Machine learning is coming of age. Thanks to new technology, theories that have been in development for decades are now filtering through to real-world applications, and there are many more to come. Here is Easy Offices’ introduction to machine learning, and how it’s helping business now and into the future.
What is machine learning?
Machine learning (ML) is when computers analyse real-world data to predict future outcomes from new data. The “knowledge” they gain can then be used to influence future actions based on these predictions.
A good example is spam filtering. If thousands of people are getting spam emails and sending them straight to the bin, machine learning can start to look for patterns and similarities between the rejected emails. These can be far more subtle than anything a human could notice. For example, there could be small parts of the email’s code that match patterns in previous spam emails. Eventually, the computer can start to predict which emails are spam and reject them automatically. It also continuously gathers new data on the spam emails, which makes the system more effective over time.
How is machine learning different to artificial intelligence?
One machine learning definition is that it is a part of artificial intelligence (AI), i.e. it comes under the umbrella of AI, but it is a very specific kind.
The most commonly used example of the AI/ML distinction is a chess computer. Programmers can teach a computer how the pieces move, and because of its speed, it can calculate the millions of possible outcomes from the next few moves, and work out mathematically which would be the best move to make. It feels intelligent, but it’s actually just a sequence of sums.
If, however, the rules of chess were suddenly changed, for example if bishops could move like the queen, the AI would not know what to do because it would not have been programmed to play that game.
If the AI was using machine learning models, however, it could start to calculate possible moves under the new rules by seeing how the game unfolds. It might take many thousands of games, but eventually it would become as competent a player as the AI was at standard chess.
How can machine learning be applied to business?
We’ve already seen one example of machine learning in operation: the spam filter. But it’s being rolled out in all sorts of applications. Machine learning works better the more data it has to work with, so it’s great at spotting trends in activities that are carried out by millions of people or objects. Some of the most interesting are listed below.
Cars, vans and trucks are learning to drive themselves, not only by taking real-life data from human drivers, but also by learning as they drive themselves. They are proving to be excellent at spotting potential hazards crucial seconds before humans do. Logistics businesses are looking very closely at this technology, as there’s clearly huge potential to save personnel costs.
Now that phones have sat nav, they can broadcast driving data back to the service providers. This data can then be fed back to drivers. For example, if a few hundred cars are driving slowly on a stretch of road, the central computer might start rerouting other drivers so they avoid it. The potential savings for any business that uses roads are significant.
Cars are even re-drawing maps by sharing their precise positions as they drive. The machine can learn when roads are closed or if a road has changed from two-way to one-way simply by reading the real-world data from drivers. No manual action needs to be taken, and that’s the benefit of machine learning.
Often, the first voice you hear when you contact a call centre is a computer. And there’s a good chance the computer has never heard your voice before. So how does it know what you are saying, with your unique accent, tone of voice and talking speed? It has listened to thousands of voices and worked out what the words sound like over a spectrum of voices. And as you speak, it is adding yet another voice to its memory.
Machine-learned call centre automation is revolutionising the industry. And the same technology is being used by Alexa, Siri, Cortana and Google Assistant to constantly improve their accuracy.
Recommendations in eCommerce and streaming
Whenever you buy something online, you’ll no doubt have noticed that you’ll also receive a number of recommendations. These are not completely random, of course. They are learnt by the computer after thousands of transactions have built up personas around shoppers. So if you buy a tent, you might be shown barbecues, bug repellant and the latest literary blockbuster. These recommendations are not programmed by humans – they are learnt by computers. The opportunities for retailers to up-sell and cross-sell are huge, and eCommerce companies are investing heavily in making the technology even more focused.
Another application of machine-learned recommendations is in streaming services. No doubt you’ll have been advised on what to watch or listen to next on Netflix or Spotify. These recommendations come directly from learning what media tend to be popular among people whose recent consumption is similar to yours.
Servers going down and systems crashing are clearly bad news for online businesses and companies that rely on their digital infrastructures. Often, it’s down to the IT team to find a fix, and that can take time and money.
Wouldn’t it be better if computers could analyse all the billions of digital events leading up to a breakdown and look for common signals? That’s exactly what they are doing. As soon as the system detects seemingly minor issues that have foreshadowed critical breakdowns in the past, it can shut down or fix the offending systems before the screens go blue.
In the beginning, search engines were manually updated by people entering each new website in a category, for example “florists” or “solicitors”. Nowadays, search engines can automatically categorise a website simply by reading its content, looking at its digital neighbourhood (where links are going to and coming from) and comparing it with similar known websites. It means pages are indexed much faster, and users are more likely to be directed to the most appropriate pages when they make a search query. For business, that means that if you make websites that satisfy the search algorithms, you’re more likely to be found by potential customers.
One thing the examples above have in common is that the amount of data collected is immense. When you have millions of human interactions and billions of lines of code entering a system, it’s almost inevitable that patterns will start to emerge.
Learn to interpret those patterns and, more importantly, how you can use them, and you can make better-informed decisions – or even leave the computer to make decisions for you.
Machine learning could not have existed in the past the way that it does now, because the digital storage and processing power were not advanced enough to make it work. Now that multiple connected devices can crunch numbers in an instant and data storage is effectively limitless, the potential for new uses of machine learning in business is growing exponentially.