Machine learning – a successful spinoff of AI
Artificial intelligence (AI) is a term that has kindled the imagination of computer scientists for the last six decades. It was their moniker for the dream to build applications that resembled or even surpassed human intelligence. The field started in the late 1950s and has since gone through several cycles of optimism and doom. Each time the ambition and enthusiasm rose to great heights, to eventually clash with the complexity of the task and the available computational power.
The last years have seen a new upturn. Lately, we’ve seen computers beat champions of chess and even Go, said to be the most complex game played by humans. We’ve seen computers dialogue, discuss, and drive cars. So this time around, AI seems to be successful. The reason? The available computational power has risen exponentially to where we are today. And the ambition of general AI has been curbed sufficiently to match that power: computers now learn to recognize patterns in huge, seemingly random datasets. It’s called machine learning (ML) and it is no mean feat.
Machine learning as we know it is either supervised or unsupervised.
Supervised ML is the technique that has been reaping the most successes so far. This is how it works. An ML system is presented with a large dataset and a task. Say an array of pixels, a map of weather data, or a history of body parameters in which it has to recognize a face, a storm, or a disease. In the learning phase, the system will be confronted with input, make predictions and then get feedback through the labels that were pre-attached to the input – correct or not. If its prediction is false, the ML system will tune its parameters (also called weights) and make a new prediction. Over and over again, until the parameters are fine-tuned to make correct predictions most of the time. After the learning phase, the system is ready to mine huge data streams on the lookout for meaningful patterns, a process we call inference.