I never made a formal tally, but rarely would a class go by without the mention of “AI,” “Facebook,” “Google,” “deep learning,” and “algorithms.” That’s hardly a surprise for a student in data science, right? Only this wasn’t a data science course, nor was it a computer science course. This was my marketing professor, and he gushed over artificial intelligence (AI).
To him, the objective of marketing was to connect an audience with a certain need to a seller who has the solution. Nothing does this better than AI. Yet, our industry suffers from a deep lack of understanding.
Why? There’s simply a talent shortage of people who can build and understand these tools, as top technology firms can’t meet their own demands. Marketers love AI. They know that this is the direction the industry’s been moving toward, but few understand the technical side of it.
How many marketers could tell you the difference between supervised and unsupervised learning? How many know what a neural network is and how it’s structured? How many of them are getting prepared for Facebook to build Skynet?
It’s important for every marketer to understand what AI is. It’s the most general term given to any machine that responds to input in a way typical of humans.
AI is a moving benchmark, if just colloquially. Yesterday’s AI is today’s responsive design. Automatic doors, motion-detecting lights, and Roombas were all simple machines that did tasks normally required of humans, but nobody would call them AI as it’s no longer the norm to have doormen or spotlight operators.
However, this isn’t what we think about when we hear people talk about AI. We think of Google finding the answer to our every question, Netflix recommending new shows to binge-watch based on our previous ratings… and who could forget all the Terminators? Of course, this is AI, too! It’s just part of a subset of AI known as machine learning.
Although most people may not be aware, machine learning has actually been around for decades. But why is everybody suddenly talking about AI now? It’s because of a fairly recent breakthrough subset of machine learning known as deep learning.
What does this mean for marketers? It means that we’re currently at the crossroads of some incredible technologies and possibly the cusp of a fourth industrial revolution. The availability of low-cost computing power and data storage has allowed for the accumulation of more data than previously possible. This data has enabled the development and training of some of the most complex and powerful deep learning models in the world. These models are the vessels through which we serve online ads.
The best way to understand the history of AI research and development is through the games we taught them to play. By jumping into three different programs, we can learn what separates AI, machine learning, and deep learning from each other. Plus, this will provide some context when you find yourself discussing AI at your next team meeting.
In 1997, IBM’s Deep Blue computer system defeated reigning world champion Gregory Kasparov in chess. This feat stunned the world. Chess is a game that requires deep thought, careful planning, and patience. With two individuals playing against each other, the game itself can be unpredictable with so many turn options to choose from. How could chess be played by a computer? This was so hard to believe at first that people accused IBM of cheating.
Deep Blue won through sheer brute force. It could compute over 200 million positions every second, sometimes calculating as many as 20 turns out. It would score all possible boards and use a decision tree to determine what the next optimal move would be. It’s what’s known as “Good Old-Fashioned Artificial Intelligence” (GOFAI) or symbolic AI. It’s so basic in its design that one of its creators denied it being AI at all!
This was a key event for AI in the public mind. It disrupted the notion that certain tasks could never be meaningfully replaced by computers. It was dubious as to whether a computer could ever beat the best human chess players, but when it happened it opened up possibilities to what could be replaced or improved upon by computers.
In 2011, IBM’s Watson defeated legendary contestants Ken Jennings and Brad Rutter in the quiz game Jeopardy. Again, this stunned the viewing audience at home. I can recall watching as Watson won handily, albeit making egregious mistakes along the way. I vividly remember Ken Jennings writing “I for one welcome our new computer overlords” on his Final Jeopardy response.
If you were watching Watson as it played that day, you would’ve been treated to a window into its “thought” process. Watson is a question-answering program. It worked by reading the clue for salient phrases or keywords and checking them against its 15-terabyte database, at the rate of one million books a second, for possible associations. Then it would take an educated guess, calculating a confidence probability as to what it believed was the best answer.
At its core, Watson was a display of the incredible abilities of machine learning. Machine learning is a subset of AI where algorithms are trained by a set of data to discover patterns and make predictions. Watson is trained by being provided question-and-answer pairs, whereby it can consult its database to map how to best reach a given answer from its corresponding question. Then it uses the model it built from the known questions and answers to guess at unsolved questions. Watson comprised of hundreds of such models to go about breaking down questions, checking its database of knowledge and scoring an answer.
In 2017, AlphaGo defeated the top-ranked Go player at the time, Ke Jie, in a three-game match. While this got the least press attention, it was perhaps the most astonishing feat yet. Nobody thought it was possible, as no current methods could handle the sheer number of possibilities to be accounted for in a game of Go. Up until this point, the best algorithms played at an amateur level and couldn’t beat professionals without being given a significant handicap. So what closed the gap? The use of deep neural networks, also known as deep learning.
Deep learning is a subset of machine learning centered on the use of structures called artificial neural networks. An artificial neural network is a loose representation of how neurons work in the brain. (This topic deserves its own blog post, but let me try to simplify it as best as can possibly be done in a couple sentences.) Your brain takes inputs from your senses and converts these into electrical signals that bounce around an interconnected web of neurons in your brain to ultimately form a response. Similarly, an artificial neural network takes a series of given data inputs and passes them through hidden layers of interconnected nodes, which function as neurons, to determine a given output. The word “deep” in deep learning actually refers to the amount of layers in these neural networks.
AI has significantly picked up in popularity because deep learning tools like AlphaGo have been shown to pull off the unimaginable. Many of the recent breakthroughs in AI and consumer electronics, like voice assistants and self-driving cars, rely on deep learning algorithms.
Hopefully this has given you a concept of what AI is, what it can do, and the levels of AI innovation. Understanding these nuances and what they mean is important for positioning your business for the future and finding solutions to the problems you experience. It’s especially critical that we as marketers understand these technologies, as some of us leverage them every day. We must also realize that our industry is one of the next in line to get swept up in the coming change. Best learn about it now so you’re ready and able to adapt.
Still, nothing would be complete without leaving you wanting more! We’re going to be diving into the synergy between AI and marketing in upcoming blogs, where we’ll discuss neural networks, the different available structures, and their applications. Also be on the lookout as we develop an AI tool to play in our office fantasy football pool.