Another Penguin question was raised in Friday’s webmaster hangout with John Mueller, this time specifically on how machine learning impacts Penguin, but the answer turned into something far more interesting about Google’s use of machine learning in the search algo.

The question was “Are you using more machine learning algorithms into Penguin, or are you relying more on machine learning when it comes to Penguin?”

While Mueller doesn’t specifically talk about the role of machine learning in Penguin, he did have some interesting comments about how Google is using machine learning in the algo, and why they need to be careful about using or relying too much on machine learning.

I don’t know specifically about Penguin.  But we do do a lot with machine learning.  I think that’s a really fascinating area.  It’s something where, I mean, we’ve worked with machine learning for quite some time now.  Sometimes it’s interesting that in the sense that these machine learning algorithms learn something that maybe we wouldn’t have come up with intuitively, that we try to figure out how did the algorithms come up with this result.

Sometimes it makes it hard for us to debug and diagnose what is actually happening there.  So if our whole search results were built with one machine learning algorithm that learns the optimal ranking for every keyword for every website for every user, then if something were to go wrong  there, that someone would go that showed me something that that was completely wrong, then diagnosing that would be really hard.

So it’s something where you kind of have to find a balance between the different types of algorithms and make sure that you can reproduce what’s actually happening there, what signals we are picking up and why we turn that into a specific ranking, and at the same time, use these advance systems to kind of bring in new ideas and see if there are ways to kind of improve our existing algorithms in ways we didn’t think about before.

Machine learning is definitely a fascinating topic, especially since we learned about the use of machine learning in Google’s search algo with RankBrain , particularly because it not only is the third most important signal in the algo, but that it has a larger impact on those never-before-seen queries.

But RankBrain, because of the very nature of it, is something that would be much harder for an SEO to reverse engineer, at least not to the extent that SEOs are able to reverse engineer some signals in the algo.  Throw into the mix the fact RankBrain is also baked in – meaning not something separate like we see with Panda and Penguin – and machine learning puts a completely new spin on algo chasing by SEOs.  And this is why so many are fascinated with learning more about it.

It does make sense with Mueller’s explanation about why Google doesn’t use one machine learning algo to rank all the search results, although it would certainly be interesting to take a peek at what the search results would look like if they did.  But debugging would be much harder than with a non-ML algo.

Here is the video: