Is Google's RankBrain About To Get A New Machine Learning Cousin?

Google, never one to shy away from technological advancements, released an impressive machine learning algorithm to the world back in 2015. Known as RankBrain, the algorithm functions alongside the Hummingbird update, helping Google understand the semantic meaning behind complex user queries.

Technologists and search engine optimizers were excited about the release, wondering how it would change over the course of the next several years, and speculating what was next for the search engine giant.

Now, it’s been nearly two years since RankBrain was released and we haven’t seen news of any major update or any new machine learning algorithm to work in conjunction with it on the horizon—are we overdue for seeing a new machine learning RankBrain cousin?

How RankBrain Works

First, let’s dig into how RankBrain actually works, as its functionality is significant to its potential future. It was released stealthily in mid-2015, months before it was officially announced, to work with the Hummingbird algorithm, which recognizes the meaning and intent behind user queries rather than merely matching keywords to keywords on-page.

RankBrain’s intention is to help Google understand more complex user queries by reducing them down to more decipherable chunks, which is becoming increasingly important due to the rise in voice search queries (which are often long and conversational). The machine learning component emerges because RankBrain can learn to make connections between words and phrases that aren’t immediately logically connected.

For example, a query like “where’s the best place for a slice?” doesn’t mention pizza specifically, but RankBrain would be able to understand, or eventually learn, that “slice” in this context probably refers to a slice of pizza, and that this user is looking for the best pizza shops in the surrounding area.

RankBrain has been reported to come into play for about 15 percent of user queries—presumably those that have long or complicated wording, or those that don’t specifically mention the exact topic the user is truly searching for.

Google and Machine Learning

RankBrain isn’t Google’s first or most impressive foray into machine learning and artificial intelligence. That honor goes to Google’s DeepMind, which is a sophisticated neural network that has achieved an increasingly surprising range of accomplishments, from mastering old Atari games with no human instruction to besting an international Go champion in a game once thought to be impossible for machines to play.

Neural networks and machine learning continue to be a focal point for Google engineers in multiple applications—including object recognition and device-based AI. But while these algorithms are impressive, none of them seem focused on improving Google’s core product—its search engine.

Object Recognition in Video

Back in March, Google announced that it was working on a video search algorithm that could automatically scan a video for recognizable images, making them searchable based on their actual content, rather than on titles or meta descriptions (as videos are now). For example, if a tiger shows up in the video, that video could feasibly appear for a search for “tiger.”

This new machine learning algorithm seems like a logical step forward for Google web search, especially since Google has been stepping up its favoritism for video content. However, this search function will only be available for videos and users of Google’s cloud storage service.

TensorFlow and Other Projects

Google has also introduced machine learning and AI software to its Android devices in the form of TensorFlow, an open-source software library custom made to make it easier for developers to produce machine learning programs. With TensorFlow, Google is putting AI in the hands of more developers and users, but still isn’t applying new deep learning algorithms in its own search function.

The State of Google Search Updates

It’s worth noting that Google search hasn’t fundamentally changed in the past several years. Game-changing updates like Panda and Penguin helped the search engine iron out some of its original weaknesses, but now the search engine appears to have reached a stasis point.

Good content and high-quality links are still the most important factors to get high organic search rankings, and updates that once saw annual revisions are now incorporated into Google’s “core” algorithm, with constant, ongoing refreshes. New tweaks appear to be made in small chunks on a constant basis, rather than being announced with a giant push. This means any new machine learning processes may be updated more stealthily, or not at all.

The Certainties

It’s hard to say whether Google will return its attention to improving its search engine with machine learning, but there are a few things we know for sure:

• Google will keep pushing for better AI. No matter what, Google is going to keep pushing the boundaries of what is possible with AI. Those machine learning ventures just may not be related to search.
• RankBrain will continue improving. RankBrain is still working well, and is improving on a constant basis. It hypothetically requires no further updating, since it learns and updates itself.
• We won't know when the next update hits. Google doesn’t often publicize its search engine updates, so if a new one is in the works, we probably won’t know about it until after it’s released.

If I had to make a guess, I’d say Google isn’t done with its core search algorithm, but it may be a number of years before a new machine learning component reveals itself in Google’s core. Until then, we have plenty of other impressive AI systems to marvel at.