The computer scientist John McCarthy first coined “Artificial Intelligence” in 1955. Since then, new terms, like “Machine Learning” and “Deep Learning”, have come into the public arena. Today, the three terms seem to be used as if they meant the same thing. This is far from true.
Artificial Intelligence
Artificial Intelligence is the application of human intelligence to a machine. At any level, it involves a machine using data to make decisions. Crucially, the level to which a machine can perform a task with the help of artificial intelligence increases with the amount of data received. There are different ways through which to achieve artificial intelligence, some of which result in a deeper level of ‘intelligence’ than others. Suppose the task in question is reading.
The First Level: Explicit Programming
You are a child, being taught how to read. Your teacher points to words on a sheet of paper and says the first word. You repeat it after him. He points to the next word and does the same; you obediently repeat. This continues until all the words have been said and repeated. This isn’t reading. You are copying the words that you hear your teacher say. You are mimicking your teacher’s intelligence.
On the most basic level of programming, a machine will receive data, such as explicit instructions, on how to perform a task. Importantly, there is a large section of explicit programming that lies outside of AI and includes areas like problem-solving. A child can be ‘programmed’ into mimicking the words that the teacher says, and a computer can be programmed into completing a particular task.
The Second Level: Machine Learning
Instead of making you repeat what he says, the teacher gives you the same sheet of words and a list of letters and digraphs (such as “sh”, “ch” and “th”). The teacher makes sure you know how to pronounce these sounds beforehand. He then asks you to read the words from the sheet. You do your best, only knowing how to make noises associated with each letter and digraph. The teacher nods when you pronounce the word correctly and shakes their head when you don’t, giving you feedback. This continues until the teacher has nodded for each word.
Machine learning is a way to achieve artificial intelligence. It includes the ability of a computer to utilise a feedback loop to make better decisions in the future. Machine learning also relies on the initial labeling of different features, and of patterns when feedback is given. In the reading example, the child gets a list of “labels” — how to pronounce every letter and digraph. They also get feedback on the way in which they pronounce each word, which they then implement to improve. When the child moves on to reading sentences, the teacher will also give feedback on sentence structure. A Machine Learning algorithm recognises patterns, but to a limited level of complexity. More complex patterns need to be pointed out by a human.
The Third Level: Deep Learning
The teacher gives you a sheet of words and asks you to read from the sheet. You don’t know how to read, so you randomly make noises. The teacher nods when you pronounce the word right and shakes their head when you don’t. They still give you feedback after each attempt. You keep trying to ‘read’ each word until the teacher nods for each of them.
Deep Learning is a method of achieving machine learning, by identifying patterns and classifying information into types. The crucial difference between other types of machine learning is that deep learning establishes patterns in data by itself, without the need for a human to point them out. In the reading example, the only ‘input’ which the child receives are the different words and letters themselves, the raw data, and not information on how to pronounce them. Crucially, a human still gives feedback, such as adjusting the importance of different variables (called hyperparameters) in the algorithm. A Deep Learning algorithm would be applicable to a wider range of data sets than a Machine Learning algorithm. It would also recognise more complex patterns, though the extent of this complexity is not yet fully known.
Artificial Intelligence, Machine Learning and Deep Learning are terms that all industries will be familiar with in the future. They will underpin the development of the global economy and have massive social impacts. Already, we are seeing the development of Artificial Intelligence in driverless cars, finance and Optical Character Recognition (OCR). Like many technologies, Artificial Intelligence will develop in ways we cannot even imagine. All we can do is understand it, use it, and build it for ourselves.
