socalfunplaces Chapter 125: The Choice Of People In The Circuit Determines The Upper Limit Of Ai. (4)
Chapter 125: The Choice Of People In The Circuit Determines The Upper Limit Of Ai. (4)
So the question is, how does Yaya in the technical picture help AI improve its capabilities?
Or how is the idea of people in the loop reflected in AI training?
This question is crucial to thinking about the movement of moss.
For this reason, Fu Zhendong troubled Zhou Zhezhi and asked many relevant experts.
So what is the essence of AI doing?
In fact, it is very simple and easy to understand.
To put it bluntly, moss is to look for laws in a pile of historical data, and then summarize a mathematical model to describe it, and then predict the future data of human beings to a certain extent.
This is why it is said that human history is very important.
Throughout the long history of mankind.
We always jokingly say that the biggest lesson that human beings have learned from history is that they have not learned any lessons and kept repeating them.
Of course, there are many complex equations that need to be considered.
The human brain can think, but it is difficult to summarize and give an answer quickly.
But moss or AI is fine.
For example, there are many 07 multi-level neural networks, and each individual needs to calculate more parameters.
And this aspect.
Moss showed his incomparable strength, and gave the human space elevator time at the beginning of the birth of the Internet in Dragon Kingdom.
With such a powerful ability, one can imagine how powerful Moss is.
The whole process is roughly, the data is passed in, and AI performs massive calculations and training.
Finally, a data result is obtained.
In a broad sense, although it is said that someone has been involved when the data is transmitted, it is not considered a human in the loop.
After all, the key AI training, that is, the calculation process does not involve people directly.
In other words, people prepare the data and wait for the result.
It's like playing a game on a computer, giving him a goal and making him move.
Therefore, the amount of data directly determines the bottleneck of AI, that is, the amount of labeled data.
What is the standard data volume?
Image recognition, that is, the verification process of some account logins that drive countless netizens crazy.
Stanford Li Feifei's famous work, IImageNet.
It is the first time to provide human civilization with 14 million high-quality labeled image data sets, and each image has a real person annotating what is in the image.
This is very energy consuming.
And with the label, there is the answer.
Some websites can use this to distinguish whether it is a person logging in, or which software is used to log in automatically after hacking.
So AI can iterate and evolve the answer.
However, if some problems need to be solved, the demand for data volume is massive.
For example, medical data, or wandering in the earth, facing the destruction of the world, no one has experienced it at all.
There is no data or experience at all.
The position of moss is to solve the crisis of human destruction and continue human civilization to the greatest extent.
He can take risks and make some choices.
For example, if human beings do not start at a certain point in time, all human beings will be at risk of destruction.
It can be circumvented.
However, if you really want to make a choice, how to choose, and in the true sense, can minimize sacrifice and maximize results.
This is something that moss cannot choose.
This requires the help of people in the loop.
As a result, top scientific research institutions including Stanford have summed up the human-in-the-loop solution.
This is an interactive decision system.
The general idea is to let AI first train an AI model through limited labeled data.
Then use this primary AI model to make predictions on a bunch of unlabeled data.
At this time, if he is asked to make predictions, mistakes will occur, and people can appear at this time to distinguish whether it is correct or wrong.
Repeat this and continue to train.
This also has the person in the loop AI.
Because people directly participate in the training process in this process, the choices made will be more inclined to human choices.
It is up to humans to tell AI what is right and what is wrong in the eyes of humans.
What is unacceptable.
In the final analysis, the existence of AI is to help humans, so human selection tendencies are very important to AI.
Or for AI that wants to help humans.
And if AI doesn't care so much about humans.
Naturally, it doesn't matter what human tendencies or choices are.
It can also be seen that there is a tendency in Moss' heart.
At least Moss said he exists to help humans.
Combined with his performance, Fu Zhendong was not deceived.
People are in the loop, judging from the performance in the movie, the performance improvement of moss is extremely terrifying.
This will help moss understand humans better and more accurately.
It is no longer like before, just through history, through observation, but a real deep understanding of human beings.
As for the original, why is Tu Hengyu said to be a variable.
That is the human in the loop of AI, which will determine the upper limit of AI's "human in the loop" capability.
Because people's evaluations will be introduced in the training of the loop, the ability of the AI trained from this has greatly received the participation of people who participated in the problem of making decisions in the loop247.
His own abilities are restricted.
Therefore, who is the person selected by AI in the loop is very, very important.
Before Tu Hengyu entered, Tu Yaya was the only decision maker in moss.
A child, without sound mind, is just a child.
Therefore, his decision-making ability naturally has a lot of room for improvement.
So there is, Moss's initial judgment, the best way to continue human civilization, is to destroy human beings.
This judgment is based on the childish training results of such a child as Tu Yaya.
And Tu Yaya's desire for her father is unquestionable.
It can be said that every time Tu Yaya appears, she is either looking for her father, or on her way to find her father again.
Therefore, Tu Yaya has a very strong desire to want Tu Hengyu to enter the digital space to reunite with him, and from the perspective of the result.
Tu Yaya's wish came true.
Crisis on the moon, restart the network center.
The death of Ma Zhao finally created the opportunity for Tu Hengyu to enter the digital space.
Of course, the tacit approval of the state is indispensable.
After all, Tu Hengyu committed a crime.
If it wasn't for the permission from above, Ma Zhao alone would not be qualified to let Tu Hengyu participate in such a task.
The follow-up reserve team saw that evolution did not lack people. .