In last week’s blog, I discussed 3 advantages to using AI in reliability. This week, I’ll give you 3 mistakes that people can make.
- Start with an Outcome in Mind – What problem are you trying to solve? Some companies import all their data into a platform and hope to find value by brute force. Start with a small problem that you understand, define what data is required to analyze that problem and review the output of the model.
- AI Makes Correlations Not Causations – Similarly to the 1st drawback, and why I recommend to start with a problem that’s well understood, AI does not know causation. It looks at correlations and draws conclusions (which can be correct or incorrect). If you imported all the data in the world, you may find things that correlate which are irrelevant. For many examples, check out this fun website called Spurious Correlations.
- It’s Only Useful on Data That It’s Seen Before – Similarly to people, although we have logic, AI is best at identifying problems that it’s seen before. If you’re using AI to detect failures but you’ve only trained the model on good operating parameters, it’s not going to be good at distinguishing different failure modes when they arise.
I would definitely recommend playing with AI as I do think it will be the future in our industry, however, I’d keep these drawbacks in mind!
Reliability Never Sleeps,