Friday, November 27, 2009

Re: Google Adds Disclaimer Ad to Results for [Michelle Obama]

This is getting absurd. First, an arguably racist photo caricature of the US president’s wife, Michelle Obama, appeared in a search for her name, and Google decided to launch a disclaimer ad campaign on those results. Then that image got removed and disappeared for some of us, but the results were soon showing a text, printed by Google, which read “Related searches: michelle obama monkey” (I saw that particular screenshot on Chinese TV yesterday while walking into a shop... it’s certainly making the rounds). And now that suggestion bit disappeared, at least for some of us, but Google is showing the text: “Related searches: michelle obama ape”. I wonder if at this point, Google has started or will start to manually tweak the related searches feature through some blacklist.

By now we’re looking at a self-powering feedback loop, where reporting on an offensive or racist result starts triggering more potential offensiveness. We know Google automates this stuff so it’s just that there happen to be a lot of people apparently searching for michelle obama monkey, perhaps in conjunction with searching for michelle obama first, and perhaps in conjunction with having heard about the issue in the news before.

In all of this, I get the feeling that Google shouldn’t involve itself as a company starting to excuse themselves for these results within the results through ad campaigns (blog posts may be something different, I don’t know). Because there’s a lot of racist or otherwise offensive content out there that won’t get these disclaimers, so if they launch one such campaign it’s almost like they’re admitting responsibility for the fact that they mirror the web the way it is, and then we have to start wondering where the disclaimers for the other billion offensive results are. At that point, Google becomes the judge of what is deemed offensive, not a role one would like them to take on, and not even vaguely neutral, scalable or automated either.

Of course in this case one could also argue that the “manually automated” ranking is off to begin with. “Manual automation” is the double-edged nature of impossibly-to-be-fully-unbiased-even-with-best-intents human programmers defining rules which then get executed by computers. (It’s easily possible to define an offensive algorithm, and it won’t help that it’s automated – as an example, one can program a search engine that defines kkk.com as a “good” seed site, turning all outgoing KKK links into authority-boosting ones. It might be harder to create such offensive programs if one never manually defines data – no URLs, no word blacklists, no specific colors or tunes – and additionally never has manual results quality evaluations, though it’s still not impossible. I can try out 50 different perfectly real-world-agnostic rulesets and then pick the one set which I favor due to my bias and put that one live.)

In the end Google is an intentionally distorted mirror that tries to highlight authority as well as opinion diversity (we often get love and hate site showing up in the same top ten results, which is good for research even when the love or hate site may express an unpopular minority opinion, and even when you as searcher may find that opinion utterly stupid). Knowing that, we have the right to ask just what distortions Google is aiming for then. And it could indeed be that this result is off because such offensiveness, even if available on the web, is not relevant enough to make it into the number 1 position. Indeed if a caricaturing site would top the web results for the name of a well-known person, I’d wonder what’s broken with Google’s algo. Somehow, I think we came to expect less representative rankings for image searches – “OK, they throw loads of pics at me, perhaps in no particular order, and three or four may be the ones I was looking for... I can decide at a glance and am not looking for the one single authoritative picture, nor would I know how to define such authority to begin with.”

If we do consider the image search ranking to be based on relevancy and do expect authority though, then the current result might point out a need for an algorithm change in image search. But even then, we also know Google prefers scalable solutions, and whatever cause triggered the offensive pic to be on top potentially also caused a million of other non-offensive pics to be on top for other queries – change the algo or the fine-tuning data behind it, and you’re affecting a lot of searches, even when the waves this current splash makes have flattened.

[Thanks Mathias Schindler!]

Please comment in the existing thread.

Re: Google Adds Disclaime ... by Philipp Lenssen

>> More posts

Advertisement