"Researchers have introduced Torque Clustering, an AI algorithm that enhances unsupervised learning by mimicking natural intelligence. Unlike traditional supervised methods, it identifies patterns without human-labeled data, making it more scalable and efficient. Inspired by gravitational torque balance, it achieved 97.7% accuracy in tests, surpassing existing approaches." (ScitechDaily, Scientists Unveil AI That Learns Without Human Labels – A Major Leap Toward True Intelligence!)
AI learns without human labels, which is a step toward real intelligence. But what does this really mean? Before we can answer that question, we must determine how humans learn. Humans need knowledge, memories, and skills to benefit our learning. However, the learning process itself is a mixture of memories and senses.
Those things make new memories that we can connect with some actions. When we learn something we need information that we can use. If we sit in a closed room without outside incentives and everything that we see is the wall, the result is that we don't learn anything.
The learning process requires information, senses, and motivations. Without motivation learning is not possible, or it's a very hard process. When we think about artificial intelligence and the learning process, we must realize that AI can make many things without deep knowledge.
If the AI-controlled robot stops when traffic lights are red. The robot doesn't need to know why the orders to stop, are given using a red light. It doesn't need to know the law texts behind the red light. It must only know that green light means "go". And red light means "stop". In the same way, humans must only know how to react to some signs like red and green traffic lights.
The thing is that learning means that the sense-and-response circuits connect new objects and new actions to the databases. The sense-and-response circuit can also be determined as a "see-and-react" circuit. When the AI sees something that event matches with a database it activates reactions that are stored in a database.
The problem with this thing is how to make the AI route observations about the event that it sees to the right database. The system can use similar, data-processing, and memory systems as human brains. They can use RAM hard disks as short-term memory. Then the system analyzes do it requires a new connection or a new skill Then the system decides. Will it trash that thing, or store that new connection between the event into the long-term physical mass memory? That saves stores in hard disks.
The self-replication is vital for the AI.
But then we can think that the AI can replicate itself. That kind of skill terrifies experts. The ability to replicate itself is the tool that the militarized AI requires. In the case of computer physical damage, the other computer can continue the missions that the destroyed computer starts.
The ability to replicate itself can also protect the AI and its code. That makes it possible to save the AI and the results that it gets if the computer faces physical damage. The AI can hover in RAM and write itself to multiple computers. In that case, the AI operates in non-centralized networks.
Those networks can be so-called morphing neural networks. And that makes the AI less vulnerable than ever before. The ability to replicate and multiply itself makes AI possible to create systems that mimic human brains. The AI can connect three computers in the entirety their larger computers act as the cerebrum and the third computer acts as the cerebellum.
https://scitechdaily.com/scientists-unveil-ai-that-learns-without-human-labels-a-major-leap-toward-true-intelligence/
https://www.space.com/space-exploration/tech/ai-can-now-replicate-itself-a-milestone-that-has-experts-terrified
No comments:
Post a Comment
Note: Only a member of this blog may post a comment.