The Nuanced Truth: Exploring Differences Between ML and AI

Everyone has heard of machine learning and artificial intelligence at this point, right?

There is an insane amount of buzz around ML and AI, and for good reason. The techniques available to us today allow for mind-blowing applications that would have seemed like magic just a few years ago.

I’ve researched all that I can across the web, and I’ve come up with an analogy: Machine learning is a kind of tool that we can use to make our computers do what we want. Every time you learn a new machine learning technique, you add another tool to you toolbelt.

When you have an assortment of tools at your disposal, you might try your hand at building something rather complex with them, something that requires all of those tools to work together in unison. It’s likely that you’d try to build something that exhibits artificial intelligence. In this way, machine learning techniques are the tools, artificial intelligence is the house, and you are the carpenter!

Let’s dive in a bit deeper and see if this analogy holds true.

Artificial Intelligence

Weak AI vs. Strong AI

Strong AI, or Artificial General Intelligence, is something that “can solve many types or classes of problems and even choose the problems it wants to solve without human intervention” [1]. There is still no real-world example of Strong AI — it’s only a concept. There are plenty of science fiction examples though, but not all of them present a favorable view. Skynet, anyone?

Machine Learning

Though closely related, ML and AI are distinct topics that must be considered and understood separately. Gary Sims from Android Authority sums it up quite well in his YouTube video on the topic [4]: Machine learning is a process by which a machine learns from its experience, and generally involves recognizing patterns and doing something useful with them. On the other hand, artificial intelligence involves creativity, abstract thinking, and analysis within a specific context. While achieving artificial intelligence will almost certainly require the use of ML techniques, it is clear that AI refers to a high-level concept of a computer’s abilities, whereas ML generally describes techniques for pattern-matching and extracting specific insights from a dataset.

So why is this so confusing? Gary hints at an answer: “artificial intelligence” tends to double as a marketing technique. When you hear a company claim that they use “artificial intelligence” without explaining how, they’re usually trying to sell you something.

Types of Machine Learning

  • Is your data labeled or unlabeled?
  • How much data will be available?
  • What type of problem are you trying to solve?

Some algorithms can be used both for labeled and unlabeled data. For example, “Nearest Neighbors” is a broad category of algorithms. Scikit-learn, one of the most popular open-source libraries for machine learning, says that their neighbors package "provides functionality for unsupervised and supervised neighbors-based learning methods" [5].

However, focusing on whether the data at hand is labeled provides a starting point for divvying up the algorithms into their most common uses.

Supervised Learning: When Data is Labeled

A few examples of supervised learning include regression algorithms, decision trees, and instance-based algorithms.

Linear and logistic regression algorithms are some of the simplest ML algorithms that we have. They are essentially trying to fit your training data to a curve, one that either looks something like y=mx+b or something involving logarithms. As long as your data can actually be modeled by these curves, it could be a great choice.

Decision trees and instance-based algorithms are some other choices for supervised learning.

Unsupervised Learning

Clustering algorithms are aimed at forming groups of data points. They try to find some kind of similarity between points, then it will form a group containing those similar points.

Association algorithms attempt to uncover how data points are related to one another [6]. Such algorithms can automatically find correlations that may not otherwise be obvious. This is the kind of algorithm that may find out that you’re pregnant before you do based on your shopping list.

Neural networks attempt to model the neurons in the human brain to process data in successive “hidden layers” of calculations. Depending on how many hidden layers are involved, the use of neural networks could be considered deep learning.

Reinforcement Learning

This awesome video shows an example of reinforcement learning in action — someone trained their computer to play Super Mario!


I hope you learned something new from this article about these topics. Please let me know if you have any thoughts on these topics (or corrections if I made any mistakes).

If you’re excited about these topics and want to get your hands dirty, check out my next post about how to install scikit-learn!








Featured image by Alex Knight on Pexels

Originally published at