In recent years, companies like Google, Facebook and Netflix have been in the headlines for developing groundbreaking machine learning algorithms. While this is fascinating, it can create the illusion, that only these huge multinationals can provide viable machine learning solutions. It is feasible to use data science to have a tangible impact on a much smaller scale. So why isn’t everyone doing it?
There are a number of, what we like to call, fears of adoption:
- the cost
- lack of data and data privacy concerns
- inaccuracy and uncertainty of algorithms
- displacing human jobs
Let’s start dispelling four of these myths.
Myth 1: the cost
Yes, incredibly complex and computationally intensive methods will cost substantial amounts. But, a relatively simple solution tailored to a specific task, a part of a business process, for example, can be developed quickly, with open source tools and deployed in the cloud for attractive prices. We regularly work on proofs of value for our customers, a short-term project meant to analyze a business case, without the typical overhead of a full solution.
Myth 2: lack of data and data privacy concerns
While data is essential for machine learning, there are ways to adapt to environments with insufficient or incorrect data, but that’s a topic for another time. On the subject of data, there are often concerns about data privacy. I think it’s important to remember that not all machine learning algorithms gather and use personal data – many times the data can remain entirely within the business, and is fully anonymized. If this is not the case, there are ways to ensure the data doesn’t land in the wrong hands, and that personal information is removed.
Myth 3: innaccuracy and uncertainty of algorithms
Another frequent concern is that these algorithms are very black-box (meaning we don’t fully understand why or how they are making a decision) and that they are not always correct. This is a more difficult concern to explain because it depends on the algorithm. There are many methods which are actually white-box, and can be analyzed by a human; some of them even have a guaranteed minimal accuracy. For the few algorithms which are black-box, there are ways to test them rigorously, and while they might not achieve perfection, they can be as good or better than humans. In order to keep an eye on a data science solution, it is possible to develop live dashboards, which display the health and performance of the algorithms in real-time.
Myth 4: displacing human jobs
Lastly, and maybe most importantly, let us briefly discuss the impact of data science solutions on the human workforce. There exists a popular misconception, that in the near future technology will replace many jobs. This does not have to be the case, as it is possible to design solutions so that they contribute to the work done by employees, and simplify or strengthen the tasks carried out by them. For example, a business process can be supported by a prediction of an algorithm, tying right into the existing system or infrastructure, displacing no one, and harnessing the power of machine learning in a constructive way.
I hope that this video gave you a quick insight into common fears when adopting data science solutions. Make sure to check out our other videos about machine learning and artificial intelligence.