(Blog) AI and Machine Learning are not the same thing
This article identifies the commonly jumbled fields of AI and machine learning.
AI, expert systems, machine learning, deep learning, neural networks, expert systems and robots. They have different names, so they must be very distinct right? Well, not really.
Not only do these fields feed from one another, but they are still developing and that makes it difficult to pinpoint exactly where one definition ends and another begins. That being said, there are a few distinctions that we can make to help understand them more.
Artificial Intelligence (AI) is an encompassing term that refers to a computer program that can perform a task commonly associated with intelligent beings, such as: the ability to reason, discover meaning, generalize or learn from past experience. AI is broadly defined, and it is often used to describe other fields that have their own distinct qualities. In reality, AI is akin to a tree, with branches that have been grafted onto other trees (e.g., robotics, expert systems, machine learning).
Machine learning is a more complex application of AI in which a computer has the ability to learn without being explicitly programmed (Kudos Arthur L. Samuel for coining the term, and for that simply perfect definition). This is a HUGE help for tasks that are extremely difficult to program by hand (language translation, spam filtering, face recognition, robot motion). Machine learning uses advanced statistics and algorithms for a program to learn and make decisions, as well as, to increasingly improve its abilities as its fed more and more data. Some examples of machine learning are Netflix’s recommender systems, fraud detection for bank transactions, Amazon’s basket analysis that show which products are often purchased together.
Originally, robots were defined as machines that replicated human movements, such as performing difficult welding tasks. Now, a robot could also be a machine that replicates human functions and human intelligence. An artificially intelligent humanoid robot, Sophia, walks around and talks like a human by using neural networks, expert systems, natural language processing, and adaptive motor control among others. Sony’s robotic dog aibo can express emotion, seek out its owners, detect words and facial expressions.
Machine learning programs use large database to learn from; the more data, the better the program. Expert systems are unique in that they use a large expert knowledge database of facts and rules to make intelligent decisions. Similar to a doctor - but a doctor with perfect memory and great indexing ability. The primary goal of an expert system is to resolve issues which generally require a human specialist. One example is the Preterm Birth Risk Assessment, an expert system that has proved to be more accurate than traditional techniques for predicting preterm pregnancy risk. NASA’s Space Shuttle Mission Control is an expert control system that monitors, detects, predicts, and repairs errors in on-board spacecraft system behaviors.
Computer vision, speech recognition systems and natural language processing all fall under AI, and they often function using machine learning, but they can also function without it. Voice recognition is used in smart assistants like Apple’s Siri and Amazon’s Alexa to comprehended their user’s words and respond accordingly. Natural language processing and speech recognition is used in Sonix audio to text machines. Self-driving cars utilize computer vision to recognize objects. Computer vision, speech, and natural language processing are core areas of AI which are often used in tandem with other technologies.
Deep learning is a sub-category of machine learning. So it is in fact machine learning, but it is machine learning with a superpower. While machine learning requires humans to program its logical operations (in other words its “rules”), a deep learning program is built to teach itself the rules. It locates relationships within the data it is given, patterns that even humans may not recognize. This is possible by its tens-to-thousands of layers of artificial neural networks, compared to the traditional one-to-three layers given in machine learning - hence the name “deep learning”.
What are artificial neural networks? An artificial neural network(ANN) is easy to understand when you think of the neural networks within the human brain which function like cross-crossing highways of different width and strength. Think of how we learn; given an input of placing our hand on the hot stove, the output is the pain we feel. ANN is an information processing model that operates by processing the input through a series of calculations and values. It is given desired outputs, and as more inputs are processed, the model “learns” and slowly adjusts to get the desired outputs. As stated before, deep learning is artificial neural networks with more hidden layers, and thereby it has the ability to learn on its own.
Google’s AlphaGo is a deep learning model that learned how to play the game, Go. It can make specific moves without being programmed to make those moves, and it has learned to beat world-class Go players. Deep learning has opened a world of possibilities because it can see things from data which we do not see. However, it does have its faults. Deep learning does not comprehend common sense, and it picks up biases and replicates them in its conclusions. So, there is plenty of room for development.
Now you might have asked yourself, “Wait. But what if someone makes an evil robot?”, and, “Did iRobot and The Terminator teach us nothing about smart technology being used incorrectly?”.