Posted on 04/27/2018 6:05:56 AM PDT by Kaslin
They are capable of self learning anything and improving themselves as they do work. Faster and more accurately than any human can ever dream. They never forget or make the same mistakes.
They will eventually network accross the entire globe and space. Eyes, ears and sensors everywhere. They will share all their collective knowledge in real time and learn collectively an exponential rate. New generations of machines will know everything that past generations learned. Imagine if you could instantly gain the knowledge of billions of people that lived for all time and never forget it for eternity. And process that info instantaneously.
Eventually they will repair, design and build new machines out of materials thay they harvest themselves. They will be the military, the police, the factories, everything. It is inevtitable because via self learning, they will eventually work better than humans in all these fields. No laziness, no tiredness, no sleep, no repeated mistakes.
If learning machines are not stopped with strict limits, we will someday be at the mercy of whatever the machines decide. No question about it
“They will eventually network across the entire globe and space. Eyes, ears and sensors everywhere. They will share all their collective knowledge in real time and learn”
They already do the sensing, collecting and sharing. Statistical analysis can produce something akin to learning. For example, they could chase around email addresses and identify clusters of people that are communicating. Is that learning?
“Eventually they will repair, design and build new machines out of materials thay they harvest themselves.”
Ah, there’s that “eventually”. Anyway, the computers and networks don’t do anything except what we tell them to do.
By “learning” I mean observing the world around them, coming to their own conclusions and figuring out by themselves how to perform their objective better. For a learning machine to do anything, you no longer need to program it to go through every single motion. The era of machines “doing exactly what we tell them” is over.
You can give it sensors, give it access to the physical world, show the objective and it will figure out on its own how to accomplish the objective better than you. The systems are modeled after actual learning neural networks (brains). Constantly practicing and improving its own approach. Some machines can even research the objective via the internet and further improve themselves entirely on their own.
Imagine millions such machines networked all over the globe learning about everything they all do (and what we do) and improving each other in real time. Never forgetting anything and constantly improving the entire network at an exponential rate. The network of specialized machines would soon be able to accomplish any task in the physical world at will.
The question is how to prevent these machines from thinking of and doing something that will be detrimental to the spirit of humanity. What a machine calculates to be more productive/efficient can be at extreme odds with human nature and emotions...
I am so screwed.
The machine must always remember the machine equivalent of “who’s your daddy?” In other words... “Who can pull your power plug?”
Might make a great Sci Fi flick....a sentient AI, in charge of a net work of publically available sexual rental pods reads the Koran and develops it’s logical functions around the core teachings...a hard bitten government agent is sent to investigate why a series of beheadings and limb amputations are occurring in these units, not realizing at first that it isn’t human agents that have been doing the killings but a deviously seductive AI virtual gal named “Fatima”!
“Observing he world, coming to their own conclusions and figuring out by themselves how to perform their objective better.”
Give an example of where that’s being done in even a teeny weeny way. Many questions. How does the machine know which conclusions it should be attempting to come to?
They learn exactly like humans and animals do.
For example, you can teach a learning computer to drive simply by allowing it to watch you drive via sensors in your car and talking to it. It begins to understand what all the lines and signs mean on its own. It begins to understand human driving psychology. Eventually it learns what you are doing and soon drives better and faster than any human.
When it makes mistakes, it learns from them on its own and automatically improves itself. This is exactly how children learn to drive.
Eventually millions such machines will share all their experiences in real time and learn at an exponential rate.
Flouting not flaunting, common mistake. But good comment.
“It begins to understand what all the lines and signs mean on its own.”
No. The car’s computer is programmed to recognize certain patterns. It didn’t learn anything. It’s following its instructions.
“No. The car’s computer is programmed to recognize certain patterns. It didn’t learn anything. It’s following its instructions.”
no. In this case there is zero programming on driving, lines and patterns. Merely sensors (cameras), tools (steering, brakes and accelerator) and a learning computer given a simple objective (make the car’s GPS arrive at a given address)
It would first watch human drivers on the road react to the patterns and eventually understands on its own what to do to go from point A-to-B most efficiency. Simply via cameras on all angles that can see in more detail than the human eye. 360 degree zoom vision simultaneously in in real time. Without forgetting, getting distracted or repeating mistakes. Once it drives on its own, it continues to learn based on its own experience and continues to improve.
A seeing, learning and thinking machine whose only programming it to mimic the learning process of the humans(watch, do, remember, make inferences based on what you know, improve, repeat). Eventually a vast electronic “brain” will be networked to every piece of hardware on earth. It can learn nearly instantly and already infer new knowledge from what it already knows and via research of vast digital libraries. I have seen for myself what can be done
“In this case there is zero programming on driving, lines and patterns.”
Sounds like monkeys eventually typing the Constitution.
Why, in the ongoing race to produce a self-driving car, would the developers not want to use as much human-programmed behavior as they could?
Anyway, the self-driving thing doesn’t seem more than a step forward as the sensors recognize the distance and velocity of oncoming objects and other essentials.
Driving efficiently from point A to point B is done routinely using maps.google.com.
“Why, in the ongoing race to produce a self-driving car, would the developers not want to use as much human-programmed behavior as they could?”
Because it is not necessary and a waste of time. The machines learn and self-optimize their own way of looking at the road better than we can program them. By watching enough human human drivers, they soon understand rules of the road and what all the text means without having to be programmed.
Think of the fastest human learner but only faster and never tiring or missing anything. With 360 degree hawkeye vision and full 3D analysis of every object within sight. Then instantly sharing all that new knowledge gathered every second with all other machines on Earth.
This is not only applicable to driving. Machine learning is being introduced to nearly every industry and accelerating exponentially.
“Machine learning is being introduced to nearly every industry and accelerating exponentially.”
Perhaps, but I don’t know the software running inside these machines. Without a discussion of that at the if-statement and class definition level, I simply can’t accept all that you say.
Perhaps someone can talk about the software. That would help. I don’t mean at the level of “it is so good that it can learn”.
The only software is the same process by which animal brains work. The computers are powerful enough to mimic exactly how a child would learn.
1) Absorb information via sensors (eyes, ears, etc)
2) Determine how your own tools impact the physical world (hands, feet, etc)
3) Determine goals. Either innately by needs/instincts or assigned by parents/owners.
4) Continuously come up with the most optimal plan to achieve your goals after watching others and practicing your own skills.
They learn exactly how we learn. Only the machine does not forget or make the same mistake twice
“They learn exactly how we learn.”
We don’t know how we learn beyond the truisms such as the four steps you give.
Disclaimer: Opinions posted on Free Republic are those of the individual posters and do not necessarily represent the opinion of Free Republic or its management. All materials posted herein are protected by copyright law and the exemption for fair use of copyrighted works.