New powerful tools are flooding the analytical market. Now is easier than ever to run an analysis on a specific data set, or apply a new methodology. Is the easy access to the analytical tools a danger to reliability practitioners?
During an informal talk in the last reliability course that I attended, we dived into the discussion that some tools are way too complex to use in our daily work life. Some simplifications are more than welcome for the practitioner. One of the instructors remarked the dangers to over simplified tools and processes. The logic behind being that people might start to execute analysis without the background, knowledge or critical thinking necessary.
I feel divided in this topic. I can understand both arguments and see their nuances, although I tend to think that a lower entry barrier to the whole reliability world would be beneficial. A wider pool of practitioners too. Check my post “3 Things to Improve Reliability Knowledge Transfer“ for more about this topic.
First, is clear that a deep understanding about the underlying assumptions of each model is an advantage. The do’s and don’ts of specific methodologies. Specially on those application that are mission critical.
For example knowing the underlying assumption that to do Life Data analysis (i.e. How to do a Weibull analysis) the model assumes that the underlying data points (failures) are independent and identically distributed (iid). This point is quite important to challenge and interpret the results. Yes, there are ways to cope with the violation of some assumptions, and to know that you need to know “what you are doing”.
On the other hand, we are seeing more and more technologies in the market, like cloud computing, that are making easier to do complex tasks. For example having access to infrastructure that some years ago was only possible for big companies. They are offering this out-of-the-box services to everyone. Are they wrong in their strategy?
The more tools that they put out there, the more chances of people misusing them. On the specific area of data analysis and modeling, there are options like Microsoft Azure Machine Learning module with a point-and-click interface to build complex workflows and models. On the side of Google, they have the TensorFlow product. This last one is not as much as point-and-click as Azure ML, but time to time. I’m sure they are working to have an environments as simplify as possible.
How others solved this problem?
In addition, this conversation has been going back and forth between the field of statistics and the new Data Science field for some time. More people “practicing statistics” without a solid background might be dangerous.
As a reference, in the podcast PolizyViz.com- Episode #69, Hadley W. -a well respected data science practitioner, developer and general forward thinker- argues that the question has been answer. He argues that that even that there will be more mistakes and dangers, the more practitioners the better. Nonetheless, there will be more than ever the need for skill professionals to correct those learning by doing.
Can we apply the new strategies to the reliability engineering sector? Is it too different? The risk outweighs all the possible benefits?
In brief, the reliability engineering world is not the first sector to be challenge with this question. Maybe will be a good idea to learn from those who has faced a similar situation, and move on. Mistakes will be made along the way, and that is why, traditional professionals will be always in need.