How the Machine Learning of Today is Driving the Artificial Intelligence of Tomorrow
Machine learning is quietly spreading into every aspect of our lives. Perhaps it is not quite yet completely common, but it is definitely becoming more noticeable. From recommendations by online retailers to better, less congested routes to familiar destinations, it is starting to make our lives easier. At the same time, of course, we are starting to rely on it more and more to help us organise our complicated lives, and live smarter.
From products to services
One really big change has been from products to services. Companies used to sell us products. The relationship was a one-off, transactional arrangement—we bought, they sold. End of story. But now that has changed.
If you go online to a retailer you have used before, it will remember you, and make recommendations for possible purchases. The arrangement has changed from a one-off transaction to an ongoing relationship, and what you are now paying for is the convenience of a personal shopper, and not just the goods that you order.
Your watch is no longer just a watch. It may also monitor your vital signs, remind you when you need to exercise, and even suggest a workout. It is your personal health and well-being guru.
In other words, the digital age has blurred the line between product and service. The use of data and analytics, and particularly machine learning algorithms, have made it possible for new businesses to disrupt traditional and long-established business models, and even turn the world on its head. While IT companies have led this transformation, there is plenty of opportunity for other sectors to follow.
Machine learning comes of age
Machine learning is not new. But it has really come into its own with the rise of big data, because algorithms need huge amounts of data in order to learn effectively. The increase in data has also been coupled with a massive increase in processing power: the perfect storm for machine learning. Techniques such as neural networks and non-linear support vector machines are able to model much more complex relationships with increasing accuracy. Suddenly, it is possible to produce models that actually behave like real life.
Of course there are disadvantages to generating complex models. The main one is the amount of data needed, and the time that the algorithm takes to learn. Interpreting the model is also more challenging, and needs more operator skill.
But for many companies, the benefits far outweigh the challenges. Machine learning is now being used across sectors and disciplines, from financial services through public sector and healthcare to the oil and gas industry. Wherever there are complex challenges, and multiple sources of data, there is potential for machine learning to help. It is particularly useful when past events can provide useful predictions for future problems.
Busting the myths
There are a number of myths about machine learning. It is not a magic bullet. It does not solve all the challenges or make decision-making entirely rational. In particular, it does not eliminate bias altogether. It can remove some forms of bias—confirmation bias, in particular, because machines do not focus only on information that confirms their previously-held views. There are still a number of potential sources of bias, however. These include selection bias, from choosing the use case, the data sets, and the machine learning methods, and bias in interpretation of the models.
Machine learning also does not happen in real time. Machine learning models can be used to give answers in real time. But the learning happens offline, from past data, and may take a considerable period of time. Only when the model has done its learning is it ready to use—and by then the data may be out-of-date. Models are only as good as the last set of data, which is why they should be updated regularly.
It is also a myth that one model fits all. Just as no single employee could do everything in a business, so one model cannot meet all needs. Businesses need lot of models for different purposes.
The human touch
Digital transformation efforts often start with technology. But technology is only a tool, not an end in itself. Transformation needs to start with the customer. Machine learning is useful in this process, because it enables us to model some of the complexity of human behaviour with much more accuracy than has previously been possible. If you have the right question, and the right data, then machine learning can give you the answer.Download an IIA paper: Machine Humanity: How the Machine Learning of Today is Driving the Artificial Intelligence of Tomorrow