AI seems to be very smart. It knows the answer to almost all questions, but the way how the AI does that thing is different to humans. That means the AI cannot estimate the information the same way as humans. The AI uses database connections to create answers. There are two ways to make that thing.
1) The AI can use internal fixed databases. In that model, all data that AI shares are stored in the computer's hard disks that run the AI. And in that case, the AI uses the data that is involved it's databases. Making that kind of system output the grab is easy. The hacker must just change the table and then the system will just give false answers. Or if nobody didn't include some kind of data like accounting terms in the system. The system can't give the right answer.
2) Or it can use the open internet. In that model, the AI uses the web browse,r that it can get information from homepages. Then it simply connects as an example information that it gets from the first four homepages. The simplest way is to connect paragraphs from those homepages by choosing the first paragraph from the first homepage, the second paragraph from the second homepage, etc.
The AI can also have parameters. That it searches certain keywords from those home pages. And in that model, the AI counts how many times that keyword repeats on the homepage. That kind of AI might have forms where users can put the keywords.
This makes it quite easy to cheat the AI. The cheater can use the homepage that involves some keywords thousands of times. And there is nothing else. If the AI is simply calculating those words it can use that kind of homepages, and in results is only one word. Of course, the user of the AI must search for this type of information. But it's theoretically very easy to cheat the AI by creating a homepage, that involves as an example word "fish". But this homepage is useless if a person asks for information about things like cars.
The AI requires an error-detection algorithm.
But the AI doesn't know what the data that it collects mean. This is the problem with the AI. The AI must control the information that it gets. And there must be methods that are eliminating the possibility to make the AI confirm that there are empty pages, that it uses for creating, the answer. The answer for making a solution that eliminates joke pages is the dictionary book is stored in the AI's memory. And the AI searches the source by using the dictionary book that confirms that there is text on that page. If the page seems to be empty. That algorithm can start to search for the next page.
It recognizes the database and home pages. But it will not know what is included in those systems. The thing is that the AI knows that there are certain keywords. But it doesn't recognize words. And that means all characters on keyboards are letters to the AI. So the algorithm can control the possibility that there is some kind of cheating page by searching other words.
Comments
Post a Comment