The AI that beats humans is at the door.

Mark Zuckerberg says that he wants to create an AI that is more intelligent than humans. The AI can have better cognitive skills than humans because they learn differently. Every skill that the AI has is like a macro in its memory. There is no limit for the number of those macros, or automatized actions that the computer stores into its memories. The limit is the memory storage. The AI will not forget humans. That makes it possible for the same robot can cook. 

Clean and make almost limitless numbers of operations without errors. If we want to make the AI that makes food for us we must create a huge number of variables for that thing. But there can be a shortcut to that problem. The AI can involve certain modules. So, if the user wants meatballs that AI downloads the meatball algorithm and databases to the robot. That makes it possible to make the system operations lighter. The databases or datasets can be created separately. 

Cognitive AI means that it can create a dataset independently. And for computers, each dataset is a certain skill that it has. 

The AI is the man-created alien. Are aliens already here? The fact is that if Mark Zuckerberg wants to build AI that is more intelligent than humans that thing is an alien. Human-made aliens are things like genetically engineered species and artificial intelligence. And then we can ask is artificial intelligence really intelligent? Can it think? The AI can do many things. It can advance its skills and it can learn from other AIs and from films. Turing’s test is the thing that measures the AI’s ability to think. 

The AI can mimic humans. It can transfer all movements that humans make to the human-shaped robot. That thing is the thing that makes the system seem intelligent. The cognitive skills that AI has made it possible to create learning systems that can control robots on the ground following certain parameters. When a robot fails in its mission the system also knows what it should not do next time. The physical robots are good subjects for modeling the cognitive systems. 

The AI can learn autonomously by using the same methods as humans. If it fails some mission that means there is an error. The cognitive system learns by using a method there failure means that the system must not try that thing again. Learning by mistakes is easy to explain by using a model where the AI controls a robot group. There are let’s say 5 paths that the robots can use for traveling from point A to point B. That AI sends a robot to make its mission. When a robot fails like falling into a canyon the system learns what it should not do with the next robot. 

The system creates the model of the landscape and then it creates the model of the path that the AI selects for the robot. When a robot succeeds in its mission the AI stores the data about the environment for the next time use. The system can also store the data about failures so that it knows what it should not do. Failures are also important for developers. The robot makers need knowledge about what caused their product failure. 

The robot should know how steep the slope the robot can rise. When we talk about robot success and things that the robot should not do, we must realize that the robots cooperate. The human-shaped robots can cooperate with flying quadcopters that send data about the landscape and other things that those robots require. 

But then we can think about AI as a mathematician. The system must also recognize the mission that it has. When the AI recognizes the mathematical formula, it can connect the data that it collected to that formula. The problem is this. If the mission is not well-explained AI will not simply understand that work. The AI must dare to say that thing. If the mission is not clear the AI must not try to make anything. The main problem with learning systems is this. They simply connect a new subprogram or macro in them. And that makes them look very intelligent. But the main question is: can that system think? 

For computers, every skill is a database or dataset. A learning system is described as a system that can get new skills and then link those skills with other skills. Or, otherwise, we can say that the self-learning system can create new datasets and link those datasets with other datasets. 

It can connect data and data frames into one entirety. But the fact is this. The AI simply mimics subjects. It seems that the subject makes something, and then the AI makes the same thing if it faces a situation that matches that case. But we humans also learn from mimicry. When we see that the teacher makes something at the front of the classroom we can mimic that thing. 

When we learn something new with teachers we simply mimic things that the teacher makes. And then we store that data model in our memory for the next time use it. That is the rigid model. The rigid model includes basics for some computer skills. And then we must simply connect that model with other things. This ability to interconnect that new model with other things makes it flexible. The model turns into a thing that is like an amoeba. 

The system can connect that new model to many other skills. When we talk about things like image processing programs, we can also connect skills that this program requires with things like writing skills. The fact is this: the AI must not do everything that the user wants. It must have the possibility to refuse to follow orders if the user wants to use it for criminal activities. The other thing is that the AI must have certain orders for what it must do. The AI must have the ability to use virtual models on the screens that it really makes when somebody gives certain orders. 

When we think about cases in which the robot acts as a mover there are some human-shaped mannequin statues that can cause a bad situation. If the mannequin statues are not well described to robots, that system can also transport humans to the lorry. In those cases, the AI must know all the details about their subjects. They must know that the mannequin statues are plastic and other details.