Devils and Details
The Laboratorium 2013-03-30
There is a necessary, but necessarily tricky provision in Google’s commitment to let websites opt out of commercial search:
Exercise of this option will not … (2) be used as a signal in determining conventional search results on the google.com search results page.
Some such provision has to be in the commitment, because severe demotion in search rankings is the practical equivalent of delisting a site. Something on the thousandth page of results does not exist from searchers’ point of view. So riddle me this: how will this promise be enforced?
Delisting is binary; it can be observed by issuing a sufficiently precise search. If I search for “hamburgers in topeka ks” and I don’t see any organic Yelp results, that might be innocent. But if I search for “yelp reviews” and I don’t see Yelp, something has gone very wrong.
But demotions can be subtler. A devious Google engineer could kick Yelp to the fiftieth page of results for any search containing the name of a recognizable item of food, but leave “yelp” unaffected. Yelp might suspect that something is wrong, but not be able to prove it just by issuing search queries and observing the results. So a demotion can be much harder to detect than a delisting.
The demotion I just described would leave behind a smoking gun: the search algorithm would have a few lines of code that special-cased the Yelp penalty. IF Google has to show the FTC its algorithm, then FTC staff can — at least in theory — spot the malevolent tweak. How, however, would the FTC know to ask? And by what legal process would it be able to demand access to the algorithms?
These may be significant consequences of the FTC’s decision not to insist on a consent order. If the FTC suspects a violation of a consent order, it has contempt remedies available: it can run to court, reopen the investigation where it left off, and send in the geeks. But a mere voluntary commitment: for that, the FTC would need to spin up its enforcement apparatus from much closer to a standstill. Which leads me to suspect that the FTC isn’t actually planning to do so again anytime soon.
On the other hand, I can also make a strong case that Google could not as a company lightly afford to put this particular commitment into a consent order. Here’s why. Imagine the Google engineer is even more devious. There is no “punish Yelp” part of the algorithm. Instead, the engineers know which sites have opted out—as a class, they’ll often be easy to identify—and they look for general-purpose algorithmic tweaks that just so happen to disproportionately hurt the Yelps of the web, the sites that have opted out. There is no explicit discrimination, just intentional discrimination in effect.
If one is seriously concerned about this scenario, then every algorithmic change is potentially suspect. You need the FTC looking over Google’s shoulder constantly, and the FTC needs to be in a position to challenge any tweak it’s concerned about. But this is precisely the kind of comprehensive regulatory review of search algorithms that Google has been fighting ferociously against in the search-bias space. That’s the victory that Google won today: not to have to say “Mother may I” each time it changes its search algorithms. Full-on enforcement of promise not to retaliate for opting out of commercial search would involve the same kind of scrutiny as comprehensive observation by a Federal Search Commission.
So now I think I may have a handle on why today’s commitments took the form they did. Letting websites say no to commercial search without saying no to general search seems like a compromise, a self-contained solution to a self-contained problem. But the equivalence of demotion and delisting is a two-way street. To really make the opt-out stick, it’s necessary to say something about demotion as well as about delisting. But once you’re genuinely worried about demotion, there’s no escaping the full complexity and controversy of search algorithm oversight. Maybe not immediately, but quite possibly down the road. Google recognized this, and for that reason was intensely concerned that the opt-out commitment not be the nose that brings the regulatory camel into the tent.
For this reason, I predict that today’s settlement will be unstable. By the end of its five-year term, one of two things will have happened. Either the FTC and the world will have stopped caring about the opt-out, and it will be broadly accepted that search engines have a free, or at least not closely watched, hand here. Or the search-bias investigation will have started up again as the regulators insist on ongoing, perhaps quite intrusive, oversight.