What is Machine Learning?
Artificial Intelligence (AI) may be the great buzzword of our times but Machine Learning (ML) is really the brass tacks. For general conversation, however, the terms have considerable overlap. Some decades ago, a chess playing programme that beat human players could’ve been crowned ‘artificially intelligent’ but now there are mundane AIs, like the ones Amazon uses to recommend purchases and ‘true’ AI that theorists reserve for a point when machines become intelligent enough to take over the planet.
However the path to achieving this is due to advances in designing algorithms, cheap hardware, access to ‘training data’— or datasets of all kinds of stuff that computers can use to ‘learn’ new associations — that have allowed computers to perform several useful tasks better.
How have machines learnt to learn?
Computer science in the 1990s had laid much of the theoretical background for machine learning namely via developing neural networks. This involved, in essence, reviving a philosophy of designing circuits to simulate the way neurons connect in the brain.
The brain with its billions neurons and each connected to a 1000 others is now the dominant metaphor for how ML programs are organised. Rather than older approaches that tried to program the most ‘efficient’ way to solve a problem (like what’s the best sequence of moves to checkmate) ML systems are increasingly organised around letting the systems figure out the rules from scratch. Circuits achieve their goals — differentiating cats from dogs and recognising signatures on cheques—by repetitively blitzing through ‘rewards’ and ‘penalties’ and are limited only by the efficiency of the underlying algorithms and computing power at their programmer’s disposal.
Realistically, what has ML achieved so far?
Using approaches of Deep Learning — an approach where layers of ‘neurons’ are hierarchically arranged to recognise an object — machines can beat human champions of games that require computation and intuition, such as Go. More usefully, it can look at pictures of biopsies and picking out possible cancers. Some ML (or AI) are being taught to predict the outcome of legal cases, writing press releases and composing music. However the sci-fi future where a machine beats a human in every conceivable department and is constantly self-learning isn’t a reality yet.
