I don’t know if there is value in posting this, so I’m going to try and if this is not valuable, I won’t do something similar in the future.
I’m in Robert Monarch’s AI for Good course. In addition to the course, I have much experience with what is going on in the startup community in Silicon Valley, USA. As an engineer, people expect me to deliver the best solution to their problem. That’s why I am posting this question.
It seems as if engineers–or even worse, people above them in an organization–are looking at AI as the first solution to try for many problems. Is this the best solution? Is this because it is easier to do than using an old, non-AI solution? Is it perhaps better to try an AI solution and fail and then fall back on a non-AI solution? Is it the case that these problems cannot be solved without AI, so AI is the only possible successful tool, so we see a solvable problem where there would be none and therefore AI is the obvious best solution to try?
I’m trying to understand when using AI is the best solution
you are interested in an output (e.g. that there is a cat or not),
and you know that the output is generated through accepting a set of inputs (e.g. a photo),
and you have a large dataset of this inputs-output pairs (e.g. many cat and non-cat photos),
BUT you don’t know how actually the output is generated (e.g. you don’t know how to construct a set of rules to mimic human’s ability to identify a cat photo),
then, you want to build a machine learning model that can generate the output based on the inputs, so that you can make prediction with it?
Sometimes a non-AI or non-ML solution is easier to explain, maintain and has a shorter path to production. To me, it’s about understanding the problem at hand and doing an initial assessment on different dimensions and implications of building an AI/ML solution. Of course, if you have some experts to consult and/or some budget and resources to work on proof of concepts and prototyping, you can get some evidences to support the decision.