What is the meaning of neural networks / machine learning?

Question

Neural networks

Hello!
Take for example this article:
https://geektimes.ru/post/291789/
"AI was taught to recognize sarcasm on Twitter by emoticons"
Why use a neural network for this case? Why not analyze a certain number of messages with emoticons and draw conclusions.
For example, we learned that smiley [arbitrary smiley] is used in 80% of cases in messages with sarcasm. And based on this probability, it is now possible to decide by smiley whether the message is sarcasm. That is, in this case, if it contains a smiley [arbitrary smiley], then 80% that the message in question is
sarcasm.
And so for each smile. Since for the neural network it is also necessary to manually label the train / test datasets, this will not increase the amount of work.
So what is the point of neural networks / ML, if so many problems can be solved simply with the help of statistics?

Answer 1

That's just the point that your decision does not give any guarantees. From the word in general.
Statistics do not give answers on a specific case.
Statistics give answers to a large number of cases - like that 10% is blue, and 90% is green. But whether a particular object is blue or green, your method doesn't answer.
Your method is good only as a helper, as a preliminary filter for human moderators.
And network training gives at least some semblance of human analysis in this case.
The network is just an attempt to replace a living person when you cannot pick out iron patterns and program these patterns.
You feed the network an example of what has already been verified by a person, and you get a semblance of a live moderator, and so on.