There are many situations where data is not available in the quantity desired, but you don’t need to wait for the ideal scenario to get started with AI. You can gain valuable insights from AI even when working with limited data.
We know that data is the fuel for AI. But how much data do we need to create an AI model? We often think that AI needs massive amounts of data to provide significant results. While this may be true in some cases, in many cases it is not. This article will discuss AI applications that can provide great value, even without having the ideal scenario of possessing all data available.
Data collection and gathering is difficult and expensive, and it is often the most time-consuming step in an AI project. Many contributing factors can lead to a lack of data. Financial or logistical constraints may be impeding your collection of data. In some cases, the data may be fragmented throughout many different spreadsheets, oftentimes in various formats that can be arduous and challenging to combine. Finally, in many cases, the data may simply not be available.
We may also have a situation where an attribute was not being collected when the AI model was created and, sometime later, this attribute became available. In this case, the AI model can start to be trained on the new attribute, along with the other attributes.
As you can see, there are many situations where data is not available in the quantity and the variety desired but, we do not need to wait for the ideal scenario to get started with AI. The awesome thing is that AI can still provide great value, even when working with limited data. Let us find out how.
How many trips do you need to make from your home to your work to know how long it takes? Not so many, right? Similarly, it is not always necessary to have years and years of data to train an AI model. Of course, AI models are not perfect and there will always be room for improvement. As the famous statistician George Box said: “All models are wrong, but some are useful”. In every project, it is necessary to evaluate how confident your models must be to be useful. Going back to our home-to-work time prediction, you may take longer to get to work on rainy days.
So, your prediction may be off on these days, and you may decide, down the road, to add precipitation as an attribute to your model. But let us stop here for a second. If you live in an area that only rains 30% of the days, you will still get an accurate prediction 70% of the time. I will leave it up to you to decide whether that is useful or not because the answer depends on each business situation. The point is that you do not need to wait for the ideal scenario to get some value out of AI.
Also, when you do need models that require intensive sampling to make an accurate prediction (such as trying to determine forest inventory), AI models can be trained on past inventory volume and on actual volume, along with other unit attributes, to “learn” the average inventory-actual variance for each unit condition.