I looked at the XAI concept in the link. This is a small evolutionary improvement in the overall AI system. The ‘Training Data’ and ‘New Machine Learning Process’ is still the basic process that you have today. When new (and old) concepts are studied by humans a website is usually established, and this knowledge is written down and explained. An AI computer can now access this knowledge via the internet and can copy and catalog this information in an index for future reference.
Currently when I use AI, I have a concept that I’m trying to implement but need some information so I can better design a device to execute my ideas. When I ask AI the proper questions then it presents me with the data, I need to refine my ideas and allow me to better design a device with my specific parameters so it will work as desired. Usually, I will have to study this information, make some calculations and/or software changes, and then try out these experimental ideas to see if my objectives are obtained. Generally, I’ll be much nearer to my overall objective but will still need to ask AI some other questions to fix some minor discrepancies and then repeat the process over again.
The new XAI model may improve this process loop by using the data it has available while incorporating my specific objectives to then apply this data and simulate the changes to the design to see if it matches the desired outcome. If not, then it could automatically try some refinements while telling the ‘User’ (me) that here’s another idea, is this a success, or not?
Nothing wrong with this whole process, but the XAI system is still basically an automated ‘data miner’ and is using the knowledge that was generated by humans and then placed on the internet.
I have been retired for several years now but still am actively working on my own hobby projects just for fun. If you want to see what AI can do for you just go out a purchase an Arduino Giga R1 computer board and then see if AI can program it for you. Sure, you can get it to program the basic ‘Hello World’ program, but all that data is readily available on the internet in the Arduino site. Give it a more difficult program to produce and see what happens. Then you see where the current technology is.
That’s not to say that eventually AI will be better than the average human and a factory won’t be occupied mostly by robot workers, but that’s a long way off. I would say that AI would be similar to a baby on the day its born into this world. The baby’s firmware is operational, and it knows how to breath and cry and how to feed off its mother, but little else. There will be daily improvements as it observes and learns how to interact with the environment, but it will take some time before it could expect to succeed on its own if it left home. A robot today doesn’t know how to explore for minerals needed for a microprocessor or a battery. How to build a power plant to power all the mining machines. How to transport, refine, design, build, ect, ect. Someday that may happen, but when it does most of the knowledge required will have come from humans over the centuries.