I was recently interviewed by Dan Gilmore on Supply Chain TV and CSCMP. One of the things we talked about that needs further explanation is artificial intelligence, especially because it is definitely one of the big trends in supply chain analytics right now.
It is one of the big trends because everyone is using the term. It feels like just last year, no one in supply chain or operations even used it, but now it feels that every type of analytics is being called artificial intelligence. At the Gartner Supply Chain Conference this year, I heard that you had to work hard to find a session that didn’t mention artificial intelligence in some capacity.
But, what does the term even mean and what should a supply chain or operations manager know about it?
Here are the three things you need to know:
- Many people seem to be using the term “Artificial Intelligence” as an umbrella term to mean any type of analytics solution.
For example, prior to 2015, things like data mining or predictive algorithms like random forests were simply called “analytics.” In 2016, they would have been lumped under the then trendy term “machine learning.” Now, in 2017, they all be part of “artificial intelligence.” My guess is that people think the term Artificial Intelligence just sounds better and more advanced. Plus, if you are a vendor, it has the added benefit of having the audience think your solution does more than it really does.
This leads to the second thing you need to know:
- Artificial intelligence does not mean full automation of a task or process.
In the news and in magazines, artificial intelligence is usually associated with self-driving cars or machines that can outplay the best players of Chess, Go, or Poker. Naturally, when a supply chain leader hears about artificial intelligence applied to inventory management, they picture an algorithm taking care of watching inventory levels, issuing PO’s, and automatically expediting with the same capability (or better) than today’s human planners. But it should be noted that AI is a long way from using more sophisticated algorithms to having a fully automated system.
- Artificial intelligence often means, specifically, deep neural networks.
The people on the cutting edge (academics, researchers, the data scientists building systems to win at Poker, etc) use the term artificial intelligence to refer to deep neural networks. A deep neural network is a type of algorithm that got its name from its similarities to how the brain works. The “deep” part refers to the fact that the algorithm can have many different steps. Breakthroughs in hardware (specifically GPUs) have allowed researchers to actually get deep neural networks to run.
Around 2012, there were several big scientific breakthroughs with deep neural networks— they showed a dramatic increase in the ability to recognize images (as well as people could) and translate language. This gave the research community a renewed energy to apply these techniques to other problems—like what you’ve read about with playing Go or Poker.
These are definitely exciting breakthroughs. And, deep neural networks show promise to transform many different aspects of business.
However, to apply these neural networks to your business, you need a lot of data (millions of records with 100s or 1000s of existing or derived attributes), a lot of hardware, and you need to tailor the neural network to your business (the neural network for playing Go is very different from the one for Poker).
Besides, you should keep in mind that a deep neural network is still just an algorithm (indeed a sophisticated algorithm), for making a prediction. There is a lot of work to go from the prediction to fully automating the decision.
This third point isn’t meant to scare you away from this area, but to give you a starting point to better understand it and research it on your own.
As you come across people and vendors talking about artificial intelligence, hopefully these definitions will help you better figure out what they are talking about.