Teaching a computer to recognize Miley Cyrus

Getty Images

Miley Cyrus

The ultimate goal of big data is to reach and perfect deep learning—when computer models analyzing massive amounts of data are so accurate they can recognize the features of various objects—cat, dog, human, even celebrity.

A famous Google example is when it trained a system to recognize faces of humans and cats from watching YouTube. The power is in the prediction, and in theory, the future is feeding data points into the computer that is able to quickly identify faces such as Miley Cyrus, even without really knowing who she is, or what a celebrity is.

"Google built a model to decipher these things, unsupervised. No one told it 'this is a human face or a cat,'" said Derrick Harris, senior writer with GigaOM on CNBC's "Squawk Box." The computer is able to identity a cat or human just by having seen enough of the objects, and based on analysis of what's called multidimensional data.

(Read more: Retailers go high tech for high return)

"My daughter looks at enough pictures of castles in Disney moves, she starts to think casinos in Vegas where we live are castles. …She's got the features right," Harris said.

Even though someone would still have to train the computer to say, "That's Miley Cyrus," the everyday living possibilities linked to deep learning include some pretty cool stuff.

Take Google's Android OS. Deep learning will attack speech recognition and learning applications, and images and text. So, you would be able to pull up photos of your trip to Europe by having the system learn what images are of particular locations in a certain region.

And your love-hate relationship with the auto correct feature on your smartphone might lean more toward love. Right now, the match works by relatively simple word-by-word prediction on mobile devices. The deeper computer learning becomes, the more it will be able to predict and base recognition on whole sentences and formality, learning speech patterns.

(Read more: Can Twitter make bad TV good?)

Any text source—tweets, LinkedIn profiles, job listings, customer service chats, transcripts—can be used to predict some sort of outcome, such as education, salary, sentiment, and customer churn.

Deep learning is that powerful because it's pretty much tailor-made for "big data." The models aren't actually that complex—and this is critical—but the more data you feed them, the more accurate they get.

The basic concepts behind deep learning are already being co-opted for corporate strategies.

Error detection in engines: Ford famously used neural nets to detect misfires in Aston Martin models several years ago, but deep learning could make those models more accurate.

Genomic data: Predicting disease is a logical end for deep learning applications.

Pharmaceuticals research: Merck ran a Kaggle competition on predicting the activity of molecules, which a deep learning approach won.

Surveillance programs: The NSA, as an example, can use deep learning to predict patterns from real-time satellite images for potential terrorist activity.

Social media: Across many industries, companies are trying to predict outcomes based on Twitter activity, and those efforts will receive a boost from deep learning.

Getty Images

(Read more: The Kit-Kat surveillance state)

Google is leading the charge with its release last week of the open source Deep Learning program Word2vec.

There is a lot of research around deep learning and some startups are getting into the game, but Harris said what makes Google so interesting is the smartphone and tablet market to which it can plug in deep learning solutions.

"It is very computational intense and what Google did last week is release a tool that is open to all researchers and lets people play around with it," Harris said. "They democratized it. They did the leg work on their end and sent it out to all."

Google has used the same open source model in the past, under the theory that the more people working on it, the faster things evolve.

"Keeping the project open source allows massive improvements, and drives innovation," Harris said. "It's Google's way of learning new applications. There is a real value in open sourcing it and it's not going to put them at a competitive disadvantage."

—By Eric Rosenbaum, CNBC.com