Google RankBrain: Where Search Meets AI
Did you hear about the computer that beat the world’s best Go player back in March 2016?
Lee Sedol is a grand master of Go, the ancient Chinese game that has been played for millennia. But when he sat down against AlphaGo for a five-game series, he came out with just one win.
You might not think that this is such a big deal. After all, computers have been winning at chess for years. But Go is different.
This is a game that requires intuition. A game that, until now, has required more of a human element to master. It’s a very different game from chess.
The fact that a computer has now beaten the best player in the world is being seen as a turning point in the development of AI.
But while AlphaGo may have got Google all the headlines, it is another type of AI that is having a greater impact on the world at large right now.
Google Gets a Helping Hand from AI Machine Learning
You may have heard of RankBrain already. It’s been out for over a year, initially being used in just a few searches in early 2015, and later being pushed out to all searches.
But what is it?
While Google has remained fairly quiet on the details, it seems like it uses a combination of AI and machine learning to provide assistance with delivering relevant search results—specifically for complex long-tail queries that may not have been typed before.
What We Know So Far…
Google has a problem when it comes to delivering relevant results to search queries.
Despite all the advances it has made over the years, it can’t always deliver the best results. One of the reasons for this is that a large number of searches (15% according to some sources) are completely new.
So out of the 3 billion searches made on Google every day, a massive 15% have never been searched before.
Google’s algorithm has to interpret these new searches—and this can be especially difficult when the searches are made up of long and complex phrases.
So RankBrain is being used to assist with these searches by making up part of the Hummingbird algorithm, along with Penguin, Panda, and the other major parts of the algorithm.
Google has stated that it is a way for it to classify results with more accuracy. RankBrain helps with rare queries by processing them and delivering relevant results. It aims to work out what the searcher wants, even when the searcher does not use the exact words they are looking for.
It is essentially another signal out of the 200 or so signals that are already used to deliver relevant search results—but we do know that RankBrain is now the third biggest factor behind links and content.
How Does It Work?
The thing is, ever since Bloomberg first reported on the story, we haven’t heard much about it from Google—and it’s doubtful that we’ll get a lot more info anytime soon.
However, it’s likely that RankBrain helps to interpret complex queries by connecting longer, more complex queries to shorter, more common queries—essentially translating them.
It sees patterns between searches that seem unconnected, and this gives it a better understanding of future searches. It might also produce some kind of score to help determine the quality of webpages based on specific queries.
We also know that it uses both AI and machine learning. Whereas AI is focused on machine intelligence, machine learning is the process of computers learning on their own. RankBrain puts both of these to use.
The reason AI is required comes down to the sheer number of searches typed into Google that have never been used before, as mentioned already.
Google’s algorithm is constantly getting smarter, and it can tell the context of searches to show more relevant results. We also have the Knowledge Graph, launched in 2012, which contains a huge database of information.
So if you search for something like the fastest man in the world, you are likely to get results related to Usain Bolt, even though you don’t mention his name.
But RankBrain probably uses AI to take this one step further.
As people use longer and more complex queries, humans cannot handle the amount of data on their own—so RankBrain interprets the queries instead.
RankBrain probably identifies patterns between words to work out how they are related—and this helps improve future searches related to that topic.
Google has also clarified that RankBrain carries out its learning offline, where it processes historical search data and tests predictions so it can constantly improve them.
The Big Question: What Should You Be Doing?
RankBrain might sound like a pretty big deal, and in many ways it is. But actually, in terms of what you should be doing now that it is being used in search, the answer is not a lot.
There is no big rankings change with RankBrain, so this is not like a Penguin or a Panda. Despite it being used in searches, it is not something that is going to require any big strategy changes on your part.
Instead, just stick to the same rules as always.
What it all comes down to at the end of the day is improving the user experience.