 |
| Image: Pinterest |
Machine learning is the ultimate tool. But we are still far away from systems that are learning like a human. In the simplest model. Learning machines record the environment, and then the human user just stores those things for future use. Then the next time, the AI can automatically adjust the system by using those recorded values. So humans are needed for making the pilot case. And then, the system just repeats that case in the cases that follow the first one.
When we think of AI as an artist. We can select a couple of paintings from the net for the AI's database. Then the AI can simply share those images with three or more areas. And that it will connect those areas to the "new" painting. There is a possibility. That the number of stripes depends on the number of selected paintings or other images.
And then, the AI will take the top area from the first painting, the second area from the second painting, etc. So AI is not producing anything new. It just recycles the old one.
Chat GPT-type artificial intelligence will take Google's position someday in the future. The thing that this system requires is a good database, and also it must learn to read. When people are using Chat GPT they accumulate its database about the homepages. But the problem with the AI is that it doesn't know what is reading on those home pages. It can use Google search and find a series of home pages about certain topics. But I think that still today.
It uses the homepages Google indexed or some other search engine. Then it simply connects the paragraphs from the page and makes new homepages by using that data. When we want to simulate the AI way to handle data in a real-life physical environment. We can give the keywords to some child who writes them to address the line.
Then that child who doesn't speak English selects five first pages from the list. And then that child copies the first paragraph from the first page. The second paragraph from the second page, the third paragraph from the third page, etc... That thing means that the result can be perfect. But it can also be horrible.
The thing that makes AI problematic is that it uses only homepages that are enlisted for it. The system can search certain words from those pages and select paragraphs there are involving most of those words. But the problem is that the system cannot read. So the results still need human controllers. Or they are full of fatal errors. And I would not use AI for my doctoral thesis.
The AI can check the computer's programming code. It can see are variables used. Or it can see are variables written right. In that case, it just compiles the list of variables and the line that the programmer is made. If there is a variable, that looks like some certain variable, the system suggests how the programmer should write the variable's name.
In that process, the AI searches for details from those variables. It will see if there are similarities in the names and text that is written in the editor. This is one of the reasons why all variables must have unique names that are easy to separate from each other.
When AI controls things like chemical processes there are many ways how to make that system learn. The simplest model is to make the chemical test in the laboratory. And the system just records the conditions like relations in gas mixtures and temperature in the chemical environment.
If the maker of the experiment is satisfied that person just stores the result on the computer. And then the system can automatically adjust the system by using those values for the next case. But if the system uses scanning laser microscopes or something like that it can follow the trajectories of the components.
https://shorttextsofoldscholars.blogspot.com/
Comments
Post a Comment