Why AI has problems with speech recognition? The major problem with AI is that. Normally, people are not speaking the literary language. That means the AI must translate dialects to literal language and then find the database. That connected with those words. The problem with AI is it can use fuzzy logic. But in its internal functions, the AI uses certain logic.
The easiest way to make the fuzzy logic system is to make many routes from things, like speech recognition software and certain databases. In that model. Where the system created so many dialect versions of one literary word. And then, all of those options are connected to the word. That activates the database. This kind of structure is technically very easy to make.
But it's hard to create a connection between every single dialect word. And the literary word connected with those dialect words should activate the database. For being smooth and effective, that the person can speak freely that kind of solution requires very large-size databases.
"In a test of ChatGPT’s ability to handle accounting assessments it still couldn’t compete with the student’s level. Credit: Nate Edwards/BYU Photo" (ScitechDaily/Humans Reign Supreme: ChatGPT Falls Short on Accounting Exams)
The problem is that the AI takes orders by using a speech-to-text application first, and then that application just drives those words. That translated to text to the database. The problem is that the texts that the application generated must be precisely similar to the name that the system uses to recognize the right database. If the system cannot recognize the word. It cannot connect it to the database.
Why ChatGPT fails in the accountant exams? The thing is that many people are acting as accountants. But the terms of that work are not often searched on the Internet. When the AI makes things like some exams it requires data that it uses in those exams. The data that ChatGPT uses must be found on the Internet. Or it must program in a fixed database that the ChatGPT uses internally.
The thing is that people are not very often search terms that accountants use. People search more often for things, like "What is the biggest lake in Africa?" than terms connected with debit credit accounting. That is the weakness of AI.
It's possible that the people who make fixed databases are not even making databases about things, like debit credit accounting. The AI is always very clever when it computes common things. But, when it must search for something unusual. That means the AI will get in trouble. If the AI uses net search, there must be some kind of variables that the system uses and the thing is that the AI doesn't know what kind of data is in the web pages that it uses.
The AI can recognize the similarity between the word that is involved in the search parameters and words, in the web pages. But the AI doesn't know what those words mean. And in that case, there is the possibility that the AI makes the solution by using the homepages of some accountant offices.
Comments
Post a Comment