This week, Microsoft’s Bing search engine was the subject of accusations that they were exporting Chinese censorship to the rest of the world. And a few days later it still isn’t entirely clear what is happening with Bing. GreatFire.org (and Herdict partner) continues to assert that Bing is censoring Chinese-language search results even for users outside of China, while Microsoft asserts that differing search results are due to the different algorithms applied to Chinese and English language searches, not censorship. But what this whole kerfuffle demonstrates is how it is getting harder to identify censorship; the more companies like Google and Microsoft use opaque, proprietary algorithms to reshape the web they present to us, the less we understand about the results we see.
GreatFire’s blog post provides a good analysis of what they believe is occurring. Putting aside the causes, GreatFire has observed the following behaviors:
- Both international and Chinese versions of Bing are now displaying partial or generic censorship notice when viewed in the United States for certain searches.
- Searching for FreeWeibo on Chinese Bing in both China and the US triggers a censorship notice.
- Searching for FreeWeibo on Chinese Bing in Both China the US returns the FreeWeibo Facebook page but not a link to FreeWeibo itself.
- Searches for identical terms on Chinese language Google return more foreign media links than searches on Chinese language Bing, which returns more state media links.
GreatFire suggests all of these behaviors are evidence of Microsoft extending the censorship for Chinese results in China to Chinese results regardless of a Bing user’s location. Microsoft offers a more benign explanation:
The reason results are different for Chinese and English queries however, is because searches in different languages are fundamentally different queries. A result may show lower in one language versus another for a variety of reasons, such as fewer users choosing that link in English results compared to users who searched in another language.
Microsoft’s algorithm is a black box (as is that of every search engine), meaning it is nearly impossible for us to determine the truth. There was a time when search results were fairly comparable; I could compare two sets of results, and draw conclusions of censorship if links were missing from one and not the other. But as search results are increasingly customized by geography, language, and user, this is becoming harder and harder to do. I can’t even compare search results with my officemate five feet away, let alone search results with someone halfway around the world. Thus we are left guessing at whether these discrepancies in Bing’s results are censorship or legitimate differences in methodology.
One frightening possibility is that the algorithms that Microsoft has created could unintentionally reinforce Chinese censorship. Search engines strive to show users the most relevant and helpful results. Showing links to sites that users can’t access is, for most people, neither relevant nor helpful. If most Chinese-language Bing users are in China, the Chinese language algorithm might begin to disfavor search results for sites that are blocked in China. And if it the algorithm is very sensitive, it may eventually stop showing links to censored sites altogether. In theory this could happen without any active or intentional removal of these sites on the part of Microsoft employees; Bing’s artificial intelligence could simply “learn” that almost no one using Chinese-language Bing is clicking the links that GreatFire believes are censored.
Thus the problem isn’t just that it is getting harder to track this kind of censorship (although it is), but that it is getting harder to define censorship. Is it censorship if Microsoft isn’t showing links in Chinese-language Bing that most of its Chinese-language users can’t even access? Is it censorship if this is the result of an AI learning algorithm that is trying to serve up the results that would be most relevant?
It may turn out that GreatFire is correct and this is active censorship. And if that is the case then Microsoft will have to explain how their behavior is consistent with their obligations under the Global Network Initiative. But if this is something else, Microsoft (as well as other search engines) should begin to think about how they can be more transparent about their search algorithms. In the absence of transparency, blaming an opaque search algorithm will become an easy excuse for all kinds of behavior that threatens an open Internet.