Is Machine Learning Taking Over?
What began in 1952, has become a major focus for the future of technology. What will machine learning really achieve?
Machine learning might sound like the plot of a science fiction film, but it’s already a reality. The science behind computers learning from their own experiences rather than merely following instructions. machine learning underpins many products and services we now use on a daily basis.
Evolution of Machine Learning
The principles of machine learning can be traced back to a 1952. This year saw a program that enabled a computer to improve at checkers the more it played. Fifteen years later, the first pattern recognition algorithm was created. The 1980s saw computers gaining the ability to discard irrelevant information. However, it was the 1990s when machine learning began to take off. Large-scale data processing allowed the increasingly powerful CPUs of the time to make decisions based on available evidence, rather than just obeying program code.
This is the technology underpinning modern machine learning platforms like voice-controlled smartphone assistants. Siri, Alexa and Cortana are designed to recognize and interpret our voices, picking out keywords or phrases and responding in the most appropriate way. Like the checkers software, they learn from their mistakes in an attempt to become more knowledgeable. Unlike the checkers program, personal assistants can respond to varied inputs and answer requests. The technology sector is ablaze with speculation about how virtual PAs could one day interact with anything. We will see them anywhere from autonomous vehicles to smart home electronics.
Machine Learning On Demand
Perhaps the best example of machine learning involves curated recommendations. On-demand media services and online shopping portals use algorithms that scrutinize our preferences. They also attempt to identify products or services we may like. These recommender systems harvest data from previous activity, public-domain information and stated preferences. The burgeoning on-demand market is pushing algorithm developers to ever-greater heights of sophistication. Especially when it comes to those “you might also like” suggestions.
Technological necessity is a key driver of machine learning advances. Another example of this involves email filters. Google’s Gmail service famously claimed to catch 99.9% of spam back in 2015. Their neural networks scan vast amounts of electronic mail and creating new rules in response to spam’s constant evolution. Machine learning represents the best way to tackle moving goalposts, since a fixed set of spam identifiers could quickly be bypassed by junk mail senders.
AI’s Exponential Value
An electronic rival to the connections in our brains, neural networking has been used to power the PA software we just discussed. The machine learning infrastructure enabled an AI system to beat one of the world’s finest Go players. The 48 CPUs powering Google’s AlphaGo algorithm set a new benchmark for neural networking. This allowed AlphaGo to win a best-of-five competition in the most complex board game ever invented. Similar computational resources have subsequently been applied everywhere from CERN’s Large Hadron Collider to autonomous vehicle trials on every continent.
Of course, autonomous vehicles have been involved in several high-profile accidents over the last year. Machine learning is an inexact science. There will always, of course, be decisions that are best left to people. Nevertheless, there is immense potential for identifying nascent health conditions, or recognizing fraudulent financial transactions. Machine learning won’t claim many headlines, but it’s going to play an increasingly key role in our lives over the coming years and decades.