Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

AI service gives Wikipedians ‘X-ray specs’ to see through bad edits

blog.wikimedia.org

Post reply on HN
Print view View original

21–30 of 52 posts

Re: AI service gives Wikipedians ‘X-ray specs’ to see through bad edits

#21

Post by Chris2048 »

cruise02 wrote: post #7
exo762 wrote: post #4

This is very peculiar. New tool is introduced, which joins a set of existing tools (bots) which address the same problem. Some comparison of quality and robustness is expected. Instead author gives us some insight into his/hers source of inspiration: > A feminist inspiration > “Please exercise extreme caution to avoid encoding racism or other biases into an AI scheme.” > Wnt (from The Signpost) And few lines below we…

If a bot can detect that you "don't buy feminist agenda," then it is evaluating based on content, not on sex or race.

The two can correlate, so "based on content" isn't enough, "based objectively/justifiably on content" is.

e.g What if the basis was about the non-/usage of gender-neutral pronouns?

Re: AI service gives Wikipedians ‘X-ray specs’ to see through bad edits

#22

Post by Chris2048 »

HarryHirsch wrote: post #8
bglazer wrote: post #6

Earlier quoted context omitted.

I agree the "feminist inspiration" part is a confusing non-sequitur. I think you're missing what they're trying to say, though. I interpreted that section as "we're making our algorithms and data open so that everyone can be sure they aren't implicitly racist or sexist". Suppose, for example, that a user repeatedly rejected edits for anyone with a noticeably feminine username. Wikimedia wants to be sure that their al…

Attempting to have processes that aren't implicity racist or sexist is a feminist practice?! If that doesn't qualify as an implicitly sexist statement I wouldn't know what would.

I think poster is skeptical that this is the attempt [of feminism] - no need to sling mud.

Re: AI service gives Wikipedians ‘X-ray specs’ to see through bad edits

#25

Post by zellyn »

MollyR wrote: post #24
zellyn wrote: post #23

Welcome to the spammer vs machine learning arms race, Wikipedia. :-) (Former YouTube Abuse team member)

Can't spammers also use machine learning, or maybe I misunderstood something.

What I meant is that I expect a huge initial win, as this catches 80-90% of spammy edits, until the spammers figure out how to get around it. Hopefully the constant stream of human-created training data (in the form of reverted or non-reverted edits) helps with this. But undoubtedly there will need to be some continual feature engineering to stay ahead.

Re: AI service gives Wikipedians ‘X-ray specs’ to see through bad edits

#26

Post by logicallee »

tptacek wrote: post #20
logicallee wrote: post #19

Earlier quoted context omitted.

Well, suppose something told me what I would decide if I gave any matter 15 seconds of consideration. It would be an invaluable tool I might use often, for example for sorting and classifying, throwing out trash and so forth. For example, it would be a fantastic spam classifier, since even after 15 seconds there is no doubt. I could use it to get through hundreds of spam messages easily, or delegate things to appropr…

It's 15 seconds of consideration you have to give to huge volumes of edits, so singling out the ones where that 15 seconds is likely to be profitable for an editor is extremely helpful. (I have no idea how well the scoring works in practice). Remember, the editorial tasks we're talking about here are enervating.

fair enough, I guess I don't like the "x-ray specs" analogy. Their example is a disruptive edit[1] where a URL is replaced by LLAMAS GROW ON TREES

perhaps rather than say it gives editors "x-ray specs", it would be fair to say it gives editors caffeine :)

[1] - https://wikimediablog.files.wordpress.com/2015/11/diff_64221...

Re: AI service gives Wikipedians ‘X-ray specs’ to see through bad edits

#27

Post by HarryHirsch »

Chris2048 wrote: post #22
HarryHirsch wrote: post #8

Earlier quoted context omitted.

Attempting to have processes that aren't implicity racist or sexist is a feminist practice?! If that doesn't qualify as an implicitly sexist statement I wouldn't know what would.

I think poster is skeptical that this is the attempt [of feminism] - no need to sling mud.

It's the Wikimedia Foundation that is heavily invested in the more radical parts of feminism - they did sponsor the Ada Initiative, who in many people's opinion qualify as fringe. It was them who called open data feminist ideology. Is all of science suddenly feminist? After all it depends on exchange of data. That's plain barmy.

Re: AI service gives Wikipedians ‘X-ray specs’ to see through bad edits

#28

Post by halfak_wmf »

zellyn wrote: post #25
MollyR wrote: post #24

Earlier quoted context omitted.

Can't spammers also use machine learning, or maybe I misunderstood something.

What I meant is that I expect a huge initial win, as this catches 80-90% of spammy edits, until the spammers figure out how to get around it. Hopefully the constant stream of human-created training data (in the form of reverted or non-reverted edits) helps with this. But undoubtedly there will need to be some continual feature engineering to stay ahead.

Yeah. We're actively working with patrollers to chase down mistakes and retrain the models. This is a system that will require continual improvement. Luckily, there's a lot of researchers working on this problem so we're drawing from a lot of brilliance beyond our little team at the Wikimedia Foundation.

Re: AI service gives Wikipedians ‘X-ray specs’ to see through bad edits

#29

Post by halfak_wmf »

jacquesm wrote: post #9
kenrick95 wrote: post #5

Earlier quoted context omitted.

Not really, but one can report a false positives and false negatives here: https://meta.wikimedia.org/wiki/Research_talk:Revision_scori... Since it is activated on Indonesian Wikipedia (I helped them to extend the tool to Indonesian language), I noticed that this tool can hardly capture an obvious vandalism yet. Some other edits are hard to verify whether it is vandalism or not, even by human. I believe this is a wor…

> Not really That's worrisome then. You'd expect a bit more rigor before putting a bot like that into production.

Actually, we openly report a wide variety of statistics about our models. E.g. if you go to https://ores.wmflabs.org/scores/enwiki, the service lists out the models that are available for English Wikipedia along with test statistics. You can also browse our test statistics in our documentation. See https://meta.wikimedia.org/wiki/ORES/damaging#English_Wikipe...

Currently, we're optimizing for AUC since it (1) is hard to cheat -- unlike accuracy or precision/recall individually and (2) it represents the ranking problem well.

See https://en.wikipedia.org/wiki/Receiver_operating_characteris...

"[AUC] is equal to the probability that a classifier will rank a randomly chosen positive instance higher than a randomly chosen negative one"

Re: AI service gives Wikipedians ‘X-ray specs’ to see through bad edits

#30

Post by jacquesm »

halfak_wmf wrote: post #29
jacquesm wrote: post #9

Earlier quoted context omitted.

> Not really That's worrisome then. You'd expect a bit more rigor before putting a bot like that into production.

Actually, we openly report a wide variety of statistics about our models. E.g. if you go to https://ores.wmflabs.org/scores/enwiki , the service lists out the models that are available for English Wikipedia along with test statistics. You can also browse our test statistics in our documentation. See https://meta.wikimedia.org/wiki/ORES/damaging#English_Wikipe... Currently, we're optimizing for AUC since it (1) is har…

Thank you. 75%, that's not bad but it's also not exactly 'x-ray eyes' grade yet. This is a pretty tough problem you're tackling, I can see now why it is not a bot but an augmentation device. It would be nice to track the accuracy over time, to see if the vandalizers wise up to how the bot makes the decisions to game it.
Post reply on HN