AI needs better management and understanding, says new research.
We know that the AI is not ready. And another thing is that AI doesn't think. It's a language model that can connect data from different sources. And if we want to make applications using AI we must realize that in the worst case, the AI simply uses preloaded code lists. Then the AI simply "kills" unnecessary code.
If the hacker can get the code tables that creates vulnerabilities in the application because the program uses static filepaths, routes, and folder addresses for its database. It's possible that hackers simply order a couple of applications from the AI, and if the database paths and file locations are the same, they can get those locations and filenames in their hands.
That thing makes the system vulnerable to so-called SQL injection. In the worst case, the hacker involves a couple of new user IDs in the access databases. This thing means that the system should inform supervisors about the changed number of user IDs and changes in the access table's size.
The AI will not take anybody's work automatically. It makes work effective, but that requires deep knowledge about the language models, or at least well-made, well articulate queries. So the user must make the query using good language and precise determination. And that means we should practice at least grammar if we want to use AI effectively.
When people develop AI and other applications by using AI they must realize that those tools are more effective than ever before. That means there must be some kind of internal security module that denies that application to create computer viruses or some killer-drone operating systems.
The fact is that laws are not an effective way to limit advances. In worst cases, laws deny the developers to make reports to authorities about suspicious-looking requests or work offers.
Always, some people say that lawsuits are needed to control AI or that AI development should stop for a while. That thing is not very effective, because people can use AI by using a VPN. And the AI may run on a supercomputer, that physical location is in a state, that doesn't limit the use of AI.
The point is a law that controls defense material exports made at a time. Where defense industrial material was physical. However, the militarized AI that can make computer viruses and deny the command systems operation can transmit over borders in the form of text files. These kinds of immaterial weapons are hard to limit. Hostile actors may make firm on some NATO countries. Then they simply hire developers and after that carry the source code to the East.
Another sad thing is this. When somebody asks to limit the AI development, at next morning, we see that the same persons start their own AI projects. So does that increase trust in those people or their motivations? We know that actors can misuse AI. The AI makes it possible to recognize thousands of people from some demonstrations. When the International Football Association started to use AI to follow the audience they argued that is a necessary thing for safety.
But the same things can used to recognize people from demonstrations. This is one of the reasons why China and other BRICS nations invest the AI development. The AI-based control systems make this possible. The drones can make escape and evasion movements if they are under AA fire. The thing is that nobody is interested the AI until it takes its job.
Then we must realize, that people. Those who work in application industries need information about AI. They really must know what they doing, when they use AI to make software. The AI doesn't remove the need to know the code, and especially what the code does.
https://scitechdaily.com/new-research-warns-ai-needs-to-be-better-understood-and-managed/
Comments
Post a Comment