October 27, 2020
As Neurisium is creating machine learning solutions, among other products meant for the manufacturing industry, we’ve seen varied approaches to onboarding new technologies. The most surprising potential customer queries of the past 3 years have been the ones, where machine learning is requested without considering the requirements or internal data readiness. This begs the question, if machine learning is overhyped and why so?
Hype vs Reality
It’s understandable that each year brings about new technologies that are more popular than others, such as Big Data, Cybersecurity and Cloud computing have done recently. Machine learning is definitely one of those elusive topics that allows for people to dream about all its future possibilities. Often though, there is a disconnect between the dreams of self-learning robots that have the potential to take over the world and the mathematical and statistical supervised learning models deployed in machine learning today.
This dream of the future is also driving the hype, motivating companies to invest in machine learning for its future possibilities, overlooking the readiness or need for such solutions today. It’s becoming more predominant that machine learning or even AI get thrown around as automation solutions to all kinds of business processes, without considering the latter:
· Data readiness: getting clean, stable and usable data, creating the data governance processes and scalable data structures before any machine learning development begins.
· Data strategy: implementing a long term data based plan instead of approaching each application build case by case. This will eliminate duplication in work assignments and create a common data architecture.
· Solution scalability: many machine learning projects are done on prototype level and scaled without considering the system-level design or reusability of the code.
Setting Realistic Expectations
The most important thing is understanding the business pain point that is being solved and putting data first as a prerequisite to any machine learning developments. Those are also the reasons why Neurisium is moving more towards analysing and cleaning data in business processes to quantify the value of deploying machine learning, instead of marketing machine learning per se.
Establishing a solid foundation for deploying machine learning is the key to successful project execution. This does not only mean data but also a change in organisational culture and processes. On boarding any new technologies requires time for adoption by the employees and machine learning is no exception.
The context in which machine learning is used should be considered as well. Currently statistical methods enable to easily automate processes or generate overviews based on large datasets, while transparency and reasoning of machine learning actions is still gathering momentum. Therefore, we are far from machine learning or AI taking over the world. What it can do, is allow employees to focus on more interesting assignments and speed up process execution.
The Bubble is About to Burst
We are getting to a point, where most companies have dabbled with machine learning and are beginning to have more realistic expectations about its application. For example, the NewVantage Partners 2020 survey showed that although 91,5% of companies continued investment in AI, the percentage of respondents naming AI as the most disruptive technology declined from 80% in 2019 to 69,5% in 2020.
There are many amazing things that can be done with machine learning, such as predictive maintenance, production parameter adjustment and visual quality control. We just need to take the time to set an achievable long-term goal and focus on developing our organisational cultures, data strategies and infrastructure. This way, we can be ready to be on the forefront of the new machine learning and AI possibilities to come.
