Last week marked a watershed moment for two stars of the business world. By chance, an abrupt end seemed imminent for the careers of Time Warner CEO Richard Parsons and Merrill Lynch CEO Stanley O'Neal. Both had reached pinnacles rarely scaled in corporate America by African-Americans. And their fates dominated business headlines worldwide last Friday, Oct. 26. Online at Google News, however, the coverage was, in a word, shocking.
A keyword search for "Richard Parsons" generated a flood of stories about the executive, accompanied by a photo of two rhesus monkeys. Clicking on the image linked users to a story on neither monkeys nor Parsons. Instead, a speculative account on O'Neal's waning support among Merrill directors appeared on screen. Other than the lynch mob's noose and the Klansman's hood, few images of racism are as offensive to African-American as monkeys. Yet the bizarre juxtaposition of image and stories persisted through the week. And even after Google was specifically contacted this week, it continued.
An unstoppable racist hacker? Hardly, according to a Google spokesman. It was an inside job. The perpetrator, however, wasn't human. The search giant blames its computers and algorithms. Despite its cutting-edge advancements, Google is simply incapable of a performing a skill typically mastered by first graders: matching the right words with the right images. The problem generally has plagued Google since June when it introduced the "Image Version" of Google News to pair the top headlines with illustrative photos.
It's an ongoing problem. Recently Google News mismatched stories on Argentina's newly elected president Cristina Fernández with a photo of the California wildfires. A photo of the late Australia crocodile hunter Steve Irwin accompanied a report on the stock market. And thanks to the fast-breaking developments at Merrill Lynch, O'Neal managed to evolve. After he was forced to retire Tuesday in the wake of massive subprime losses, Google News illustrated the story with a 1939 photo of Katharine Hepburn taken during the filming of "The Philadelphia Story."
Google acknowledges the situation, but declined to openly discuss details of the primate episode. Surprisingly, it was even unwilling to go on record and explicitly disavow any racist motivation. "While we don't comment on individual stories on Google News," spokesman Gabriel Stricker told NEWSWEEK, "crawling thousands of sites across the globe is a complicated task, and we're confident that the quality of the crawled pages is extremely good for the vast majority of news sources on our site." In an oblique nod to the problem, he cites the need for "more work to be done," adding, "we're always working on improvements to Google News to ensure that the experience for all of our users ... continues to be great."
If the "Parsons" search result is, as Google indicates, merely a remarkably unfortunate techno blooper, the search giant's explanation in one respect is still as potentially unsettling as the primate imagery is distressing. Short of shutting down searches, the problem will persist for now. Google News relies almost exclusively on algorithms and automation to build its pages. For now, manually correcting one mismatch may fix one issue but simultaneously create many more. When asked when the problem would be resolved, a Google spokesman answered: "As soon as possible."
Despite Google's technical explanation, there's a twist to the Parsons-primate episode that could lead online users to suspect hacker involvement. Over the weekend, the company appeared to have corrected the mismatch, at least temporarily. At times, the primates photo was substituted with a photo of a formally attired Parsons posing with an actor in a Bugs Bunny costume. (The character is owned by Time Warner, which declined to comment for this story.) Later, however, the rhesus photo reappeared, supplanting Parsons and Bugs Bunny. Google denies it was hacked and insists the incident was again, simply a result of its computer systems.
This isn't the first time automation has managed to offend. In January 2006, as USA Today reported at the time, a feature on Wal-Mart's Web site that generated recommended purchases linked a "Planet of the Apes" DVD to films about African-Americans, including Martin Luther King Jr., Tina Turner and boxer Jack Johnson. The retail giant, citing errant automation, apologized to any offended customers and shut down the recommendation system.
Given the sheer size of what Google does, the chances of getting everything right all the time would appear to be statistically nil. In September, the latest data available, Google handled 38.2 billion searches--some 63 percent of the total 61 billion searches worldwide, says Andrew Lipsman, senior analyst with Internet measuring service Comscore. Of Google's total, Google News accounted for some 103 million searches. According to Google, it "crawls" 4,500 English-language news sites, and thousands more worldwide in other languages, to seize freshly posted stories and images for storage in Google's servers that it taps to respond to keyword searches.
Google won't disclose the scope or details of its mismatching problem. But the problem clearly seems to be associated with material from at least one of the world's major originators of news--Reuters, which declined to comment on the situation. The primates and associated Parsons and O'Neal stories, for example, were both retrieved by Google from Reuters--as were the mismatched wildfire photo and Argentine election story, as well as other examples.
Reuters makes its stories and photos available separately for retrieval by search engines, Google foremost. When the search giant dispatches a crawler to snare a story from the Reuters site, it simultaneously deploys a crawler to Reuters's photo gallery. But the image crawler isn't discriminating. It simply grabs the first photo in the gallery, never mind that the particular image often has nothing remotely to do with the text-based story.
Google says it is trying hard to fix the problem. But for now, it seems a primate could do a better job.