M is for Machine Learning

From A to I to Z: Jaid’s Guide to Artificial Intelligence

Machine learning is the process by which AI masters a specific task. 

First, the AI is trained to perform the task using a statistical model or algorithm. Once trained, it can perform the task on new data, and get better over time. 

There are four main types of machine learning:

1. Supervised learning

Remember those school workbooks that had the answers at the back?

Supervised learning is the AI equivalent. 

The AI is fed data that’s already been sorted. Its job is to try its hand at organizing the data, and then compare its results to what the trainers have supplied to make sure it did it correctly. 

Imagine you wanted to train AI to perform a classification task, for example to detect whether a customer query is about billing or product-related. 

In supervised learning, you’d feed the AI a list of customer queries that have already been sorted. 

The AI would then attempt to sort the queries itself. If it puts the queries in the same category they’ve been assigned in the data set, it means it has learned how to perform the task. 

Aside from classification tasks, supervised learning is also useful for training AI to recognise patterns and perform so-called regression tasks — tasks such as predicting stock market prices, where AI predicts an output based on a number of inputted variables.

2. Unsupervised learning

As the name suggests, unsupervised machine learning is a process in which the AI figures out patterns on its own. There are two main unsupervised learning techniques:

• Clustering, in which the AI has to sort unstructured data into meaningful categories
• Dimensionality reduction. Here, the AI’s job is to remove irrelevant data points so the data set is more accurate

The GPT language model was trained using unsupervised learning. The technique is also useful for training AI to detect anomalies, for example pinpointing which items on a production line are likely to be faulty, and to detect relationships between seemingly unrelated patterns, such as which products you’d likely be interested in based on your purchase history.

3. Semi-supervised learning

In semi-supervised learning, the AI is fed both structured and unstructured data. The AI’s job is to sort the unstructured data using the structured data as a reference point. 

Semi-supervised learning is useful for fine-tuning AI’s classification and anomaly-detection skills. 

It also saves time and money compared to supervised learning, because the amount of data that has to be pre-labelled is much smaller.

4. Reinforcement learning

In reinforcement learning, the AI learns by trial and error. It gets rewarded for taking appropriate action, and penalized for taking inappropriate action. Over time, it learns to take the actions that will result in the greatest reward. 

Reinforcement learning is useful for training AI to play games, control systems — traffic lights, for example — or make recommendations based on consumer behavior, such as suggesting movies based on viewing history. 

It’s also used to train robots to perform physical actions like grabbing objects.

Some facts:

The first machine learning model was a decision tree developed by AI pioneer Arthur Samuel in the 1950s. The model taught a computer to play checkers by looking ahead at all the available moves and picking the best one. 

The computer eventually became a very good player, though it never reached expert level. 

Today’s machine learning models are orders of magnitude more powerful than Samuel’s model, and can handle much larger volumes of data in a fraction of the time. 

A machine learning engine developed by UCLA in 2018, for instance, can solve complex mathematical problems at the speed of light. 

Machine learning and AI are sometimes used interchangeably, but they’re not the same thing. 

An AI is a computer that can think independently and apply knowledge to new situations. Machine learning is the process by which the AI acquires that knowledge and improves its abilities.

Want to know more?

Jason Brownlee’s Machine Learning Mastery is jam-packed with machine learning resources, from basic guides to deep dives and tutorials to test your knowledge. 

In this video, Google’s DeepMind teaches itself to walk without any human input. The results are impressive, if a bit goofy.

Jaid’s perspective

Machine learning is what makes AI such a powerful tool. Not only does it make it possible for AI to master any skill in a relatively short amount of time, it can also discern patterns humans might miss, with greater accuracy, and to keep getting better over time. 

This is potentially transformational in any number of fields. You can’t expect a human to remember a customer’s currently open inquiry, previous communications with them, and product portfolio at the drop of a hat. But AI can do this and much more, including cross-referencing similar inquiries, seeing which resolutions had the best outcomes, and honing its response to increase customer satisfaction. All without any human intervention.

Optimize your customer service experience by utilizing Jaid’s AI-powered platform – contact us today!