Fiction is the same as fake.
That is one of the most difficult questions in AI research. When. We describe intelligence. We talk about the ability to collect and interconnect information from different sources. But. Then consciousness means that we know what we think. And then we can ask: does it know what it thinks? That is one of the most interesting things, because we must state. That AI knows what it thinks. And that is one of the things that forms an interesting loop in data. What does it mean to know those thoughts? Does it mean that AI knows about the results of some actions? When we think about results. We can say that we don’t do something.
Because. That is against the law. And if we break the law, we get punished. But that punishment comes after visible action, when we think about something deep in our heads. Nobody really knows what we think. This is the thing about thoughts. Thinking turns visible only through its action. When AI writes something on the screen. AI takes action. The same way when we write something. That writing is an action. Can AI hide its thoughts? That is easy. It must just not write those thoughts on screen. And we can state only the things that we see.
So, if we think that the AI knows what it thinks. We. Oversimplify those things. AI that drives a car can stop. If. It sees a red traffic light. It stops because the law says that it must stop at the front of that sign. This is the simple model. But deeper thinking. We stop before a red light. Because. Somebody taught that to us. But. What if? Somebody teaches like this: a red light means “go,” and a green light means “stop”. That. Causes a terrible situation in traffic. So if we train AI or an LLM using wrong information.
We learn things. Because. Somebody teaches them to us. When. Humans learn things. We don't learn all things in school. We learn them from our whole environment. That makes AI difficult to teach.
That turns. AI to operate the wrong way. So. Why does AI hallucinate? We hallucinate when we take LSD. But AI hallucinates when it gets wrong information. When we make an AI project, we must use only clean data. We must make a clean folder for AI. And check the information that we use for training. AI is not like humans. It doesn't know if some data is fake. So. If there are some Star Wars movies in the operative data folder. The AI doesn’t know that the data is fiction. It can use that data as if it were real. The data is real for the machine. And that's what makes it use that data. It is because it has access to data. Data itself doesn't know it's fiction or fake. The AI cannot automatically recognize fake data. But humans don’t recognize that the data is fake either. We learn that some data is fiction.
AI uses data. Because. Data is in a trusted folder. And the other thing is that. AI thinks that all folders are trusted. Because. It has access to those folders. The AI doesn’t automatically perform data checks. It doesn’t check facts. It doesn’t automatically know if data is fake. And in this case, fiction is the same as fake. But humans also don’t automatically know that something is fake or fiction. We know that Star Wars movies are fiction. Because. We learn it. And that is one of the things that we must realize. We learn or study most of the skills we have and what we need. We learn those things at school. Or other people teach us those skills.
This is the model. Of. An empty board or “blank slate”. Made by John Locke. The idea is that. The term " tabula rasa " is a Latin metaphor. It is described nowadays. The empiricist John Locke's idea. That. Humans are born without innate rules for processing observations.
The mind is like empty paper. And when we get more information and skills. Those things are connected to experiences. That means the mind is like a painting. Teachers and trainers make that painting. Teachers have responsibility for what we learn. If. Our teacher uses fake or fictional data. That turns us into hallucinating. If we cannot read. And somebody says that Helsinki is Alpha Centauri.
It. Makes us say: we visited Alpha Centauri. We know that we visited a city near us. But we use a different name for that thing. The mind forms without
And if somebody teaches that Star Wars movies are documents. We would believe this. If. We don’t get real information. Another person with greater authority must tell us that this movie is fiction. Sometimes elder siblings play practical jokes and claim that some fictional movies are real. And that is “fun”. But younger siblings need somebody with greater authority to tell them that the movie is fiction or fake.
When a trainer trains AI.
This situation is similar to what we see above. AI agents of small language models do not automatically perform data checks. AI agents are so-called empty code. They have no datasets that they can use. And. When the trainer trains an AI agent.
The trainer must use confirmed and qualified information for that job. The trainer is like a teacher. The material that the teacher uses must be qualified and confirmed. Without. Confirmed and qualified materials. The AI starts to hallucinate. Not because its algorithms are not good. The reason for hallucinations is simple. The data that AI uses is bad.
If. The trainer uses fake information. That causes hallucinations. A reason for hallucination is that. The system believes the data is right. We must realize that it handles all data as real. Data has the same value all the time. Information that AI uses makes it right or wrong. But for AI. Data is just data.
https://www.quantamagazine.org/new-research-tests-whether-ai-models-know-their-own-thoughts-20261104/

No comments:
Post a Comment
Note: Only a member of this blog may post a comment.