Ok, maybe not strictly AI in the truest sense, but a few days ago, in the wake of new Google “over-company” Alphabet earnings reports, Google revealed that for several months already, an artificial intelligence system nicknamed “RankBrain” has been responsible for handling “a large fraction” of their daily user queries.
We’ve always know that about 15% of queries received by Google every day are unique. In this sense, Google means that they are queries which have never been received before. (Most of them are probably very long-tailed queries, or “conversational” queries, which will be phrased differently every time.)
Google’s investment into AI is also something that has been commonly known for some time, and there has been speculation in the past, that the ultimate end goal of Google search is in fact the development of true AI. That theory got a little boost a few days ago, when they confirmed that they are now using a machine learning system to handle those unique queries.
A Significant Signal
Greg Corrado, a senior research scientist at Google explained to Bloomberg Business that while there are hundreds of different signals that inform the algorithm that determines search results, in the few months that it has been active, RankBrain has become the 3rd most important of these signals. According to his explanation, although the other signals are based on insights of people in information retrieval, there has been no “learning” aspect to those signals. RankBrain is different.
The result of a year-long effort by a team of engineers, the system so far has shown a greater accuracy at predicting pages that the search engine would display for queries than human search engineers, with an 80% accuracy rate, compared to the 70% rate humans managed. Subsequent tests have found that excluding the system’s input has resulted in outcomes which they describe as being as bad as “forgetting to serve half the pages on Wikipedia.”
Not A Replacement
It’s important to keep in mind that this is not a new Google algorithm, or a replacement. This is just one part of the algorithm, which deals mostly with queries that Google has never seen before, and most importantly, learns from those queries. Processing 3 billion searches every day means that 15% is quite a serious chunk of daily queries.
RankBrain appears to allow Google to determine patterns between ostensibly unconnected searches, and learn how they are similar to each other, whichit can then use to deal more effectively with other queries that share those subtle similarities.
What Does RankBrain Mean For Search?
In short, we don’t know yet. In true Google fashion, despite them revealing this information, they still haven’t really said very much about it. There have already been a few speculative analysis of the patent that appears to be involved, but for now, the long (and even medium) term impact is unknown.
We can probably infer with a fair degree of safety though that it’s participation in the algorithm will increase over time, but for now, it just looks like they’re going to be really good at resolving those long and complex search queries, and getting better all the time.