Above this text is the model of the neural network system. Every point in that diagram is the ability or skill that the network has. And every line in that image is the connection between those skills. So every point in that image is the database. And the line is the database connection. Every single database can have a limited number of connections. Everything that the robot must do is stored in databases. And there is a series of actions that are connected to a certain table.
The neural network can be physical. It can be the network of physical systems like computers and surveillance cameras.
Or it can be virtual. The neural network can be the network of skills. And whenever the system learns a new skill it expands. The network of skills means that every single action requires sub-actions. The sub-actions like turning the steering wheel can use in many places. Same way robots can turn cars, tractors, and forklifts. So that means the robot can use the same skill to turn every vehicle that has steering wheels. The virtual system means a large number of networked databases.
The things like red traffic lights can act as triggers that launch a certain reaction. So when a robot sees red traffic light. That thing launches certain action. A great number of databases means that AI can search solutions more effectively. If the system already knows that the screwdriver is the tool it can search that thing faster. In that kind of operation, the robot goes to the toolbox. If there is no screwdriver it might ask alternative places. So the operator tells that maybe that tool is on the table. So the robot searches that place. If that screwdriver is there, the robot would involve the table to list places where the screwdriver could be.
Of course, the robot can search automatically screwdriver from the floor. The AI uses image recognition for separating the objects. If there is no screwdriver the AI can search the image of that tool from its memory and ask is this screwdriver? The human operator can check that there is the right image in the memory of the AI. If the image is wrong the operator can change it to right. The robot is not necessarily physical. It can be the algorithm that collects data from the Internet.
The learning system means the number of databases increases. And the number of connections between them also increases. That means that if there is a lot of databases at the beginning of the independent learning that system can use more connections at the beginning of operations. And that makes the system can use databases versatile if there is a large number of data for use at the beginning of the self-learning.
Self-learning or autonomous learning process means that the system can increase the number of databases and database connections without human assistance. There is a theory that all databases on the internet can interconnect to one large database entirety. And that thing makes it possible to create the ultimate artificial intelligence that can interconnect all computers and other systems to one entirety.
The self-learning process can turn more effective. If the databases are pre-sorted by using certain parameters. In that model the database groups are sorted under topics like "visiting shop", cleaning the house" etc. Those databases involve what kind of things the AI must use for completing the mission successfully.
So when an AI-controlled robot is taking the order to go shopping it can interconnect the databases that are involving things how the robot should walk to the shop. Where are traffic lights, how must react to traffic lights etc.? The thing is that robots can use the same databases for multiple uses. The knowledge of how to react when traffic lights are red. Can use in all missions like walking, driving cars, and other things.
Image)https://www.quantamagazine.org/computer-scientists-prove-why-bigger-neural-networks-do-better-20220210/
https://thoughtsaboutsuperpositions.blogspot.com/
Comments
Post a Comment