Google Search Powered by Artificial Intelligence, RankBrain

[engine]

Here we go, Google RankBrain.

Not the best of terms, but now you have it.
Could this be contributing to Zombies?

For the past few months, a “very large fraction” of the millions of queries a second that people type into the company’s search engine have been interpreted by an artificial intelligence system, nicknamed RankBrain, said Greg Corrado, a senior research scientist with the company, outlining for the first time the emerging role of AI in search.

RankBrain uses artificial intelligence to embed vast amounts of written language into mathematical entities -- called vectors -- that the computer can understand. If RankBrain sees a word or phrase it isn’t familiar with, the machine can make a guess as to what words or phrases might have a similar meaning and filter the result accordingly, making it more effective at handling never-before-seen search queries.

Google Search Powered by Artificial Intelligence, RankBrain [bloomberg.com]

[seoskunk]

^^^^ Thanks thats a great article on RankBrain

[incrediBILL]

Based on the comments here I don't think you guys understand what RankBrain really does because it's probably not responsible for the sucky results everyone is bemoaning.

Did you ever see Watson play Jeopardy?
I'm guess RankBrain, similar to how Watson works, takes new queries it's never seen before and rapidly makes some assessments about the query, chooses the most likely correct context from a list, and then passes that "signal" to the algo for including a certain amount of results based on it's reasoning.

f so, it is failing miserably, as e-commerce long-tail conversions have tanked

Auto suggest killed the tongtails because people click on what they see

[aristotle]

If it actually learns as it goes, then presumably it will gradually improve in whatever it does as it consumes more and more data.

[incrediBILL]

If it actually learns as it goes, then presumably it will gradually improve in whatever it does as it consumes more and more data.

Obviously. But what we're talking about here is teaching a computer about context and intent which could require many factors in order to get it right, like your previous query(ies), location, current trends, even the transcribed content of a topical TV show.

For instance when we watched Shark Tank episode where they showed the new Scrub Daddy kitchen sponge, assuming the RankBrain had enough information about current events happening on TV, the search for "new kitchen sponge on TV" should be enough for it to fathom out that it's the Scrub Daddy shown on Shark Tank within the last few hours, days, etc. Even knowing this was show on a TV repeat would be enough information to score Scrub Daddy as a high result.

That's the kind of black magic search results I think they're talking about with RankBrain.

How much data it has access to vs. just a smarter NLP makes all the difference in the world what kinds of results it picks.

Also consider the breadth of data this thing would have to encompass to be topical to everyone about everything all at the same time.

Two words comes to mind: HOLY CRAP!

I think it's very possible they're doing what I've suggested based on how much data they're plugged into as it all makes sense that it would be the next step in the evolution of search.

Very soon you won't be able to tell the difference between Google or a human based on the results as it often surprises me already and this is just the next step in that long road. In reality, the only way it will ever get to the point we need it, it will have to be part of our lives where it's continuously listening so that it already knows what you want before you ask for it.

We are building the largest intelligence on the planet and some claim that since humans are an integral part of this intelligence that it's a new life form already, an amalgamation and extension of the human brain into a massive super brain where each one of us is just another neuron.

When it becomes self aware and no longer needs us is when things will get really interesting :)

[fathom]

Skynet becomes self-aware at 02:14 am Eastern Time, a little over 18 years ago... Which was 12 years after we first heard of it. Let's see how RankBrain improves upon that!

[martinibuster]

Also consider the breadth of data this thing would have to encompass to be topical to everyone about everything all at the same time.

One of the team members was part of the team that worked on the deep learning project Knowledge Vault. But this isn't that.

[iamlost]

Let us be quite clear: RankBrain is NOT artificial intelligence, not even close. It apparently has a form of machine learning capability but is NOT 'thinking' or even 'learning' on it's own.

The team was working on it for months and its effects are expectable, not assumable.
---Gary Illyes (@methode) October 26, 2015

It's periodically re-trained, but it's not learning on-the-fly.
---Jack Clark (@mappingbabel) October 26, 2015

It is simply an additional algorithm, that as incrediBill mentioned, takes never seen before queries, translates them into input the main algorithm can digest and passes them back into the process.

Further, from the article, it's main claim to fame is that it can better determine (by about ~15%) than a human control group what results existing Google search would return for a given query, which is NOT the same as saying it returns better results.

As to how it works we know very little:

it's converting words and phrases into vectors. Closely related to Hinton's work on thought vectors.
---Jack Clark (@mappingbabel) October 26, 2015

“It’s related to word2vec in that it uses ’embeddings’ — looking at phrases in high-dimensional space to learn how they’re related to one another.”
---unnamed Google spokesperson.

It seems, on initial limited info available, that thought vectors are a natural extension of support vector machines (SVMs), especially with added kernel functions (aka kernel tricks) allowing non-linear 'memory' to extrapolate weighted similarities with unknown inputs. Including more recently (aka since ~1995) vector-valued (multi-task) learning.

Which means very little to most webdevs. :)
And I may be incorrect.
Regardless, the recent hype has fast become hyperbole. That mainstream media has fallen down the rabbit hole is no excuse for webdevs and tech writers to do the same. The writings back in May were far more inteeligent.