Definition of Machine Learning

Machine learning (ML) refers to the ability of a computer to self-learn, without explicit instructions. It can perform certain tasks autonomously. These algorithms improve automatically through experience, or learn from historical data.

We need to feed the machine lots of data before it can learn and predict. In many cases, humans learn faster. Yet, intelligence systems have more dimensions than humans. They can process huge datasets, recognize more patterns and make better predictions.

Self-driving cars

Self-driving cars is a field where this self-learning ability is crucial: the system that controls the car – that drives the car – has to assess all the different traffic situation and determine what action to take.

Driving a car is much more complicated that it may seem if you are navigating your way on the highway. Traffic is chaos, it’s not a predictable, linear process. In fact, when you are driving, you are improvising every day.

Self-driving cars need to navigate through chaos.

Traffic is chaos.

A first-generation computer system would have needed an instruction for each and every traffic situation, which is of course impossible to do. Self-learning systems don’t need all situations specified. It learns along the way.

That is where machine learning comes in handy. It can combine the various scenarios that the programmers foresaw when they designed the program to a multitude of unique traffic situations. Like you, it can improvise. It is autonomous.

INTELLIGENT MACHINES: WHAT’S DIFFERENT?

The first generation computers tended to follow orders. You, the user, would direct it to perform certain tasks and the computer would follow a struct set of rules to do what you told it to. It could only do what it was programmed to do. That limited the performance to what the persons that designed the programs taught the machine to do.

Computers are becoming more independent now. The intelligent systems built on machine-learning algorithms can be instructed to deliver a certain end-result and the computer itself figures out how best to achieve that – even if nobody installed that knowledge onto the machine. That is called self-learning ability.

See also General AI, Supervised learning vs unsupervised learning.