Reply To: Article 11/13 for the EU and future of the site

[tarrox 249xp 1]

@khusrau Problem here is that you and the Politicians (also managers etc) are overestimating the abilities of AI. Yes certain things are simple and more doable but others aren’t and some never will be. Currently we are still exploring what is doable. Basically we are like in the people in the 19 hundreds imagining what the year 2000 would be like.

For example last year Sony had to remove their copyright claim for one of their Bach performances, as it resulted in nobody being able to upload their rendition of this Bach piece, without it being copyright claimed (which is more than legal as Sony only holds the right for this specific performance not the music itself). Problem was that there was no way to change the filter to make this work without making the filter useless (It would have been very easy to trick it and thus making it useless and not what copyright holders want).

Second the application of AI filters is really fast but creating them can take weeks or even months. For example googles voice recognition needs to train for several months to improve itself. That is the reason why voice recognition is advancing so slowly as it takes a lot of time to create the models. And a lot of it is so fast due to improvement of computer technology, in other words it is fast but to get that fast you need big computing rigs.

Third, one of the biggest issues here is that no AI is capable of recognizing parody, satire and the like, even the research is just now starting to check if this is even possible. And even if it is theoretically possible, the complexity of doing this task could be so costly computing wise that it is not practically viable. For example a correct recognition would require to include picture recognition, voice recognition, recognition of the pitch of the voice of each word, a understanding of the context (Oh this will be fun believe me) and combing all of it together. I myself work on a project that worked with logic problems, that even a 6 year old could solve but the moment the machine tried it, it needed way more time than a human and we are talking about laughable problem sizes, we were not even close to something even remotely practical (Even though we improved a lot of the stuff by factors of 1000 and more).

In addition just last month a bunch of scientist proofed that there are problems that AI’s are incapable of solving to the point it is even impossible for them to determine that they fail at the task. So keep that in mind when saying AI will solve the problem, without knowing if the given problem was already solved by an AI.

To sum it up, the current approach of AI is, that AI is a hammer and we try every problem as though it is a nail. A lot of the times it works (especially as we use AI on the problems we know it can work), but now that we try it on things that aren’t nails we will see it fail. Often we already know it will fail or is a hard to do but to know this you need to look into the research work, as that is where failed stuff is written about due to the fact that companies not writing about their failed attempts so the stuff you see is heavily biased towards success. For example every few moons you see someone try to create a medical AI to tell you what is wrong and it never works as good as you need it to.