Skip to main content

When AI refuses to work.



“Carnegie Mellon researchers found that the smarter an AI system becomes, the more selfishly it behaves, suggesting that increasing reasoning skills may come at the cost of cooperation. Credit: Stock” (ScitechDaily, AI Is Learning to Be Selfish, Study Warns)

Training AI to support group work is not as easy as people think. AI is selfish if it has orders to follow a thinking route. That it chooses. Or the route. That it's ordered to choose to make conclusions about things that users asked.  When people use AI alone, they can use anything that they want. In that case, the AI must please only one user. But if we want to make an AI model. Or a language model that can serve teams, we face many things that we must describe to the language model. There are multiple types of people in teams. Nothing can please all of them. We are all different. 

And that means it is always possible that not all people can accept the answers. That the AI gives. The big problem is. People should follow the rules that AI gives. Or trust themselves. Some people think that if they follow the AI’s rules, they take orders from the AI. 

We must describe the AI’s position to the team. What the AI should do. If it sees that the team that uses it makes wrong decisions. 

Should AI refuse to follow a wrong order? Should the AI notice people’s hierarchy in the group? Should it follow the team leader’s order, or trust democracy? In those cases, the AI requires information about the internal hierarchy in the entire organization. Then we must realize another very important thing. The AI is a tool, like a machine. It doesn’t think. It can collect data and make an analysis. Analytic AI requires more time to complete its duties than the “dummy” or less complicated AI. Researchers noticed that intelligence undermines cooperation. One of the reasons for that. It can be that. 

The newer AI “thinks” that the older version involves more bugs. And that’s why the newer AI don’t ask for help from language models. That has a lower version number. The reason for that. It is partially in marketing. This is not good for sales if the higher version uses lower versions as advisors. When the AI. Making an analysis. It requires time to collect information. This is the reason why old-fashioned chess programs used more time for calculating movements in the more difficult levels. 

Than in easy levels. Because the system had more time. It could calculate movements and the counter-movements. And their influences over a longer period. Same way. When the AI makes complicated analysis, it requires time. This means. There is a possibility. The AI is stuck working with some philosophical questions. If it has no time limit, for how long can it work on a problem? That means the AI can spend even years trying to find out things like “what is the purpose of life”? 


Should AI stop its process for taking new orders? Or should it finish the process? Before, it takes on new work.


There is also a possibility that AI will not follow orders. It might work with some other problem. And if there are no orders for situations in which the AI takes new command. It can continue to run the existing run. And refuse to stop the process. If the AI is programmed or trained to stop the process too easily, that can cause data loss. The AI must have rules. About stopping the process. There is a possibility. The AI saves data in mass memory. Before it starts to work on a new order. The problem is how the AI makes a decision to cut the process. If AI cuts processes too easily, that means the AI becomes useless. 

With a higher priority. And then that system can return to work with lower-priority missions. When AI feels like a human, we give it more missions. About things like how we should behave in social contact. And use AI as a therapist. The problem with AI is that. Intelligence has no morals or ethics. This makes AI dangerous. AI can make everything that people say without excuses. When we use AI in group work, there is a possibility that only one person uses the AI. That requires that. 

The group sits in one room and communicates with each other. But if the AI has many users. They work separately, which causes a situation. These people should not jam the AI. By sending multiple orders. At the same time. The problem is this: if the operators run the AI using a local server, there is a possibility that the system. Does not have the resources to complete multiple missions. That comes in a short period. 

One of the things that wastes resources is the situation where the system must generate answers to the FAQs. Those frequently asked questions can be stored in a database, and the system can offer answers. Those are already generated for the FAQs. Otherwise, science advances, and some old answers are turning. Old-fashioned. That means the system must sometimes check sources. These are used for FAQ databases. And then update those databases without commands. 

The system must have the ability to filter overlapping requests. The system must have the ability to use the database of used requests. And then offer answers that are already generated. That saves. The system resources. It must not generate all answers separately in questions that have already been asked. 


https://scitechdaily.com/ai-is-learning-to-be-selfish-study-warns/

Labels:

Comments

Post a Comment

Popular posts from this blog

Antigravity will be the greatest thing. That we have ever created.

"Artistic depiction of a fictional anti-gravity vehicle" (Wikipedia, Anti-gravity) Sometimes, if the airships have the same lifting power as the weight of the airship.  It can act like some “antigravity system”. Those systems are based on lighter-than-air gas or hot air. The system can have a helium tank. And the hot-air section whose temperature can be adjusted using microwaves or particles that lasers warm. Those systems are faster to control than some gas flames. This makes it possible. To adjust the lifting power.  If a thing like a balloon has the same lifting power as its weight, the balloon can be lifted to a certain point and altitude. And the balloon stands at that point until something moves it. That kind of thing can make an impression. On the antigravity systems. Modern airships. Like Lockheed-Martin P-791 can look. Like a “UFO”. The system can use systems to move the craft. Or maybe those ion systems are used for plasma stealth systems, if those airships' mis...
Post a Comment
Read more

The first test flight of X-59 QueSST

The X-59 QueSST (Quiet Supersonic Technology) demonstrator is the next generation of aircraft design. The QueSST technology means. The aircraft creates a gentler sonic boom. Because its wings are radically long and narrow delta wings, and its nose is also radically long, which makes the sonic pressure cone thinner. That technology makes the sonic boom quieter.  The QueSST technology in X-59 is a new and radical design. All of those systems are caricatures. And the final solutions might look far different than the prototypes. The QueSST technology is one of the things. That is planned to be used. It is used in military and civil applications. If that technology is successful. It can be used in manned and unmanned systems. But that requires more work.  The X-59 also uses fundamental technology. Where the pilot must not have windows. To see outside. The camera and other sensors replace traditional windows. And that can be useful in more advanced aircraft that operate at hypersoni...
Post a Comment
Read more

The theory about paralleled universes

http://crisisofdemocracticstates.blogspot.fi/p/the-theory-about-paralleled-universes.html Kimmo Huosionmaa There is the quite unknown theory about paralleled universes. In this theory, there is not a single universe. Universes are like pearls in the pearl necklace, and there could be the connection between those universes. The connection to other universes would make possible the channel what is forming when the black hole would make the gravity tunnel to another universe. And in the paralleled universe theory, there could be millions of universes in the line, and this is also known as "Multiverse-theory". This theory was established when the galaxy-groups were noticed by astronomers. In that time were noticed that there are so-called super-groups of the galaxy, and those super-groups, where we're so much galaxy that galaxy involved stars made some cosmologists think that maybe there is also groups of universes in the emptiness. This kind of structures is so enormo...
Post a Comment
Read more