By Nazra Noushad, TKS 2017
Currently: Community Manager at The Knowledge Society
Originally published in 2 separate articles on Medium
With Artificial Intelligence reaching its inflection point, there are a few basic concepts that everyone in the tech industry should understand. One question that I always make a mental note of is the difference between A.I. and Machine Learning.
What is Machine Learning exactly, and is it the same thing as Artificial Intelligence?
By the end of this article, you should understand what Machine Learning is! Until then, the answer my second question is; nope, ML ‘isn’t just a fancier way to say “A.I.”
Defining Artificial Intelligence and Data Science to improve our understanding of Machine Learning.
What is Artificial Intelligence?
Artificial Intelligence is the ability of computer systems to perform tasks that usually require human intelligence. These tasks can fall under the domain of visual perception, speech recognition, and the ability to make decisions.
Earlier concepts of A.I. included that they were pre-programmed computer “intelligence.” An example of this is the ability to play chess, but only through a series of pre-programmed moves and counter-moves. In other words, an intelligence that can only behave based on things it’s already been told to do.
There are also two types of Artificial Intelligence: Weak A.I., and Strong A.I. Weak A.I. is the ability to complete a specific task (such as playing chess). Strong A.I., also known as Artificial General Intelligence (AGI) is the computer ‘system’s ability to perform any task, from driving a car, to detecting skin cancer. It can learn and deal with multiple variables.
What is Data Science?
Data Science is the study of scientific methods and systems to extract knowledge or insight from data in various forms. This interdisciplinary field consists of Data Mining and Data Analysis.
Data Mining implies pulling information from webpages, excel sheets, and databases. Whereas, Data Analysis is using this data to make a point, usually through charts and graphs. Finally, Data scientists can use this information to make informed and better business decisions.
So then, What is Machine Learning?
Where does Machine Learning fall within these two incredibly exciting fields of study?
Well, right in the middle. ML is a sub-field that falls under both A.I. and Data Science.
Machine Learning is a method used to create Artificial Intelligence because it is used to complete tasks that usually require human intelligence
Most importantly, it can create a Strong A.I. as the computer can upgrade its own intelligence beyond its pre-programmed intelligence. To do this, ML uses a combination of Data Mining and Data Analysis to gather knowledge to make a decision.
Machine Learning is the ability of a computer system to teach itself through 3 steps:
- Recognize Patterns
- Make Predictions
- Learn/Update its own pattern recognition
Through these three steps, machine learning can be used to do anything.
Artificial Intelligence has the opportunity to disrupt every industry, including; government (AI Judge), literature (AI writing Horror Fiction), and even healthcare (able to recognize skin cancer better than a dermatologist!).
Some may argue that ‘there’s a long way to go before this happens, but we’re living in an era where technology is changing exponentially. So you need to believe and understand that more than ever is becoming possible at faster and faster rates. So believe me, it’s close than you think.
I get what Machine Learning is, How does Machine Learning Work?
“Wait, so, you mean there’s more than one way in which a machine learns?”
The first part of this article made it sound so straight forward, didn’t it? A machine recognizes patterns, makes predictions, and learns.
Well, these next three super-simple steps are going to now help you understand the different ways in which a machine can learn. It may sound a little intimidating, but don’t worry; this next bit doesn’t get harder and I promise I’ll make it easy to understand.
1. Machine Learning Through Supervised Learning
Supervised Learning is a type of shallow machine learning, and it’s exactly what it sounds like.
This type of learning is when the developer labels the variables that the machine is using. Within this domain, there are two sectors of learning: regression and classification.
Regression is the machine’s ability to recognize numbers and group them to form predictions. An example of these variables can be the total area of a house in square feet, the number of bathrooms it has, and the number of bedrooms it contains.
Through linear regression, the machine can predict the cost of a house by grouping examples of different houses and learning to predict the price based on the variables and costs.
Classification is the machine’s ability to identify images, or binary things (yes’ and no’s). Think of this like playing flashcards with your machine! Your stack of cards contains different types of cats and dogs. The side you (the developer) can see is the image, and the machine sees the back of the card. The machine uses a number between 0-1 to guess what the animal is. 0 would mean the image is completely wrong, 1 would mean the image is right. It keeps doing this until it can distinguish between cats and dogs.
Machine Learning Flowchart, Type 1: Supervised Learning (regression & classification)
2. Machine Learning through Unsupervised Learning
In the first example of playing flashcards with the machine can be used to understand unsupervised learning. Except for this time, the machine is playing flashcards with itself.
On one side, is a cat or dog. On the other side, is the blank back of the card. The machine can randomly notice something special about the cats that are different from the dogs.
Let’s say that it notices that the cats all have pointy ears, while the dogs have floppy ones. The machine puts the cats into one pile, and the dogs into another, this is called clustering.
The machine can also notice several things at once. Perhaps it notices that the cats have four legs, green eyes, and pointy ears. It also notices that the dogs have four legs, brown eyes, and floppy ears. It can reduce excessive variables attached to the images, such as the four legs, and can distinguish the cat from the dog using fewer features; this is called dimension reduction.
3. Machine Learning through Reinforcement Learning
Reinforcement learning allows us to teach the machine through a reward system. The machine is given rewards for making the right decisions.
Let’s imagine the scenario of a machine paying a game and benefiting from scoring more points. Before the machine begins, it has 1000 points. The objective is to maintain and increase the number of points that it has. When the machine encounters a dead end, or takes the wrong turn, and hits a trap, and loses points. Then, when it makes it through the maze, or takes the right turn, leading him to the right path, it gains points.
If it makes the correct decisions, the machine is an incentive to make accurate predictions and actions.
When the machine can model the maze in its mind, and plan out a series of directions and steps it needs to take in advance, then it is using the model-based form of reinforcement learning.
If the mouse is learning through habit, by going through the maze multiple times, it is learning through the model-free method.
Machine Learning Flowchart With Definitions of Each Domain of Learning
Building AI Means Using Multiple Machine Learning Methods
These different types of machine learning are best for different functions. Supervised learning through regression and classification is better for speech and visual data prediction. Unsupervised learning is for random assortments of data, and it’s use cases include advertisement suggestions.
Reinforcement learning, which is learning via rewards, is used for creating physical robots. Note that despite their specialized use cases, it’s important to know and understand each learning model as they can be used together for various applications.
How you can learn more!
If you’re excited to learn more, find a youtube channel or an online course that helps you go deeper into ML! I’d recommend Andrew Ng’s course on Coursera.
Sign up to stay in touch with TKS and join us for events, workshops, or maybe even be a part of one of our future cohorts!
P.S. feel free to use and share my diagrams if you find them resourceful!