Deep learning and AI ramblings

 SF looks and movies about making a digital uploaded copy of a human or about good/evil AI often assume a single instance copy is some sort of universal rule… why? I assume it mostly has to do with it being impractical to tell a story where the main character suddenly spins of into billions of copies …. it would be quite hard to follow the story after this. However, there is no such universal rule - the low cost of producing copies once a DL model is trained and sufficiently good for the task is key to the success, No need to educate people for the task and, more importantly no need to educate new people when. your needs increase or current staff leaves. I can also spread to all areas where it is useful - a model for detecting intruding “killer snails” is quite costly to develop but once it is done it can be used wherever there is a fear of them enterings and gardens… (I really need a better example here :)

AI hypes have come and gone. My first was during the late 80s and since then I have been very sceptical… hower a few years back I saw an impressive demo of an image recognition Deep Learning implementation that was a bit of an eye opener. I also happened to read this by now old but excellent blog  https://waitbutwhy.com/2015/01/artificial-intelligence-revolution-1.html   since then I have done a fair share of reading and have also gotten a bit of real life experience of developing DL tools, most for video analytics. So, in this blog I thought I could gather some of my thoughts …