I’ve read more than my fair share of articles about all-things AI, including ones on machine learning and deep learning, and two things are clear: (1) my brain is about to short-circuit faster than a vintage robot, and (2) more often than not, the media uses these terms interchangeably even though they’re not the same thing.
So, what’s the difference?
First of all: know that machine learning and deep learning are related and that both fall under the AI umbrella. Think of the three as concentric circles: deep learning is a type of machine learning, machine learning is part of AI, and AI encompasses the entire field of study.
What is machine learning?
In a nutshell, machine learning is an automated process that uses data and math (algorithms) to uncover (“learn”) new information without human intervention.
Because the machine continuously “self-learns” (or “self-trains”), humans don’t need to write code for each process along the way (huge time saver!). Self-learning also means the computer can identify new methods on its own, so it may find solutions that we mere humans wouldn’t have considered.
How machine learning works (at a very, very high level):
For example, let’s say our favorite cat lady got the help she needed and is ready to meet the man of her dreams on “Catch,” an app that uses machine learning to make [human] love connections. Catch would help our kitty fanatic find likely love interests (desired objective) by understanding what she likes/dislikes about each candidate (e.g. cat fur craft enthusiasts), and continuously refine criteria to present more fitting stud muffins for her consideration.
Marketers use machine learning in a similar way. For example, Magnetic’s AI platform uses machine learning to sift through real-time data of about 260 million individuals, and identify those who are most likely to click on an advertiser’s ad, view a video, or complete a conversion action. In many ways, identifying high-value prospects for advertisers is like finding crazy cat lady’s ultimate love connection.
What is deep learning?
Deep learning is a subfield – and some say the “cutting edge” – of machine learning. Deep learning takes machine learning to the next level by better emulating human brain functions.
What does it mean to emulate the human brain? Does that mean deep learning also forgets to do laundry and randomly busts out Beyonce song lyrics at inappropriate times?
It means this: your brain is a category-creating, pattern-identifying system. Both the brain and deep learning software use patterns and categories to compute and “think” in a highly effective way, better than machine learning alone.
One of the coolest examples I have seen of deep learning at work is automatic colorization.
Deep learning algorithms can identify objects and their context within a black and white photograph to determine the colors that are most likely associated with every object. Pretty cool, huh? Try it for yourself here.
So, how does deep learning apply to marketing? One [of many] ways is that deep learning provides more reliable and rich descriptions of customer buying behavior. For example, it can find and use contextual data to identify specific events that may be high-value marketing opportunities – like an impromptu shopping spree or sudden interest in a product or service. This allows marketers to respond in real-time to capitalize on the opportunity.
How are they different?
Machine learning and deep learning are parts of the same whole (per our diagram above), but they have their differences.
Machine learning allows computers to learn on their own. Deep learning does too, but it goes, well, “deeper”. Where machine learning can identify consumers with a high propensity to convert, deep learning can identify erratic and chaotic shoppers who may not seem to be interested in an event or product – but may actually have the most potential to convert.
For example, let’s say AI is being used to identify people’s emotions in photographs. Machine learning would find pictures with faces and put them into the system. Then deep learning would recognize facial patterns and identify the specific emotions.
One important difference between the two is rooted in data breadth and depth. Deep learning works best with large, rich data sets; smaller, more narrow data sets may not give deep learning algorithms the info they need to understand details and context that is necessary to perform well.
How will this affect marketers?
Marketers are notoriously focused on short-term outcomes. We put little-to-no effort into strategies and tactics that reap long-term rewards, like building brand affinity and growing customer lifetime value.
Much of our aversion is rooted in risk. Marketers can’t say – with confidence – “we can invest in this strategy, do it correctly, and get a huge return in X years.”
Deep learning will change this. It will tell us much more about multi-touch success. Marketers will be able to account for every consumer event, action, touchpoint, consideration, brand interaction, (etc.) when modeling short- and long-term strategies and associated return.
For example, with deep learning, a brand could reduce or mitigate harm to customer relationships. Let’s say crazy cat lady has a poor experience with her Catch dating app. She calls the company, writes an email, chats with them online, but still gets no resolve. Deep learning algorithms could interpret all of these communications, relate the events, and indicate that crazy cat lady is a progressively irate customer. In turn, it could automate a response that bubbles the issue up the chain so crazy cat lady’s concerns are addressed immediately, and she remains a Catch customer (and maybe even tells her crazy cat friends about how great Catch is).
Needless to say, the possibilities are somewhat endless. But marketers can be sure that deep learning will have its time in the limelight in the relatively near future.
Magnetic is a digital marketing and artificial intelligence company. We use machine learning and AI to deliver smarter, faster, and more effective advertising. Our powerful AI platform continuously analyzes the attributes of 320 million live user profiles alongside real-time inventory supply and bid opportunities to deliver highly performant and profitable campaigns for our clients.