Nick Cowen

The phrase ‘machine learning’ can mean different things to different people. To those out of the loop with regards to the latest advancements in technology, it can recall every nightmarish science fiction scenario in which mankind’s recklessness in the development of Artificial Intelligence (AI) put paid to civilisation. Movies such as The Matrix, The Terminator and Transcendence leap to mind. But to those more acquainted with it, machine learning is not only one of the more useful processes developers have created; it takes place in millions of devices every day. The smartphone in your pocket is engaged in machine learning every time you pull it out and fire up an application. It is the process that allows AIs to learn, thanks to data shared between devices. The collecting and sharing of data allows them to make helpful predictions based on a user’s experience of a specific device. One of the more common examples is when a smartphone owner decides to use the virtual assistant on their phone to bring up a contact or open an app. Siri and Google Assistant began life as simple voice-activated shortcuts, but because of machine learning, they can now understand nuance, semantics and even tone. As smartphones become more advanced, the internal tech has become more geared toward the task of machine learning and both virtual assistants have evolved in their capabilities. The aim with both has been to make using a smartphone a more intuitive and personal experience, while providing Apple and Google with a ton of data about the likes, dislikes, preferences and aesthetic tastes of millions of consumers. If you’ve ever wanted to know how Google’s algorithms know which adverts you’re more likely to click on, the answer is machine learning. The more information you share, the better your personal experience with an app or device is likely to become. But this is just the tip of the iceberg. As developers become more ambitious in their quest to improve algorithms, bots and AI, they’ve turned their gaze to grander projects. As AIs gain access to an ever-increasing mountain of data, their ability to interpret and analyse it is improving all the time. A lot of the research on machine learning is geared toward passing off tedious tasks to automated assistants, WHAT’S UP Machine learning Everything’s going to be AI which frees up humans to spend more time on bigger projects. Google’s DeepMind project is arguably one of the world’s biggest machine learning endeavours. Up until a couple of years ago, the AI had mainly learnt how to master the ancient strategy game Go, and it wasn’t long before it had established itself as the world’s best player. Not content with this success, Google engineers have branched out into other games, including chess. The company has even tapped up the world’s most bankable videogames publisher, Activision Blizzard (makers of World of Warcraft and Call of Duty, among others), to help DeepMind’s capabilities expand. In order to take its AI to the next level, Google has had the AI start playing real-time strategy game StarCraft 2, in which pieces are assigned certain skills, but are always moving as each match progresses. Activision Blizzard has opened this experiment, allowing approved AI developers to help with improving Google’s results. Perhaps the most impressive advancement in the field of machine learning took place last year in March, when Google Brain announced AutoML, which is an AI that’s capable of making its own AIs. As has been the case in the past, the aim behind this machine learning – or in this case, what Google’s researchers have called ‘reinforcement learning’ – was to create ‘child AIs’ to take on specific tasks and carry them out faster and more efficiently than the parent AI. According to the researchers, one of the child AIs, called NASnet, was tasked with recognising objects – such as cars, people, kites etc – in video, in real-time. The parent AI would evaluate the results in order to improve the recognition by repeating this action thousands of times. Astonishingly, this process led to NASnet outperforming all the humanmade AIs that had come before it. Google has said that AutoML and its ‘children’ could have a wide range of applications and has opensourced the AI to allow developers to build on and improve it. The development is being handled responsibly, however. Silicon Valley’s leading companies – Facebook, Google, Amazon, Apple etc – have banded together to provide such technological advancement with checks and balances. After all, no one wants to see the rise of the robots anytime soon.