In case you haven’t noticed, Artificial Intelligence is the It Girl for marketers these days. Just about every ad tech tool now claims to be powered by AI components. Agencies and consultants are touting their AI capabilities. CMOs are bragging on their AI initiatives. As a result, the definition of AI has become stretched beyond recognition. Many people are defining it by what it does (language processing, data manipulation, visual recognition, etc.). But the real mark of artificial intelligence is how it does it. A chess-playing computer could be AI or not. If it plays chess by analyzing every possible move on the board, assigning it probability of success, and selecting the move with the highest probability, then it isn’t Artificial Intelligence. If it plays chess by applying certain principles, and refining those principles through feedback gained from multiple games, then it is AI.
While the more technically-minded would have more specific and accurate definitions, marketers can practically characterize Artificial Intelligence applications by two things:
- The ability to perform tasks for which they’re not explicitly programmed
- The self-directed capacity to improve that ability as it is exposed to more tasks.
At the core of the computing age is the power of If-Then processing: If a certain condition is true, then perform a defined operation. For example, you might program a traditional application by coding “if Speed is greater than 55mph, Road = Expressway and Weather = Rain, then reset Speed to 45mph.” This is an explicit direction. Using AI, you could effectively program an application more along the lines of “If driving conditions become less safe, then slow down to meet the conditions.” Using the definition outlined above, AI would use a variety of input (weather, traffic, type of road) to determine what is “less safe” and the speed appropriate to the conditions. Based on feedback(e.g. how many accidents occurred), it would improve its definition of “less safe” over time without a developer having to go in and explicitly recode every conceivable definition of “less safe”. So while the first explicit program will never change its behavior, an AI program would learn to distinguish between rain falling at 70F and rain falling at 32F. In essence, AI allows applications to refine their If/Then engine autonomously.
There are many things being touted as AI that really aren’t. For example, increased processing power is allowing for even more extensive and sophisticated searches of ever larger databases. So, if you ask a Dinner Bot for recipes using tofu, asparagus and mayonnaise that’s quick to make, it can search multiple recipe databases, correlate those ingredients with preparation times under 20 minutes and send you back the 10 most popular recipes fitting that description. It can do all that using explicit instructions. While it may be impressive, it is not AI. Here’s one test to apply to see if something is really powered by AI. Will the results of the applications change over time depending on who is using it? The answer should be “yes.” In other words, two identical AI applications should start to yield different results if they are used by two distinct groups of users. That’s because they’ll be exposed to different situations and different feedback that should affect its autonomous development.
For marketers looking how to start thinking about applying AI to their businesses, Shelly Palmer provides a typically insightful view of the topic.
The Ryan Report 