I work in image processing and I have seen demos of AI doing amazing things in this field. For the sake of job security I would like to learn how to make AI do useful image processing things, but for any real world problem I have encountered it is so much more straightforward to do it with traditional methods.
That is not to say that traditional methods are superior to AI, but AI is AI is less accessible to a regular individual like myself who just wants to solve some specific image processing problem. Take image recognition for example. I'm not going to gather a bunch of data and train it for this purpose. I could try to find an already trained model that does a similar type of image recognition as me, but then I have to learn how to interface with it, and it might not recognize exactly the kind of object I want it to recognize, and now I can't modify/debug it because a trained model is basically a black box. The traditional method of thinking of mathematical features the object has and filtering based on these is much easier to get it to do a reasonably good job than AI methods.
With traditional image processing methods, I have a toolkit in my mind of strategies I can use, which over the years I add to. But AI is basically just one big blob which either "does it all for you", or it doesn't work, there is no granularity.
So I am stuck. On one hand I see people giving amazing demos of AI image processing and I want to be able to do the same thing myself. But AI solutions just wouldn't be practical for someone like me. So I am stuck producing work that looks like it is from the stone age compared to what AI can do. And then I worry that there will come a point that AI can "do it all for me" and everything I worked on will become obsolete, and since AI models are a black box there is nothing I can do as far as trying to improve it, or learning from it, so my job becomes just to deploy AI and that's it.