What are the problems in building dialogue systems?

Question

Machine learning

I read a bunch of articles about using neural networks to build intelligent task-oriented dialog systems (seq2seq, attention model, hierarchical RNNs, using RL - in short, all relevant articles from the last two years). Those. things when the phone does not say "press button 1 if ...", but there is a voice dialogue with the bot. And the question arose: why didn’t it take off before? Was the problem in the accuracy of phrase recognition through semantic analysis? In the accuracy of prepared phrases-answers? Accurately tracking the state of the dialog? Is there any infa somewhere about failed or successful implementation attempts with a description of technologies or resulting indicators?

Answer 1

Read about the philosophical law of the transition of quantity into quality.
There just became enough research on this topic, programmers, businessmen pulled themselves up, productivity increased - as a result, something could be done. Well, they did.
PS. If you would have done it 3 years earlier, then 3 years ago you would have asked this question. And they would get the same answer.