Artificial intelligence (AI) and Machine Learning (ML) are quite the buzzwords currently for techies and opinion leaders tuned to these technologies of the future. In certain references, the two technologies are even used interchangeably. But how accurate is that? When should you be using the term “artificial intelligence” and when should you be talking about “machine learning”?
In this article, we will try and dissect what these two terms actually mean and try to draw a clear distinction between the two that might help people, particularly writers, journalists, bloggers, and opinion leaders, to tell them apart when discussing these technologies.
The best way to draw a clear line between the two is to treat artificial intelligence as a general term used in describing machines capable of intelligent, reactive and proactive behavior. Machine learning, on the other hand, is more of a specific subset of artificial intelligence. It is an approach or paradigm to artificial intelligence and is a broad field in itself.
Limited Pattern Recognition
Machine learning can be described as one of the (most practical) approaches to general artificial intelligence. There are several approaches to artificial intelligence such as artificial neural networks (ANN) although some data scientists argue that machine learning is the only practical approach. Machine learning is premised on the concept that machines must rely on data models or data points to make seemingly “intelligent” decisions or predictions (machines making analytical data models from data points).
Machine learning is very much like predictive modeling and some AI purists may not regard it as a complete or strong AI. But the machine learning approach to artificial intelligence in itself has gotten so complex that it has become a (reverse) “black box” in its decision making in that scientists are unable to predict certain machine learning responses beyond a certain level of complexity. In this way, they have started to express some little “autonomy” though it is still what data scientists and AI researchers consider a narrow AI implementation.
“Machine Learning Enables a System to Achieve Artificial Intelligence”
ML is premised on the fact that it must follow pattern recognition based on the processing of thousands or possibly tens of millions of data points or patterns. The implication is that if a machine can continuously process so input data with some “regularity”, then it can develop pattern recognition which is a form of intelligence and use this to forecast more accurate outcomes. But this is not the “classical” artificial intelligence as we envision it. That is why machine learning is referred to as a “narrow artificial intelligence”.
Machine Learning should therefore be treated as a single field of technology within a broad AI or one of the possible implementations of artificial intelligence.
Then why is machine learning often treated as “the” artificial intelligence?
Machine learning has achieved synonymy with the general field of artificial intelligence because it has been the most successful, the most promising and the most practical path to developing strong AI. This is so much so that over 90% of artificial intelligence products are now based on machine learning.
The confusion is further amplified by the fact that most companies that develop machine learning products market these products as general AI products. Which they are! But given the mind-boggling possibilities that we envision in an artificial general intelligence, machine learning can be regarded as only a basic or rudimentary implementation.
So for general information, marketing purposes or practical purposes, people who use terms “machine learning” and “artificial intelligence” interchangeably are not way off the mark. However, for technical purposes and for accuracy, it is important to draw the distinction between these two and know where one ends and the other begins.
For example, it will be totally inaccurate to refer to the technologies used in Facebook face recognition and autonomous cars as “machine learning” as these technologies, though they are AI-based, are built on rule-based systems. There is also a quest for an AI “Master Algorithm” that will not be based on machine learning.
If you are still lost and trying to imagine what a general or strong artificial intelligence would look like now that machine learning is considered a less than satisfactory implementation, try to imagine The Terminator! That is a self-motivated machine with human-level intelligence.
