RankBrain- This Deep Learning algorithmic inclusion to Google’s search has changed the Internet, is currently affecting your search results, and will likely continue to do so. Here are the questions you need to be asking right now….
“What does Deep Learning have to do with me?”
SEO’s need to know these terms – Machine Learning, Deep Learning and Machine Intelligence, due to the addition of RankBrain, Google’s machine learning artificial intelligence (AI). The more you can learn about how Google works, the better you can optimize sites to rank on Google.
(Image Credits: fourdots.com)
“Why is RankBrain so important for us?”
Because out of the 200+ factors used in Google’s search algorithm, RankBrain is the 3rd most important of them all. Since they have never shared what the first and second ranking factors are, this tells us two things:
1. Google thinks RankBrain is worth knowing about and understand why- it must be important to them, and since they make (most of) the rules, we should keep up.
2. To get the best bang for your buck when optimising your website, you need to understand what RankBrain is in order to maximise ROI.
“So, how does it work?”
RankBrain is a self-improving, deep learning system created in part by Geoffrey Hinton (who has a brilliant course currently on Coursera, “Neural Networks for Machine Learning”). He has spoken about creating its neural network. A neural network is a computer system modelled on the human brain and nervous system. RankBrain uses its neural network, and “learns” the strength between connections, gaining a level of understanding about the connection between each neuron – whether strong or weak and its relevance to its search.
This gives a better understanding to RankBrain of which things interlink, and teaches it by creating patterns. New input is added to the database throughout this process, and informs future patterns as RankBrain continues to learn. This is proving to be especially effective for long tail keywords, multi words, complex and more conversational searches. It’s been taught to learn what it means when the search input is less clear, and will get progressively more capable as more information is processed by it. This will make our search results more accurate in the future. So far, pretty useful.
“As computers have gotten faster and bigger, the seemingly impossible a few years ago is now possible”
“The head of AI is now the head of search?”
Maybe you heard that the newly appointed head of search for Google was previously the head of Alphabet’s (Googles sister company) Artificial Intelligence sector. This doesn’t seem coincidental. Since Amit Singh retired as head of search, John Giannandrea has stepped up to lead Google’s searches into the future. With his strong background in AI, he is very much expected to continue RankBrain’s development.
(Image Credits: marketingland.com)
“What has Google said about this technology?”
(Image Credits: neilpatel.com)
Greg Corrado, a senior research scientist at Google, has referred to RankBrain’s way of learning as “gradient descent” or “gradient learning”, which means the system makes tiny adjustments over and over until it gets it right. He states that progressive AI in our everyday lives is becoming more prevalent, whether it be in anti-lock braking, autopilot technology, spam filtering, language translation or recommendations in your searches.
“We are all using AI in our everyday lives, without much thought about its integration.
“Well, how does RankBrain learn?”
Think slow and methodical, mimicking a toddler in style. Much like when a toddler learns, all the input /answers are not always correct – this is known as unsupervised learning. It is not only shown pictures of one subject to teach a certain subject; some answers will be wrong. It could be shown thousands of pictures of something and use AI to understand patterns between similarly linked ones. Engineers do not even understand exactly how RankBrain works, as it connects seemingly incomprehensibly linked images or words to inform its results, prioritising one random looking vector over another, but they know it does, and it does it well.
“RankBrain was right 80% of the time for which searches would rank higher vs 70% for Google’s own search engineers.”
“Is RankBrain used much right now?”
(Image Credits: stonetemple.com)
Oh yes. RankBrain is utilised in “very large fractions” of searches today, helping with the 15% of queries never searched before. The learning RankBrain does now teaches it to further understand complex searches and how they relate to particular topics. Most importantly, it can then associate groups of searches to return results it thinks the searcher will appreciate the most.
Since the beginning of 2015, when Rankbrain was gradually being rolled out in sandbox tests on Google (where it ran upwards of 20,000 different experiments), it’s been fully live and functional for months now. You may have noticed the “did you mean?” function leading you to a more likely search when you have misspelled in the search bar – this is due in part to machine learning understanding that you probably meant to search for ibuprofen, not eyebrooprofen.
“So far, it sounds pretty cool”
I know! This is a brilliant demonstration of AI usefulness, but placing our searches in the hands of a math based learning machine does lead to less human control over the results given. Personally, I can’t wait to see how Rankbrain will develop with Google. Will it only get more impressive as we advance? Or will we have tough questions to ask about the use of AI in everyday life?
I’d love to hear your thoughts, please comment below.