Google Brain Turns 3

Even in our office, which is populated by young, tech-savvy, social-network-obsessed, college-educated people, few things stir up fear like images of Boston Dynamics’ Atlas humanoid robot.

The world’s most anatomically and dynamically correct android, it’s the way the Atlas moves that makes us the most afraid — because it reminds us, of us.

A humanoid machine made of metal with grimacing edges, wires, and metallic joints, we’ve seen this kind of thing too much in the realm of sci-fi to not immediately associate it with the inevitable and ultimate struggle between man and technology.

In 2013, Boston Dynamics, the company that gave us other biologically inspired machines like the “Big Dog” and the “Cheetah,” was acquired by Google’s private, highly secretive experimental technology branch, Google X.

But as much as the appearance of machines like Atlas and its cousins may scare even the least tech-phobic among us, it’s something far less physically imposing that should be truly striking terror into our hearts.

What’s 3 Years Old and Loves Kittens?

When the technology I’m about to tell you about made news more than three years ago, the main point of interest wasn’t its determined, aggressive gait or its bone-crushing fingers and hands.

The most impressive thing it could do was something just about any toddler past the age of 18 months could do equally well: It could tell when it was looking at a picture of a kitten.

Also a property of Google X, the Google Brain was described by the New York Times as “16,000 computers dedicated to mimicking some aspects of human brain activity.”

Not a very impressive task when compared to things like calculating orbital speeds and trajectories required to send a capsule to Pluto, for example, but it’s the way this “brain” processed information that was most revolutionary.

Instead of following strictly set guidelines to achieve the task, this artificial thinking machine relied on something known to artificial intelligence researchers as “deep learning.”

It’s defined as “a branch of machine learning based on a set of algorithms that attempt to model high-level abstractions in data by using model architectures, with complex structures or otherwise, composed of multiple non-linear transformations.” Deep learning, or DL, involves using basic commands to fluidly create associative links in an organic, free-form manner.

It’s not easy to explain or understand on a technical level, but essentially, it’s how all of us learn to learn.

After being exposed to more than 10 million images of cats, the Google Brain was then able to recognize cats in images regardless of camera angle, positioning of the cat, background, or other features in the image.

It formed this understanding the same way our own brains form neural pathways — through sheer volume of exposure to data.

Technology Mimicking Nature

Pediatric neuroscience has estimated that by the age of two, the average human has seen the equivalent of trillions of individual images coming in a constant stream of data every waking second of every day.

Only a small percentage of those images can make an impression strong enough for us to form associations that lead us to complex cognitive functions like recognizing an object for its specific shape and nature, recognizing what an object is called, what it feels like, and what role it plays in day-to-day life.

But the enormous volume of data, combined with the human mind’s ceaseless appetite for it, allows us to pick things up fairly quickly and move on to more complex learning.

Google’s Brain, with its cat-recognition capabilities three years ago, was able to mimic a level of cognizance equivalent to that of a child between the ages of 15 and 18 months — about the same point at which they start to recognize their own reflection.

Little kids and kittens… Sounds cute, I know.

Only it’s not.

Because while kids at this age are at the peak of their learning and mental adaptive abilities, the Google Brain is a machine and works on a completely different curve.

Moore’s Law, which dictates the speed at which computer-processing power increases over time, looks like anything but the graph mapping out human learning potential as age progresses from childhood to adulthood.

The Machine Learning Curve

Instead of gradually flattening out (the rule of stating you can’t teach an old dog new tricks), Moore’s Law states that processor speed doubles once every 18 months or so.

It’s proven extremely reliable over the four decades since it was first proposed, and it continues doing so today.

Apply that curve to the technology behind Google’s neuralnet, and the result is truly terrifying.

At 18 months, a child and the Google Brain exhibited roughly the same cognitive ability.

18 months later, that three-year-old’s brain is still learning at a dizzying speed, forming more new neural pathways than it ever will again.

The Google Brain, in the meantime, will be twice as smart.

18 months after that, this 4.5-year-old child is now speaking full sentences, counting, and reading.

The Google Brain, now also 4.5 years old, has doubled its learning potential yet again.

By age six, our hypothetical child is ready to go to first grade, learn arithmetic, read and hear more complex stories, and gain a deeper understanding of natural social hierarchy.

At age six, Google’s kitten-loving artificial brain is learning at 16 times its initial rate — making associations on top of associations, only accelerating its level of advancement.

By age 17, Google Brain will be 1,000 times as fast and efficient as it was at first.

By the time it would be old enough to legally drink — a point in life when its human classmates’ neural pathways are starting to become rigid and learning potential starts to plateau (not to mention is killed off by alcohol) — Google Brain will have 8,000 times the learning power of its toddler self.

And so on, and so on, and so on.

From Deep Learning to Chaotic Thinking

Predicting the speed and efficiency of its processors is one thing, but when it comes to deep learning and the multi-leveled conclusions that this brain will be making years before it gets this powerful, forecasting the results is nearly impossible.

There is simply too much chaos stemming from too many factors, too many options, and too many potential outcomes for any viable model to be created.

By the time Google Brain is in its late 20s and is more than 100,000 as “smart” as it had been in 2012, the only way to tell what it will think and how it will react to stimuli will be to actually ask it.

Of course, by that time, it would have far exceeded the collective cognitive potential of every Google employee put together, so chances are our interactions with this level of technology will be far more complex than simple question and answer.

Now, here’s where things get really mysterious…

Where Are They Now?

Since we first heard of the Google Brain in the summer of 2012, there has been very little public talk about it.

What we do know is that on January 26, 2014, Google acquired DeepMind Technologies.

DeepMind was at that point a four-year-old start-up and already a global leader in deep learning research and development.

The company was successful in creating a neuralnet capable of teaching itself to play video games and accessing short-term memory in the same way a human mind does — which is part of the reason Google is rumored to have paid more than three-quarters of a billion dollars for this 70-employee firm.

The actual sum remains a closely guarded secret.

And yet news on the Google Brain is sparse, to say the least. It’s gone virtually unmentioned since the initial announcement, despite years of accelerating development.

In case you’re keeping track, however, that brain should now be at least four times smarter than it was when the New York Times first wrote about it.

What effect the addition of DeepMind Technologies has had on this project and others like it is something only the people at Google X know for sure.

One thing is clear, however: When it comes to ushering in the age of machines, it will be artificial intellect, not artificial anthropomorphic bodies, that are most transformative.

I would keep your eyes on this technology. Chances are, sooner or later, it will start keeping its eyes on you.

It’s only a matter of time.

Fortune favors the bold,

Alex Koyfman

Check us out on YouTube!

His flagship service, Microcap Insider, provides market-beating insights into some of the fastest moving, highest profit-potential companies available for public trading on the U.S. and Canadian exchanges. With more than 5 years of track record to back it up, Microcap Insider is the choice for the growth-minded investor. Alex contributes his thoughts and insights regularly to Wealth Daily. To learn more about Alex, click here.