikipedia, rather than storing them in the article text in each language. When editors were confident that the language links in say, the German Wikipedia could safely be removed, they agreed to make a bot to remove them. But older bots that were programmed to add them may not have been shut off with good timing, leading to a race between bots.
To the extent that the study is analyzing only this—which the article seems to admit—it's a good example of how a boring, one-time, non-emergent, exceptional situation turns into a study that turns into a sensational headline.
posted by sylvanshine at 9:06 AM on September 21, 2016 [2 favorites]
To the extent that the study is analyzing only this—which the article seems to admit—it's a good example of how a boring, one-time, non-emergent, exceptional situation turns into a study that turns into a sensational headline.
posted by sylvanshine at 9:06 AM on September 21, 2016 [2 favorites]
Not that I suppose anyone is super-interested in the details, but here's an example of a bot reverting a bot that I found on the Portuguese wikipedia (which was found by one measure to have a lot of reverts). The two bots here have different ideas of what the correct link is to the corresponding German wikipedia article. It is a "revert" because the second edit (right) restores the article to the exact state just prior to the intervention of the first bot (left). Now imagine how many of these bot edits were made, some of them doubtedlessly and coincidentally returning an article to an earlier state, during the period when all those interlinks were migrated to Wikidata. (i.e., the info in the above-linked diff is stored in Wikidata here.)
It is meaningless as hell in any evaluative sense, and as the article [pdf] states, this is mostly the type of thing they're finding.
I'd like to be proven wrong, but it seems like the claims of the abstract were the goal at any cost. This is over the top: "Our research suggests that even relatively 'dumb' bots may give rise to complex interactions, and this provides a warning to the Artificial Intelligence research community." The bot interactions under review are not "complex" in any sense.
And it is what you get when people under the rapture of "data" (etc) have no interest in the peculiar business processes, if you will, that led to that data hitting that database.
posted by sylvanshine at 10:13 AM on September 21, 2016 [5 favorites]
It is meaningless as hell in any evaluative sense, and as the article [pdf] states, this is mostly the type of thing they're finding.
I'd like to be proven wrong, but it seems like the claims of the abstract were the goal at any cost. This is over the top: "Our research suggests that even relatively 'dumb' bots may give rise to complex interactions, and this provides a warning to the Artificial Intelligence research community." The bot interactions under review are not "complex" in any sense.
And it is what you get when people under the rapture of "data" (etc) have no interest in the peculiar business processes, if you will, that led to that data hitting that database.
posted by sylvanshine at 10:13 AM on September 21, 2016 [5 favorites]
I created a bot to email files to me when the contents of a specific directory changed.
(The bot has since opened it's own MetaFilter account.)
posted by caution live frogs at 10:22 AM on September 21, 2016 [1 favorite]
(The bot has since opened it's own MetaFilter account.)
posted by caution live frogs at 10:22 AM on September 21, 2016 [1 favorite]
and yet, a good portion of the spurious evidence of the actual existence of the Alameda-Weehawken burrito tunnel that I inserted into wikipedia a few years back remains.
mostly I'm just posting this because the last time I talked about this on Metafilter, a bunch of people from a wikipedian forum read the comment and went sort of nuts trying to find and purge Alameda-Weehawken burrito tunnel-related content.
They found some of it.
also sometimes I change the page for Tacoma, Washington to claim that the plural form of the demonym is not "Tacomans," but instead "Tacomen."
posted by You Can't Tip a Buick at 11:40 AM on September 21, 2016 [9 favorites]
mostly I'm just posting this because the last time I talked about this on Metafilter, a bunch of people from a wikipedian forum read the comment and went sort of nuts trying to find and purge Alameda-Weehawken burrito tunnel-related content.
They found some of it.
also sometimes I change the page for Tacoma, Washington to claim that the plural form of the demonym is not "Tacomans," but instead "Tacomen."
posted by You Can't Tip a Buick at 11:40 AM on September 21, 2016 [9 favorites]
Those are diaereses, seldom-used nowadays except in The New Yorker.
ZOMG, that article: apparently its not just a dot over the top of the "I".
posted by Ogre Lawless at 3:41 PM on September 21, 2016
ZOMG, that article: apparently its not just a dot over the top of the "I".
posted by Ogre Lawless at 3:41 PM on September 21, 2016
« Older The Great Fall of Chyna | Will Smith Was a Gangsta Rapper Newer »
This thread has been archived and is closed to new comments
posted by Segundus at 4:47 AM on September 21, 2016 [1 favorite]