As a digital agency in Melbourne, we’re constantly abreast of new and exciting concepts being floated in the digital industry. A growing catch-cry in tech industries (and one that is exciting a lot of people) is machine learning. Instead of programming a computer to complete tasks in orders A-B-C, machine learning is essentially about teaching a computer to look at data and build its own systems to process and provide insights. But rest assured, sci-fi fans, because this isn’t going to be like The Terminator, with machine’s eventually rising up to work together and overthrow humanity.
Instead, machine learning looks at data collection and analysis, setting up tasks and processes designed to teach the computer how to learn about user behaviour. This is the next step in web evolution, having started with business-driven Web 1.0, moving to user-driven Web 2.0, and now leading us to machine driven Web 3.0. Eventually, the internet will be nothing but suggestions to the user based on their data. This step is a long way off however, and you’ll find that at the moment, big companies like Google are invested in building the technology that will eventually allow this to be the case. But what does this actually entail?
If you look at a flow chart of what machine learning is at its heart, you’ll see data being churned by an algorithm that pumps out insights. Machine learning holds these insights and applies them to new sets of data, hoping to find the same result. See the below diagram.
For machine learning to really work, it needs clean data. Google believes that more than 90% of all data online is unstructured and messy, and can’t be parsed into something usable. For example, unstructured data doesn’t tell you what the above diagram is, simply giving you a title and an alt tag. Most of the work for machine learning at the moment requires data cleansers to give far more detail on almost everything, telling the machines that the diagram above is blue, highlighting data going into the algorithm and then the interaction between the algorithm and any insights found. Right now, more than 75% of all work done on machine learning is based on cleansing the data. Once that is completed, there are algorithm specialists who are constantly working to edit and adjust the algorithm to process the data effectively and deliver actionable insights.
Think about how Facebook knows where faces are in pictures. Before machine learning, each image was unstructured, and meant nothing to a computer, just a image of something. Facebook pushed its programming but building a set of data (random faces) and cleaning the data contained in each image (highlighting eyes, noses and mouths), ran the data through an algorithm designed to know how to highlight these factors combined to create a face, and then delivering that in the form of facial recognition. When Facebook felt as thought the algorithm was performing well enough to understand the data, it released it to the greater community, resulting in the facial recognition that you now see when posting images on Facebook, or even something like applying filters on Snapchat.
As more and more companies work to create machine learning-based API, websites and applications will begin to implement them into their product offerings. Facial and image recognition is just the start; this is about teaching computers to make relevant and useful suggestions in your everyday life. When you wake up, and look at your phone, machine learning will allow the software to immediately open Facebook, showing you relevant news articles that you may be interested in reading, or saving those for when you’re going to work an hour later. Machine learning allows much greater individualisation of your digital usage experience. As a digital agency in Melbourne, we’re very excited by the prospects that this can offer, and we think you should be too.