When NOT to use Artificial Intelligence or Machine Learning – Today lets talk about when NOT to use AI or ML in your products, First, I know, these 2 are the buzzwords right now and make your products look sexy and… while every other startup in 21st century uses these terms while explaining their next innovation, is it actually required in every scenario? Well, in order to know when not to use ML or AI, let’s first see when you should actually use them:
When we want to find a useful pattern
Yes! if you need to discover a pattern between dependent variables and an independent variables and if (that’s a big if!) we could find it, it would not just make our product better, but also, could help us explore a more brighter & newer side of the program we are building.
When we want to generalize the established pattern
If you are someone who’s working in the data science industry, you’ll soon realise that its not enough to establish a pattern in our data. To say the truth, its quite easy to establish a pattern. The challenge comes when you want to make your establish pattern into a generalised one.
Well, a generalized pattern is the one which stay relevant even when certain scenarios or conditions are over… and its quite hard to confess that all patterns are not generalised in nature when it comes to machine learning. As a reminder, I would like to reinforce the fact that “Finding patterns and using them is what machine learning or AI is all about.” and it is our job to look for better patterns that help us move forward and make progress with the available data.
Now, I believe, you are having some background on when machine learning or AI is generally employed. Next, lets take up scenarios when we should not use AI or machine learning,
When there’s no pattern to connect input with output.
Now you might think, is there any such thing like this? Well, yes, one such example would be to calculate how human brain develops a feeling called love… Ahh! I know what faces you are making right now… but you cannot correlate factors and tell a machine why humans love their pets or sibling or someone they just met a workplace.
When you lack “Good” data
We all know that machine learning algorithms require large amounts of data before they even begin to give useful results. The larger the architecture, the more data is required to produce viable results. And while you might try to pull in every data you find relevant on the web, not all might be of use to you, or can be applied to your business use case & can even cause your algorithm to perform poorly.
Also, do note that Reusing data is also a bad idea, and data augmentation is useful to some extent only. Though having more data is always the preferred solution, having “good” is something that should be a priority. By “good” here, I mean data which directly affect the end result of your model.
When we are dealing with Nonstationary universes
This translates to a situation when all the data that you have isn’t relevant anymore — perhaps because a pandemic like COVID-19 changed the rules — it won’t matter how good is the information that you have, as it is all irrelevant now. If now you want to build a machine learning model, you need to prepare data as per the new rules.
Simply because if your past data all of a sudden don’t apply at all the nonstationary future you’re having, you’re not allowed to use yesterday to predict tomorrow with a straight face. With that, I hope this video was helpful to you and served value.
Web enthusiast. Thinker. Evil coffeeaholic. Food specialist. Reader. Twitter fanatic. Music maven. AI and Machine Learning!