Can we ever create artificial general intelligence (AGI)? That is a good question. When we think of ourselves and our way of learning and thinking, we must realize one thing. No human can make everything. We know stories about programmers and other kinds of specialists. That cannot make food. That is the thing that we can transfer to the hypothetical AGI.
The thing is that the AGI is useless if it doesn't make something. For making physical things. The AGI requires access to physical things like microwave ovens and robots. Another thing is that. The AGI cannot make everything, that it cannot find from the database. This is one of the problems with the AGI. The system requires large-scale data storage, and the ability to learn things like humans.
When humans learn something, our brains make a database about the thing. The problem is that we need a database for everything that we do. Our memory cells collect data that the operation requires like mosaic. Memory operates like neteye. It collects necessary information from multiple sources. And humans have very many memory neurons. There are billions of neurons and their connections in brains.
The number of human neurons is not static. The connections between neurons make it possible to create virtual neurons. And every virtual neuron, or neuron combination, acts like a physical neuron.
This is one of the reasons why humans have superiority. There is no computer with billions of memory blocks and memory handling units. The futuristic computers can have multiple processors with integrated memory units. But that means those systems require lots of energy. But if mass memories are integrated straight with the processor and every memory unit is operated by an individual processor. The memory operates faster. That kind of neural-network-based system can have multiple sub-computers, the processor entireties that operate as virtual quantum computers.
(FreeThink, LLMs are a dead end to AGI, says François Chollet)
But the problem is that if we want to make robots and AI that make food. We have a limited number of variables that the robot must handle. The robot must only know where certain dishes are. The robot can search for a match from receipts like the word "pepper". Then it can search for similar words from dishes. Then the robot can search for the details of what pepper looks like. If the robot doesn't know where it finds pepper it can search every bag and dish. When it finds the right thing, it can mark the position in its memory.
The kitchen involves limited space and a limited number of things. The robot can search for things when it has free time. In the kitchen, robots can simultaneously search and move every item, that they find. But at the city level, there are lots of variables. The robot cannot open every box that it sees, or the job takes a very long time.
The robot can look very intelligent if it goes to the shop. When a robot operates at home, it can use the home computer and surveillance system to see things like where it is. And when it goes out. The same body can connect to the city traffic control and GPS or other navigation system. Then the robot travels to the shop. And then it can connect itself to the shop's computer. That system knows where the right merchandise is. Then the robot can collect stuff from the shelf. The thing is that the robot is three robots. Mission control and databases are the things, that determine what robots can do.
The system that can operate a little bit like AGI can operate as modules. The neural network can connect those modules into new entireties. The module is like a room and operations. The AGI might not be possible, but that kind of modular AI can make it possible to make robots that follow orders like "Go shopping and take a milk bottle".
However, the problem is that the system cannot handle abstract thinking. It can calculate probabilities, but it cannot think. But the AI can predict many things. If two AI-controlled cars face and there are identical AI systems that control them, the cars can predict how the other car reacts. The AI can use probability calculations to predict the way, where people go. The system must only calculate how many people choose certain routes. Then it can make a prediction, of the probability that some person chooses the route.
https://www.freethink.com/robots-ai/arc-prize-agi
Comments
Post a Comment