So what the heck is RankBrain?


You won’t have failed to notice in the news this week certain headlines along the lines of “Google reveals new machine learning algorithm” or “Google turning its search over to artificial intelligence” or “The machines have taken over, run screaming for the hills.”

That last one is taken from my own street-pamphlet The Sensationalist Times.

The buzz (both warranted and unwarranted) comes from an interview published by Bloomberg a few days ago, in which Greg Corrado, a senior research scientist at Google stated, that a “very large fraction” of the millions of queries a second that people type into Google have been interpreted by an artificial intelligence system, nicknamed RankBrain.

It is also stated that RankBrain has now become the third-most important signal contributing to the result of a search query. This is pretty significant and you can understand the ensuing interest from anyone with even a passing interest in search.

As Corrado even humbly states himself…

“I was surprised… I would describe this as having gone better than we would have expected.”

So let’s answer a few questions based on the available evidence out there at the moment…

What is RankBrain?

It’s an artificial intelligence system developed by Google that helps process its search results using machine learning.

Machine learning is basically a form of AI, where computer programs are created that can teach themselves how to develop and change when exposed to different types of data (in this case, search queries).

How does it work?

Here’s where we refer back to Google’s senior research scientist… 

RankBrain embeds vast amounts of written language into mathematical entities into a format that a computer can understand, then if RankBrain sees a word or phrase it isn’t familiar with, using this information it can ‘guess’ what words or phrases might have a similar meaning and filter the results accordingly.

What purpose does it serve?

Basically, it’s used for handling ambiguous or unique questions that have never been submitted to Google before.

According to Google, brand new queries make up to 15% of all searches a day and as Search Engine Land pointed out, Google processes 3bn searches per day, which means that 450m per day are entirely unique in nature.

Clearly there is a need to use machine learning to cope with the sheer demand of its users. RankBrain is also showing signs of improving on Google’s own search engineers. Those behind the search software were asked to look at various pages and predict which pages would be ranked at the top of the Google results. The humans guessed correctly 70% of the time, RankBrain guessed correctly 80% of the time.

Should I be terrified?

Yes absolutely. As I said in my street-pamphlet that you probably threw in the trash immediately after I forced it in your hand, “run for the hills.”


No. I’m sure everything will be fine.


Source link