As the concepts of big data and operational automation have seen their profiles rise over the last years, so to has the concept of machine learning – the ability of automated systems to intelligently evaluate and manipulate data, and modify its own approach, based on the data and information that it encounters.
And as machine learning systems continue to evolve, they begin to show promise across a range of industries, from financial through utilities to healthcare, among many others. Machine learning systems give us the ability to rapidly gain insights and make adjustments to our approach in a much more efficient and timely manner than we’ve been able to achieve before.
Myths Behind Machine Learning
That said, sometimes the hype behind big data exceeds the reality of its use. While machine learning is a powerful tool, it’s not necessarily meant to be used in all scenarios, and like all technologies it works better in some use cases than others.
With that in mind, let’s explore some of the more pervasive machine learning myths in the market today:
1. Machine learning removes human bias completely
It’s true, machine learning does remove a certain level of bias. However, to suggest that bias is completely removed is simply not true. That’s because the initial algorithms, the data sets analyzed, and even the platforms chosen are all done by humans. And so, while it’s true that machine learning does remove much human bias, it doesn’t completely remove it.
2. Machine learning is real-time
This is actually a more common myth, which doesn’t make a lot of sense once you have even a rudimentary understanding of the topic. At a very high level, machine-learning technologies are meant to run a series of algorithms against data, and build useful models for organizations and systems to use based on the results of that analysis. It’s true, those models are meant to be used in real time, for the most part. But the learning and analysis that built those models is not.
3. Machine learning will produce results from any data, in any situation
This is also a common myth, the idea that any data can be introduced into the system, and the machine will automatically produce useful data, no matter what kind of data it is. Also, this simply isn’t true – an old saying in computer science is ‘garbage in, garbage out’, and that is just as true in machine-learning systems as anywhere. It is true that machine-based systems can often find patterns and insights that were previously hidden. But that doesn’t mean that one can simply feed garbage data into a system without any shaping or preparation, press ‘Go’, and have useful insight instantly present itself. It doesn’t work that way.
4. Machine learning is really only used for predictive analytics
This myth is based on some truth, and it’s certainly no secret that predictive analytics does present a category of use cases that are well suited to machine learning. But that’s a limited view of the uses for machine-learning systems. Machine-learning algorithms have a large range of uses in the classification, regression analysis, and clustering of data, all of which can provide useful information to the enterprise. Plus, these use cases are typically less complex and expensive to implement than predictive analytics use cases.
5. Machine learning only applies to Big Data
Again, another myth that has some basis in reality. It’s true that the rise of big data has given rise to machine-learning systems, and those systems have a unique applicability to analyzing large data sets and providing useful insights that really can’t be produced any other way, at least not in a timely fashion. But machine learning can be applied to traditional relational and structured data as well.
6. Machine learning is expensive
This myth certainly had some basis in reality at one point in time. As the use of machine-learning systems was developing, the practice relied on expertise and tools that were prohibitively expensive for many enterprises. But, as with any technology, over time the costs have dropped as it gains acceptance, and new tools enter the market. Likewise, machine-learning platforms at providers offered by providers like Microsoft now incorporate the ability to utilize machine-learning systems on a use basis, no longer requiring the same capital outlay as previous years.
As machine learning progresses into the market place, other myths will undoubtedly emerge; the usual vendor marketing and competitiveness will more or less ensure that. But machine-based learning systems are like any other technology – they have use cases that make sense, and some that don’t. And like any technology, as long as a balanced and realistic approach is taken, it can be an enormously powerful tool. As long as that kind of approach is taken, machine-learning approaches can produce value for the enterprise for years to come.