What mean to AI to understand?

 What mean to AI to understand?

What mean to understand anyway? We can do many things. And we might not understand them. When we are trying to think about the question of what means understanding? We are facing an ultimate question do we still understand anything? The fact is that I can put seven years old child to read the text about quantum physics and that child might read those texts pretty well. But does that child understand those words? The fact is that I can also use the Google text-to-speech application to make that thing. And this application will make its job well. 

The AI can make many things and if that application doesn't know about things like the mark of the sum or something like that. The programmer needs to store the characters of those things on the computer. The mark of the sum will store in the database, and then there will make the text that is connected to that mark. So the sigma-mark will trigger the words that are connected to that mark. 

1) To know means: That a character knows what to do in certain situations

2) Understanding means: To realize why actors must do things in a certain way.

When we are making the robot do something it makes things what we programmed in it. If we want to make a tennis robot that plays tennis with us, we might make a robot that hits the ball. The robot might have gesture control.

If the ball is coming to the robot it will hit it. Then robot must have some algorithms for how it aims at that ball. If the referee will give the pass to the robot, it must know the gesture and then make the pass. Or movement series what makes it pass. The robot must calculate many things like the right hit point and power. But then it might have one problem does it understand anything? 

The pseudo-understanding is that the AI can give pre-programmed answers to certain questions. 

It knows how to react to the ball. And simple gestures that are making the person who sits on the chair.  But could that robot play tennis in a real match? Does it separate the referee from the audience that might show similar gestures? The robot must "know" that it should not follow any other than the referee's marks. So the robot knows how to punch a ball. 

The ball acts as a trigger that activates a certain series of movements. The robot might have orders where the punch must and where the ball should not go. The robot would not strike outside the field, because it's programmed in there. And if somebody asks about why the robot doesn't hit the ball outside the field area, it can answer: "that's dangerous". 

If the programmer is put that answer to robot's computer. Or it might have an answer "that's prohibited" whenever the person asks it to make something that is not programmed in its memory. If somebody asks a robot to punch a ball to humans or vehicles robot might say "it's prohibited" and make a report to its operators. 

The reflex robot recognizes that some action is filling the notes that are stored in the database. After that, the action triggers the database. And then that database begins the response to that action. 

Those actions are programmed in the program of the robot's programs by programmers. The robot does all the time same things. There is a series of triggers that are activated by certain actions. So that robot has a reflex. A certain action activates certain types of reactions. 

The reflex automation is simple to make. When somebody says "good morning" to the computer, it might answer by saying "good morning". And then that computer might have a voice or image scanner that connects a certain workspace to it. If the computer uses an infrared camera or ultrasound-based system. It can also recognize a person. Even if that user has a beard or is in flu. The idea of those deeper-than-surface systems is to benefit the static components of the human body. 

Of course, the system can ask the person to identify self. The command that the operator gives for access to the workspaces. Might be "I'm Eric, open my workspace". In the place of that name is the operator's name. That means the system can also recognize if somebody tries to play as that operator. The system recognizes the face but asks the name of the operator. That uncovers if the person tries to use some other user's accounts. This is one version of artificial intelligence called "reflex automation". 

The thing is that the machine has some kind of model in its memory. When some action fits some models. That thing activates certain actions in the system. This type of system is effective. Artificial intelligence-controlled robots might make many things like activating traffic lights or bringing tea or coffee to certain persons. They know how to respond to some kind of command or action. But those computer programs don't know why that response is given. 

A robot or computer program has a series of reactions to how to react to something. And if something that is outside its databases is asked robot might say "I cannot do that thing". Or it can say that it transmits the problem to the system supervisor who is making an algorithm for that thing. And the time for machine learning starts to dawn.