Skip to main content

Neural networks cannot learn anything without information.

We can think that the neural network is like a factory. Information is the raw material that this factory uses. Sensors send the raw information to the neural network. And this information factory's product is processed information.

During information processing, the neural network interconnects information that it gets from different sources. And learning means that this system creates behavioral models for similar situations. Using ready-made templates. That made by using experiences makes reactions to similar situations easier and faster. 

Learning neural networks are systems where the data sources are interconnected with data-handling units. But without information those networks are useless. 

When neural networks categorizing the trusted pages that they use, they can make that thing automatically. The AI can use the user as an assistant, and show homepages to that person who estimates the data. And then the user can say if the data has enough high value. Or the user can simply put the home pages to a trusted list. 

That means if the user wants to use AI for physics that means that person can put things like "Phys.org" and "scitechdaily.com" into trusted page lists. And that means the AI uses those pages to select data. That kind of thing requires high skills to analyze information. So in this model, the system has a prime user. Who selects sources. That the AI uses. And then other users can use trusted data in their work. 

All systems need information for learning. And the problem is that when the system learns things. The system doesn't know whether is information that it uses is real or fake. This is the thing with all neural networks. And one of the neural networks is the human brain. The problem is this: without information, the system is like an empty paper that cannot do anything. 

Learning is that the system makes the models using the information it gets from sensors. Then the system can escalate that model to all similar situations. That thing means machine learning and human learning are similar processes. The machines get information the same way as the human brain. But there is one big difference between those learning models. The AI doesn't understand the meaning of the words. 



AI is an impressive tool. And there are trillions of observation systems that are transferring information to AI-based systems. NASA and other organizations send space probes to take information from distant planets, and the AI can interconnect that data with other systems. In this version, we are talking about AI-based systems in astronomy. 

The AI can search and follow the changes in the brightness of stars. That helps it to find new exoplanets. But without simultaneously taking images the AI cannot compare the object's brightness within a certain period. So without the telescope, the AI is useless. 

When we are celebrating things like ChatGPT and other AI systems and telling how good answers they can make, we must realize one thing. Those systems use certain categories in how they choose the homepages where they get information. And that thing means that the system might not know. Is the information on those home pages trusted or faked? 

There are certain parameters in how the AI selects homepages where it gets information. And there is a theoretical possibility that somebody makes a practical joke to the person who uses AI for making the thesis. The joke could be that some other person would change the information that is on the trusted page for a moment when AI uses that page. 

In this version, the AI makes mistakes because it doesn't know what the text means. And there is the possibility that by changing the text on the trusted homepage to things like lines from Donald Duck the AI would put that thing to the thesis. Fixing that error is quite easy. The AI can use two or more home pages. And if there are big differences, the system can introduce those homepages to the user. The user makes the decision that is information valuable. 

So that means AI requires a new type of skills in working life. The person who uses the AI must have skills to estimate the text, find conflicts and then estimate the source. 

Science advances very fast. And that thing means the AI must have parameters that it selects only the newest possible data. And that means the AI must use only the last updates on trusted homepages. The problem is that our "practical joke" would be the last update. 


Comments

Post a Comment

Popular posts from this blog

Einstein’s light cone and five-dimensional spacetime.

“Light cone in 2D space plus a time dimension, more commonly referred to as ‘space time’” (Wikipedia) The light cone introduces the model. That time focuses information on the hypersurface of the present. But why can't we see into the past or future? Entropy is introduced as the reason for. Why can't we see the future? Or to the past. This means that entropy scatters information into such a small mosaic and twists it. That information exists. But. We cannot collect it into a form. That. We can use it.  Term spacetime, or. Space time. Means the space and time. “In physics, spacetime, also called the space-time continuum, is a mathematical model that fuses the three dimensions of space and the one dimension of time into a single four-dimensional continuum. Spacetime diagrams are useful in visualizing and understanding relativistic effects, such as how different observers perceive where and when events occur.” (Wikipedia, spacetime) There is a connection between time and space, an...
Post a Comment
Read more

Supermassive black holes from the universe's dawn.

“An AI-generated visualization of a distant galaxy, containing, besides dust, gas, and young stars, three massive, active black holes (black spheres, not to scale) with bright accretion disks. Other distant galaxies are shown in the background. And. A few stars in the foreground. Credit: MPE (generated with AI).” (ScitechDaily, Three Supermassive Black Holes Found in a Galaxy From the Dawn of the Universe) Supermassive black holes from the universe's dawn are among the most interesting objects in modern astronomy. Reseachers found three supermassive black holes in one galaxy. And the famous “little red dots” from the dawn of time. They are so-called black hole stars. Or at least some of them are.  Those objects should tell us about the history of the young universe. And. Those distant objects. Tell us about gravitational interaction behavior over very long distances. We know that gravitational waves have a speed. Gravitational waves travel at the speed of light. This means. That gr...
Post a Comment
Read more

A superkilonova, which has never been seen before, may be spotted.

"A puzzling cosmic blast detected in both light and gravitational waves may hint at a previously unseen type of explosion, challenging astronomers to rethink how neutron stars are born and collide. Credit: Shutterstock" (ScitechDaily, Astronomers May Have Spotted a Never-Before-Seen “Superkilonova” Hidden Inside a Supernova) "When the largest stars exhaust their fuel, they end their lives in powerful supernova explosions. These blasts scatter heavy elements such as carbon and iron into space, helping enrich the universe. A different and far rarer kind of cosmic explosion, known as a kilonova, happens when two neutron stars collide. Neutron stars are the dense remnants of dead stars, and when they merge, they can create even heavier elements, including gold and uranium. Materials like these later become part of new stars, planets, and other cosmic structures." (ScitechDaily, Astronomers May Have Spotted a Never-Before-Seen “Superkilonova” Hidden Inside a Supernova) A...
Post a Comment
Read more