Artificial Intelligence vs Machine Learning: What’s what?
You’ve probably heard the words Artificial intelligence (AI) and Machine Learning (ML) a lot. As an industry professional, I hear these words too. Most of the time, these terms are used interchangeably. While these two concepts are different, there are some commonalities between them.
In this article, I hope to break it down for you:
Let’s start by understanding what “Artificial” from Artificial Intelligence really means. “Artificial” means something made or produced rather than occurring naturally. Okay, that’s a start but then, what does “natural” refer to? In this context, “natural” refers to a “human” — you and I. The field of computer science is fascinated by humans. Why? Because understanding and replicating human-like intelligence within a machine is a huge undertaking. Scientists have been working for decades to advance the research and implementation in this field. Robots and driverless cars are some examples of human-like intelligence, but there’s a reason you don’t have robots cleaning your house or doing the laundry for you (yet!); there is still a long way to go as far as using human-like objects in our daily lives.
So now that we know machines aren’t human, we can safely say: without any type of intelligence or logical instructions, machines are remarkably dumb. But what they lack in intelligence, they make up in huge amounts of computing power. Computations that might have taken you maybe a second to do, a machine can do 1000 of them in a fraction of the same time.
Furthermore, these machines will do exactly what we instruct them to. Nothing more, nothing less. They can’t change the course of action beyond what they have been instructed to do. If I give a set of 10-step instruction to both a human and a machine, the machine will follow those 10 steps exactly while a human has the ability to course correct with additional steps if the 10 steps provided are not sufficient to result in a desired outcome. Essentially, we need to provide machines instructions for what to do in the event of every possible scenario that a human can think and act in given context. Sound hard? Yeah, you bet it is.
To understand artificial intelligence, we need to try to understand how human intelligence works. In a nutshell, there are three underlying components as to how human intelligence works (there are other biological and scientific phenomenon that also drive the workings of human intelligence, but we should leave that for the neuroscientists to explain!)
- Perception through a multi-dimensional data source
If I give you a small object and ask you what it is, you’ll use your eyes to see, hands to feel, nose to smell, ears to hear to recall any information you may have about that object.
2. Pattern recognition within the data set referred in step 1 above
You may or may not have seen this object before. Your mind has constructed patterns in order to catalog the thousands of objects you’ve encountered over your lifetime. If you’ve seen this object before, the pattern recognition technique will help you identify/predict what this object is. If you have never seen this type of object before, you may not be able to recognize it at all or may recognize it incorrectly.
3. Decision making within the given context
As you recall from your learning (understanding) from your memory in step 2 above, logical argument and an evaluation process takes over. As human, we tend to (over) rationalize things (don’t tell my wife that!) and emotionally make decisions using what we call our “gut feeling.”
Take a moment and think about which of the above components can be replicated for AI. Turns out, they are all pretty difficult, but the first and third components are the most difficult to replicate. At a high level, if all three components can be replicated or nearly-replicated in a machine, then we have an artificially intelligent system that can almost think, behave, and act like a human, depending upon the extent to which all three components are implemented. As a thought exercise, how do the three components articulated above apply to a driverless car that has to make it out of your garage, down your driveway, and onto the street?? Share your responses below.
Let me state that machine learning is not equivalent to AI, but rather, an essential part of AI. To that effect, machine learning is really centered around the second component in the section above: pattern recognition. Machine learning helps to identify patterns within data sets and thus, tries to make predictions based on existing data it has learned from.
Before we dive deep into machine learning, let’s refresh on what human learning is to relate it properly to machine learning.
Let’s say I ask you, “what day is it tomorrow?” and your response is, say, “Sunday.” Take a moment to think about the process from learning to retrieval of your response.
You must have learned when you were a kid in pre-school that there are seven days in a week (Sunday-Saturday). That is learned data, my friends, and is stored somewhere in your memory.
Along with the data, there is a pattern (learning) stored in your memory as well. In this case, learning refers to how a given day (input) is associated with the next day (response).
In the table above, learning is stored as input and response. Please note that this learning is very specific to association of the current day to the next day. There can be different learning on the same set of data to answer different types of questions e.g. “What was the day yesterday?”
So, when someone asks you “What day is it tomorrow?” you will use the current day (today) as the input and refer to the learning table above and find the response. Initially, you may not realize that you’re following these steps, but trust me, this is what is happening behind the scenes.
Please note that this is a very simplified explanation of human learning and deals with a very small set of data (1 input, 1 response and 7 rows). In real-word business problems, there can be thousands of input columns and millions/billions of rows to get a response (output). As a dataset scales in size with more inputs and rows (observations), the logic (learning) becomes increasingly complicated. This is where becomes a challenge for humans, and prime for some good ol’ machine learning. Machine learning is the development of algorithms /models by data scientists and/or machine learning engineers to make computers learn, recognize patterns, and make predictions.
Amazon’s product recommendation and Facebook’s news feed are two great examples of machine learning. In the case of Facebook, when someone frequently reads or likes a certain post, he/she will see more of that particular friend’s activity in the future. Behind the scenes, your navigation data is captured and stored. A machine learning algorithm is able to learn from these patterns or human signals that you, human, are providing. The technique that discovers pattern is called an algorithm or a model. To make things simple, we focused this discussion on pattern recognition, but machine learning is not just limited to pattern recognition; it can also predict (forecast) an output value based on automatic discovery of relationships between several inputs and response (output) variables.
In future articles, I hope to explain more technical concepts in artificial intelligence and machine learning.