@kel This is so sad! People who think that AI can make music are people who never knew what music was in the first place (same goes for the visual arts, for science, and anything that involves true creativity).
I think this has been in the making for quite a while. The societies we live in mostly fail to teach people to really *see* what is special about human output, to ask questions and seek the unexpected.
I'm sorry for you. Please keep making music. Music needs you!
@minimalparts @kel Same as the above musician. I have just closed my translation business after a 40-year career. The technical/patent-related work I did has gone. It's good that autonation can help more people communicate but companies have gone overboard on the hype. Some translation will always need humans, but as with coders, how are they to get trained & experienced?
@annehargreaves @kel A close friend of mine is a translator, so we often have long conversations about 'how you would translate X or Y'. When I see how much thought, argumentation, linguistic and world knowledge goes into those conversations, it is very clear to me that a mere statistical model of language will never do what a translator does. It is so important that people do not forget this. One thing I am most worried about is that we are experiencing a devaluation of language itself.
@minimalparts @annehargreaves @kel
if
(( we care about human language )
&&
( hyping AI makes you rich ))
then
we need to change the system;
@minimalparts @kel Translation automation is positive up to a point - it would be impossible to provide human translation for all the films, clips, social media posts, software etc. so here it aids communication. The hype has meant that people now think *any* translation can be done by AI, which is not the case. Just 1 example - Scots law into the language of any other jurisdiction. Not possible & never will be. There will never be a big enough corpus for training just for a start.
@minimalparts @annehargreaves I studied computational linguistics under Winograd in the early 80s and when Google decided to replace its previous automated translation system with an ML system in 2010 I decided to investigate it. Naively put: ML based systems are good (enough) at translating non-idiomatic factual statements between languages that have reasonably similar grammars and sufficient corpus for training. The farther you get from that the worse they do.