As our society becomes more and more digital we produce exponentially more data. In order to remain competitive businesses need to leverage the data that they have available to them to make plans for the future.
Many methods can be used to process this data in order to make predictions about the future using data analytics. This can be done via several methods such as traditional spreadsheets and other visualisation, Machine Learning, or Artificial Intelligence.
What Machine Learning Isn’t
While Machine Learning and Artificial Intelligence are sometimes used together, they are not generally accepted as the same thing. Artificial Intelligence or AI is used in an effort to have a computer take various inputs; be that from data, sensors or images etc. and assess the optimal action based on this information.
At present it is generally a human programmer training the computer to take an action which a human can do better than a machine. This may change in the future as technology and algorithms improve.
A few examples where AI is prevalent in your day-to-day life may be; a virtual assistant such as Siri or Alexa, spam filters, data analysis and customer segmentation.
What Machine Learning Is
Machine Learning or ML is used where a machine can learn on its own from the inputs without being explicitly programmed. There can be specific algorithms employed to do this, however, the results should be the results of the computer incrementally learning more from each input.
Examples of ML in your day-to-day life may be; Traffic Predictions for GPS navigation. Facial Recognition, Search Engines.
There are many places where ML is used to optimise AI solutions; virtual assistants for example will learn from failed attempts at following instructions in order to know how to better perform the command in the future. A sophisticated spam filter will incrementally learn from how people interact with their emails and improve on how it treats future emails for both that person and other people it has learnt have similar tastes.
How You Utilise Machine Learning In Your Business
One common example would be financial forecasting. With many business inputs and outputs, it can be difficult to determine how each affects the end results. While a human can get very good at predicting the end results. They may disregard or under-weight an effect which has had more of an effect than thought, or sometimes worse, the opposite of over-weighting an effect.
In healthcare, ML is being used to diagnose skin cancer. A computer receives data in the form of images and tries to predict if it is of a cancerous tumour or benign. At the start it will be a 50/50 guess, it will then be told if it is true or not and will eventually learn markers that are common to all positive or negative results. Eventually, after millions of iterations machines are now being reported to have diagnosed tumours that doctors have originally felt were harmless.
ML is being used by many firms for dynamic pricing to maximise profits by recommending or automating price changes. The results will be measured to maximise a set output and the pricing will be optimised each time to improve future pricing decisions.
Key Types Of Machine Learning
Supervised
Supervised Learning is the most commonly used. It’s task-driven; its goal is to take known values, and determine an algorithm in order to predict the next value. This can be used for things such as predicting prices, financial results and image classification.
Unsupervised
Unsupervised Learning is a little bit more complex as it is not labelled in a human way. One of the most common uses is in clustering. It will take the information given, often categorical and try and classify them by the various parameters. This can be used for things like fraud protection, gene clustering and digital marketing by automatically clustering customers and delivering advertising to similar audiences to your current customers.
Reinforcement
I like to think of Reinforcement Learning as training a tiny digital toddler, hopefully slightly faster. The basic concept is programming a positive reward into the computer and allowing it to try and accomplish this reward through several iterations. One example is in robotics where a robot is programmed to stand up; Each time it tries to stand it will try a random method, and if it fails, it learns that those exact steps are not correct, if it gets a bit closer, it learns a bit more until it eventually succeeds.
The same principles are used to develop autonomous cars and how chat-bots. It is also how Google taught a computer to play the game AlphaGo, a strategy game that was historically thought to involve too much instinct for a computer to truly learn.
Another application of reinforced learning is in encryption, one very fun example of this is how Google used 3 computers named Alice, Bob and Eve to compete against each other. Alice’s job was to send a secret message to Bob. Bob’s job was to decode the message. Eve’s job was to intercept the message and decode it. Bob and Alice both started with a simple code cypher or key, Alice had to convert the text into a jumble and send it to Bob who would convert it using the given key. While this works well, to begin with, Eve quickly learns the key and begins to win. So Alice started to change the encryption and Bob had to keep trying to interpret it. This continued over thousands of iterations where at first Eve would occasionally win. Then after a string of successes in a row Alice and Bob suddenly leaped forward and started to create incredibly complex algorithms which they were able to both interpret but Eve would fail. After 15,000 iterations humans cannot understand how the encryption works but it is highly effective and has countless real-world applications.
Should My Business Be Using ML?
In a lot of cases, an ML application will help with the efficiency of doing business as well as optimising sales outcomes. For example, chat-bots can drastically reduce workload as well as improve customer experience. An automated marketing system can be used to A/B test digital campaigns to decide on the most profitable option. If your business would like to dynamically price products and services ML can be used to do this very effectively. ML can be used to serve product recommendations to customers like when Amazon recommends a complimentary product to increase each cart’s value.
All of these examples and many more can be implemented using out-of-the-box services that can be plugged into your existing platforms with little to no technical knowledge. I hope after this article you will know a few examples of products that will be useful for your particular business.
If you have any questions please comment below or contact me at paul@paulclarke.com.au and I would be happy to point you in a direction to start.
Leave a Reply