A couple weeks ago I was fortunate enough to be invited to a talk by Qi Lu at the University of Chicago's Computation Institute. The President of Microsoft's Online Services Group, Mr. Lu's talk was on opportunities for collaboration with academics on the Bing search engine. Microsoft had recently announced Facebook integration with Bing, and I was particularly interested in what he had to say about it.
Mr. Lu's talk-- and particularly the Q&A that followed-- provided some interesting answers to the often-contemplated question "What is Microsoft thinking?", and thoroughly convinced me that Bing is not the search engine for me.
On Facebook integration
Mr. Lu was excited about the possibility of using recommendations from friends-- an element of real-world decision-making-- to influence Bing search results. I first heard the news about Bing’s integration with Facebook in a blog post questioning the value of decontextualized "like" data, so I decided to ask him about it.
I figured it was a pretty soft-ball question, given the number of computational linguists in the room and the premise of "collaboration with academics". In the last year or so, services like twendz have emerged, providing automated analysis of whether tweets are positive or negative. Why not develop similar systems to analyze "Like" data? One could look at the follow-up comments on a news article or web page-- if there's content criticizing the article for being biased, or incorrect, or poorly-written, then the "Like" shouldn't be valued positively. Similarly, if someone has "Liked" a product or business so they can write critical messages on the product's Wall, those wall posts could be factored into the valuation of the "Like".
When I mentioned the ambiguity of "Like"-- given the many motivations for doing it, including gaining access to a product's Wall-- and asked what Microsoft was doing to re-contextualize the data in order to correctly assess its value, Mr. Lu didn't quite seem to understand what I meant. He agreed that "Like"ing an article can just indicate that the person finds it interesting (rather than necessarily liking the contents, particularly in reference to articles about tragedies), and he said that his group is looking into hiring social scientists to follow the emerging trends in people's behavior on-line and why they "Like" the things they do. I re-emphasized the point about "Like" being more than "this is interesting"-- it can be used to enable a form of protest. Mr. Lu's replied that people will always find ways to "abuse the system", but he feels that "the good will win out".
It's understandable that from the point of view of the business being criticized, "Like"ing a product in order to protest it amounts to "abusing the system", but I was taken aback to hear the same sentiment from a company that-- one would think-- would be interested in correctly valuing the data being used to influence their search rankings, regardless of the sentiment. Then again, the way Bing apparently does its search rankings to begin with doesn't exactly inspire my confidence.
Bing's search results: brought to you by the masses
A colleague described to Mr. Lu his recent experiment looking for a map of Budapest on Bing. He typed in Budapest, figuring the city's tourism page would come up first, and that would have some appropriate map. The day he searched, all the top results were about Angelina Jolie and Brad Pitt filming a movie in Budapest. As he scrolled, he saw a headline about Brad Pitt accidentally walking into the wrong house and, feeling guilty ("They'll think I found what I was looking for!"), he clicked on it.
Mr. Lu replied that Bing results depend on what people are searching for, stating that searches for "Columbia" right before the space shuttle disaster probably referred to the country, but immediately afterwards people were searching for the latest news, and the system has to adapt nimbly. And since my colleague clicked on something he found interesting, wasn't that, in a sense, a successful search?
Apparently with Bing, a dancing baby-- or whatever else is entertaining the masses at the current moment-- is always the right answer. "We don't rank our results," said Mr. Lu, seemingly with some pride. The Borg Collective of John Q. Public (which, as I look just now, has informed the default Bing drop-down that I'll be looking for 'Megamind movie') takes care of that for you.
How do you prevent your search results from being invaded by vacuous pop culture? Mr. Lu mentioned that Bing is very respectful of your privacy, but later noted that Bing tries to provide the best search results based on everything it knows about you-- which seems to suggest that if your interests and inclinations differ from those of the masses, you have to surrender some of your privacy to get reasonable search results.
The future of mobile search
Mr. Lu was enthusiastic about the strides Microsoft is making in turning data about the world around users into search queries, remarking that entering words into a box isn't the most natural way to search from a mobile device. Instead, why not take a picture of something and let that be the search input? If you take a picture of a business or church*, you could see the associated website, opening hours, menu, etc.
Personally, I despair at a future where such a system is the default input for search queries, nor am I sure that it would even work. Even if you're looking for a product to buy, it's likely because you don't have one and therefore can't take a picture of it. Maybe the hope is that you would take photos of billboard ads, to facilitate your easy purchase of things that companies want you to buy? If I'm looking up a restaurant to find the hours, chances are I'm not there at the moment to take a picture-- otherwise I could look at the hours posted on the door. And what if you're looking for something more abstract, like the history of touchscreens, and don't just want one to buy? There's already mobile apps that let you purchase an item based on a photograph; what else would Bing offer?
Profits, not products
The intensity of Mr. Lu's focus on profits took me aback. Before the talk started, we all introduced ourselves briefly, and the audience included a number of scholars from the Computation Institute (including linguists), a number of IT staff, two business school students, and someone from a nonprofit. Notably absent were the economists, but why should they attend what was ostensibly a talk about opportunities for collaboration on a piece of technology?
Knowing there were no economists in the room, Mr. Lu persisted in directing his offers for collaboration to these absent economists. He described the University of Chicago as a great school for economic theory-- while arguably true, I can't help but wonder if it wasn't a little insulting to the people who were in attendance. Mr. Lu wants to collaborate with scholars in developing a better model for a market to sell adwords for Bing. He referred to a second of human attention as the most valuable commodity, destined to become even more valuable as "great products designed especially for you!" compete for your attention. It seemed pretty clear that the focus of Bing is on profits-- getting people to see as many ads tailored to their demographic as possible, and developing more effective ways to extract money from advertisers-- and not on getting people the search results that they actually want.
Maybe Mr. Lu should wait until the Milton Friedman Institute opens at UChicago. He might just find the collaborators he's looking for there.
* It's a small thing, but I was struck by how Mr. Lu repeatedly used the example of looking for information about a church. While not as bad as getting the name of your host institution wrong, it's a detail that suggests some degree of unfamiliarity with UChicago. Everyone I've known in the Divinity School has been agnostic, if not atheist.
(As usual, the opinions expressed here are mine alone and not those of the University of Chicago, a place where the world’s leading minds come to be agreed and disagreed with.)
Bing pop can by Jason Walsh (CC BY)