Despite the feeling of being rational and in control of our decision-making, psychology is full of examples of ways that our decisions can be influenced without us even being aware of the influence. Indeed, Richard Thaler was just awarded the Nobel Prize in economics for his work on nudge theory, the idea that understanding how people think is essential in getting them to do what you want.
For example, changing from an opt-in to an opt-out pension policy in the UK lead to a 285% increase in participation in 4 years. We're more apt to buy a new toy for $4.99 than to pay $5.00 for the same toy, much more so than can be explained by the one-cent discount. And that bottle of shampoo with "50% more for free" is very attractive, especially if it's also "new and improved!"
While increasing retirement savings is a great goal, I've always assumed that the extra 50% in the shampoo bottle was not actually free, and that whatever was new and improved about it had absolutely no impact on the core function - getting my hair clean.
I worry that Artificial Intelligence and Machine Learning are becoming the new "new and improved!" It's a rare company in the tech space that doesn't tout the use of AI or ML in their product. But saying that you use AI or ML doesn't mean that your product performs its core function any better.
Here's what I do to try to separate hype from substance. Marketo recently announced Predictive Content, and has a web page with the headline "Predictive Content: Engineer Higher Conversions with Machine Learning". The first thing to ask is "why machine learning?" Is the problem a good fit for learning, or is learning simply a buzzword? In this case, personalization of marketing content to customers is a laborious process that's hard to scale. So, yes, learning to do that personalization automatically seems like a good idea.
Is there enough data? Again, for organizations with long-running Marketo installations there could be enough data to learn about user engagement patterns. And, finally, are there any details at all about how they actually do it. In this case, the answer is again "yes". There is material online that even refers to specific learning problems and algorithms and how the solutions to those problems can be used to surface content to customers. That doesn't tell me anything about how well it works, but it does tell me that the use of "machine learning" here is not just about the latest buzzword.
I won't go through an example of what seems like a vacuous use of "machine learning", but you get the idea. If you see no details on how it works, don't understand where the data might come from, or don't see an easily understood justification for the use of ML, there's a very good chance that the 50% more machine learning in the box is either not free, or won't help get your hair any cleaner.
If you'd like to read more on Machine Learning in business, try "3 Reasons To Start Implementing Machine Learning Before Your Competition Does".