When is ML/ AI an appropriate solution?

Original article was published by Sandi Besen on Artificial Intelligence on Medium

Case 1:

Your credit card company needs help determining whether a charge is valid or fraudulent.

Credit card companies lend to millions of people most of whom charge daily. It probably wouldn’t be efficient or timely for someone to manually check if the charge is valid. You don’t want your credit card company letting you know something looked fishy last month…These events are time sensitive and require immediate action.

Additionally, let’s think through the pieces of information that we need to know in order to determine if the charge in question is valid. We might want to know the customers spending frequency, average charge amount per transaction, typical location of charges, or any other variable that could help us predict whether this transaction is valid. This seems complicated! It sounds like ML will be able to solve this problem timely and efficiently.

Case 2:

You need help budgeting discretionary income or allowance.

So you want to figure out how much shopping budget you have this month? Some excel spreadsheets and simple formulas could be very helpful, but you are working with small amounts of data and the answer is not extremely time sensitive — this doesn’t look like an ML appropriate problem.

Case 3:

Your smart watch is streaming the progress of your current workout to determine how much longer until you burn 100 calories.

Streaming data sounds like there is a lot of data to be analysed! If certain metrics (calories burned thus far, energy output, current workout time, etc.) are recorded almost constantly, and you are using that data as inputs to predict how much longer until a certain calorie count is reached…. It sounds like you have something ML can help you with!

Case 4:

You would like to predict which country will have, on average, the happiest people next year.

Based on historical data, you would like to predict a future outcome. Prediction is a traditional use case for ML. For this problem you would use various features such as life span, % of improvised population, etc as inputs to appropriately predict an output. In fact, in industry we call this topic predictive analytics and there is a whole field centred around it!

In Summary

ML / AI is a helpful tool for complex problems that involve tons of data and/or require time sensitive results. However, not every problem is an ML problem! If the task can be completed timely and efficiently without the use of ML — do it. When ML is not the best fit for the problem, it can lead to higher budgets with more use of technical resources, and greater amounts of labour. The next time you are not sure if ML is an appropriate solution, review the heuristics!