THE GAP
I was listening to a podcast called “AI TODAY” by Cognilytica the other day and their topic about this blog’s title was quite intriguing, so I want to share it here.
According to TechRepublic, 85% of AI projects eventually fail. One of the major reasons for that is due to lack of understanding of the difference between traditional software development and AI projects. Thinking AI projects too much as the functionality like normal applications.
What makes an AI function actually has nothing to do with the code itself. For example, if Amazon is recommending you something that you are interested in, it’s not the code itself making the suggestions. It’s the data used to create the model that’s making the suggestions.
THE APPROACH
The agile methodology has become the mainstream for software development because the waterfall methodology is hard to adjust when it comes to huge projects which can be inconsistent and unpredictable. But when we try to use the agile methodology for AI Projects we run into a problem.
For traditional software development, the agile methodology focuses on iterating over the functionality of the application. Maybe discussing, what features were added or revised since the last iteration. But this functionality-centric iteration does not help you consider the most crucial part of an AI project, which is data.
Therefore, when it comes to an AI project, you also need to consider the data-centric iteration, which runs together with the functionality-centric iteration but in a different timeline.
I’ve created a diagram with an example of an agile project creating a chatbot. If you only consider the functionality-centric iteration, it seems like there is only one iteration, but as you can see, if you also think about the data-centric iteration, you’ll find out having a lot more things to consider.
Thinking about this ahead can help you not only prevent setbacks but also recognize the feasibility of the project with the current data resource you have at an early stage.
It is vital for us and the client to keep this in mind.
