If you were born in the 80s or 90s, then there’s a good chance the idea of machine learning conjures images of Arnold Schwarzenegger as The Terminator. It sounds like a far future concept that may be synonymous with AI and ‘the rise of the machines’.
But as it happens, the rise has actually already begun. Machine learning is no science fiction concept but rather a concrete tool that you likely encounter on a daily basis already – even on the Android device that’s currently sitting in your pocket. Learning. Waiting.
In this post, I’ll attempt to demystify machine learning a little and demonstrate a few ways in which it is already a big part of your life. And far from wanting to ‘take over’, you’ll hopefully see that it’s actually just here to help. For now, anyway.
What is machine learning?
For the best summary of machine learning, you should definitely check out Gary’s article on what machine learning is, along with the accompanying video. To cut a long story short though, machine learning is not the same thing as AI, although the two subjects are closely related. In fact, machine learning also shares a lot in common with data mining and statistical analysis.
Machine learning is concerned with helping a program to get better at a specific task, often through the collection and subsequent analysis of large data sets that allow patterns to emerge.
For example: say you were you to speak with an AI on WhatsApp, many aspects of its behavior would be pre-programmed examples of artificial intelligence. But if it could also asses the language that you were using and your responses and then use that information to come up with more realistic and human-like vocabulary, then that would be an example of machine learning. This might work using a database of common phrases, responses and interactions, for example, that could be added to and iterated over time.
That’s just a hypothetical, but there are plenty of much more remarkable examples of this that you already encounter every day.
Spotify and Netflix
Here’s another example of something you might use every single day that is constantly collecting data about you as you do: Spotify. Spotify uses a form of rudimentary machine learning to work out the kind of music you like. It can then use this to create playlists of suggested tracks – something I always look forward to on a Monday morning while I’m working. At least until it starts playing Westlife because my wife has been using it. Deezer and Pandora also use similar features.
Partly, this is based on genre or other categories, which are likely tagged by a human and thus not an example of machine learning. The more advanced systems though will look at the behavior of similar users and at statistics such as the pitch, the tempo and the length in order to identify the type of music that you play the longest and most often.
Of course, similar strategies are also used by Amazon to recommend products you might like and by Netflix to suggest the next show you might want to watch.
Gmail is just the first example of Google using machine learning in its products – there will be plenty more on this list. Here, machine learning is used to categorize your emails and filter out spam. Spam filters of old would have worked based on a number of key ‘spam words’ that the programmers would have programmed manually. Machine learning takes this a step further though by learning from the way that users react to certain types of content. Eventually, patterns emerge that allow Gmail to intelligently distinguish between the kind of content that we don’t want and the kind we do.
Skynet: protecting you from handbag and Viagra adverts since the 1980s.
A slightly more impressive example of machine learning is Facebook’s ability to identify people in your pictures and recommend them for tagging. This is, of course, based on data collected over the course of many, many tags that allows the program to spot patterns emerging. These numbers might represent things like skin tone, contrast and the width of a person’s nose and those patterns can then be used to identify individuals. It still doesn’t always get it right though; Facebook once suggested that I tag a bowl of fruit as my friend Tiller!
Gradually, we will begin to see similar technology employed in CCTV cameras to aid with the apprehension of criminals, and in stores to recommend the right products to the right people.
Gradually, we will begin to see similar technology employed in CCTV cameras to aid with the apprehension of criminals, and in stores to recommend the right products to the right people. The scary part? This has already begun.
Computer vision and digital assistants
The previous example is not only an example of machine learning but also of another area of computer science: computer vision.
Computer vision is what makes things like AR and mixed reality possible. This is how a computer can view an image and identify what’s in it, which one day might allow Google to purely search through the actual content of images rather than just sorting them based on file names.
Computer vision isn’t entirely accomplished via machine learning but is a big part of it. Machine learning is what enables the collection of the necessary data sets that can be used to navigate surroundings and eventually allow machines to identify new objects correctly – we’ve already seen Bixby Vision on the Galaxy S8 attempting something akin to this. Similar techniques are also used to improve voice recognition, which in turn gives rise to services like the Google Assistant and Siri.
Both these examples show how one technology can give rise to another and the really exciting part is that Google is actually letting developers access their Cloud Vision API – so you can take advantage of this technology in your own apps.
If you want an example of machine learning that does sound a little like it came from a George Orwell novel, then consider the case of advertising. This isn’t what you think. For the most part, the adverts that show you things you’ve already looked at (or things very similar to what you have looked at) will use other methods like cookies in order to part you with your cash. Cookies are files stored on your computer that allow a website to recognize you, so there’s no learning going on there.
Stare long enough into the net and the net stares back.
But the position of the advert and even its color and size may be a result of machine learning. That’s because pay per click advertising platforms can observe which positions and which ads get the most clicks and then rotate and arrange them accordingly to ensure the publisher (and therefore the ad network) gets the maximum profit.
Stare long enough into the net and the net stares back.
While machine learning might be costing you money through advertising, it could be saving you money at the bank. Here, software is used to look for patterns that might point to debit or credit card fraud. No team of human employees could check the accounts of every single client – but a bot can, and it can learn your ‘normal’ patterns of behavior in order to effectively pin-point anything that seems out of the ordinary.
In the future, we might also see machine learning guiding investment strategies to a greater degree. Maybe it will be a computer program that finally ‘finds alpha’.
Science and medicine
While you’re using machine learning for your social media and ‘Netflix and chill’ activities (yes, Tinder also uses machine learning), the same technology is meanwhile being used to treat illnesses and potentially make scientific discoveries. Doctors are now beginning to use machine learning in their diagnoses, looking for patterns that can point to eye disease or cancer.
In particle physics, machine learning has been used to identify patterns in the exponential quantities of data generated by the Large Hadron Collider. In fact, it played an instrumental role in the discovery of the Higgs Boson. In future, scientists hope that it could uncover ‘new physics’ – ideas that no human has yet thought of!
There is an exciting future for the possibilities of machine learning in research and medicine. Learning computers might represent the first rung on the ladder on our ascent toward the singularity as they transform the way we do business and make breakthroughs that no human could ever have dreamed of.
But machine learning also has a huge amount of potential for our daily lives; in entertainment, retail and communication. The effects of that are already being felt across many of our everyday tasks in fact, and perhaps that is the most exciting part of all.