The clearest way to describe the relationship between data, artificial intelligence and automation are to break them down this way: Data drives artificial intelligence. Artificial intelligence drives machine learning. Artificial intelligence drives deep learning. Although machine learning and deep learning are built on the same foundations of data collection and artificial intelligence techniques, there is still a degree of difference that must be considered before using the terms interchangeably.
Machine learning pipelines are constructed by channeling data into a training model and then deploying the model to complete a certain task. When new data emerges, the machine must be retrained. Data scientists have noted that machine learning tends to plateau. There are limits to its learning. For instance, once a streaming video service learns that you like horror movies you will continuously receive horror movie recommendations, even when Halloween is long past or after you have moved on to documentaries or musicals.
Unlike machine learning, deep learning relies on greater amounts of data. From this data, deep learning technologies are capable of learning from their own mistakes. No retraining is necessary. The mechanism is self-taught.
In a recent piece in The New Yorker, Pulitzer-prize winning author and physician Siddhartha Mukherjee investigated deep learning’s potential to improve diagnostics. In the piece, he describes deep learning as a neural network system (as is machine learning) but with layers. Information passes through each layer and as it rises the connection between the data and the process strengthens, much like recurring thoughts or processes strengthen neural connections in the human brain.
Mukherjee explains the difference between machine learning technologies and deep learning technologies like this:
“Such programs [recognition software]have no built-in mechanism to learn: a machine that has seen three thousand X-rays is no wiser than one that has seen just four.”
In short, deep learning relies on greater amounts of data and has the ability to learn, whereas machine learning relies on rule-based algorithms and can be successful even when data sets are small. Both, however, are only two means of harnessing the power of data and artificial intelligence.