Author | Quantum Jun
Source | Quantum School (ID: quantumschool)
Behind the AI
In 2015, AlphaGo and the human Go genius Lee Sedol went to a decisive battle.
In the fourth game, Li Shishi judged that there was a chess in the black air and played a white 78.
Li Shishi, an epic "hand of the gods", embodies the peak of human intuition, computing power and creativity.
Five years later, Li Shishi, a talented chess player, has retired.
AI has forced humans to retreat in a row in various areas of intelligence.
In 2016, DeepMind defeated Ke Jie, who was ranked number one in the world at the time.
In 2017, Libratus won the Texas Hold'em Poker War.
In 2018, the accuracy of Watson's lung cancer treatment reached 90% and exceeded that of human doctors.
In 2019, AI began a stormy revolution in "deep learning".
…………
It is also said that AI in the natural sciences can defeat humans.
But in the field of art, artificial intelligence has no way to compete with humans.
Does AI really know nothing about art?
Even if it can't write a classic like the Ninth Symphony,
Can't you create catchy children's songs?
Even though AI cannot "freely think and express subjectively",
But it's always okay to help mankind in art.
Can AI really not "subjectively create"?
What is the thinking mode of AI?
What is the difference between AI's intelligence and human intelligence?
To answer these questions, we must first study "Bayes' Theorem".
Because it is the cornerstone of intelligence hidden behind AI.
The "unscientific" Bayesian formula
There are many geniuses in history, who were unknown when they were alive, and worshipped after death.
The 18th century mathematician Thomas Bayes was one of them.
The "Bayes" theorem is derived from the papers written when solving the "inverse probability" problem.
Before that, people would only calculate the "forward probability" .
What is "positive probability":
Suppose there are P red balls and Q white balls in the bag. Except for the color, the other properties are exactly the same. You reach out and touch it. The probability of touching the red ball can be calculated.
But whether the reverse can also be calculated, we can regard it as the "reverse probability":
If we don't know the ratio of red and white balls in the bag in advance, but we close our eyes and draw out some balls, then we can guess the ratio of red and white balls in the bag based on the ratio of red and white balls in our hands.
This problem is the inverse probability problem.
In layman's terms, like a superstitious HR, if you encounter a Virgo candidate, HR will infer that the person is mostly a perfect person.
This means that when you can't accurately know the essence of something, you can rely on experience to judge its essential attributes.
This research seems unremarkable, and the little-known Bayesian is not noticeable.
The paper he wrote was not published by a friend in 1763 until the second year after his death.
The pearl is covered in dust, just like Van Gogh in the painting world. No one cares about the paintings during his lifetime, and they are invaluable after death.
Why has Bayes' theorem been hidden from view by scientists for more than 200 years?
Because it is contrary to the classical statistics of the time, even "unscientific".
In classical statistics, the law of numbers comes from random sampling and calculation.
The Bayesian method is based on subjective judgment. You can estimate a value first, and then continuously modify it based on objective facts.
Starting from subjective guessing, this is obviously not in line with the scientific spirit, so Bayes' theorem is criticized.
In 1774, the great French mathematician Laplace also saw the value of Bayes' theorem.
But he knows the common problems of mankind and always uses tradition to oppose new ideas.
He was too lazy to argue with others and directly gave mathematical expressions:
Bayesian formula works like this
Bayes' theorem is simple, elegant, and profound.
Bayes' theorem is not easy to understand, and every factor has deep meaning behind it.
How does it "serve the people"?
For Bayes' theorem, referring to the above formula, we must first understand the events corresponding to each probability.
P(A|B) is the probability of A occurring when B occurs;
Also called the posterior probability of A, it is the re-evaluation of the probability of A event after the occurrence of B event.
P(A) is the probability of occurrence of A;
Also called the prior probability of A, it is a judgment on the probability of A event before B event occurs.
P(B|A) is the probability that B will occur if A occurs.
P(B) is the probability of B occurring.
And the meaning of Bayes' theorem is self-evident: first estimate a "prior probability", and then add the experimental results to see whether the experiment strengthens or weakens the "prior probability", and after correction, the result is closer to the facts. "Posterior probability".
I know you didn’t understand... Let’s give an example!
Let's take the COVID-19 epidemic as an example.
Assuming that the incidence of COVID-19 is 0.001, that is, 1 in 1,000 people will get sick.
A virus research institute has developed a reagent that can be used to test whether you are sick.
Its accuracy rate is 0.99. That is, if you do get sick, it is 99% likely to be positive.
Its false positive rate is 0.05, that is, if you are not ill, 5% may be positive (that is, a headache in the medical field "false positive")
Here comes the scary thing: if your test result is positive, how likely is it that you are indeed sick?
This is a terrible question, you must want to know the result, so you have to take a good look at the following inferences.
Assuming that event A represents illness, then P(A) is 0.001, which is the "prior probability".
Assuming that event B is positive, then P(A|B) is calculated, which is the estimated incidence after the test.
P(B|A) means positive in case of illness, that is, "true positive", P(B|A) is 0.99.
P(B) is a kind of total probability, which is the sum of the probability of occurrence of B in each sample subspace. It has two sub-cases, one is the "true positive" without false positives, and the other is the "false positive" with false positives. After applying the full probability formula:
A reagent with 99% accuracy, you are tested positive.
You may be scared and lost, is life like this 88?
But in the eyes of Bayes, this credibility is only 2%.
There is no other reason. The false alarm rate of 5% is very high in the medical profession.
Regardless of the superficial data, we have to believe in Bayesian mathematical conclusions.
The seemingly cold Bayes theorem will gently comfort you:
Don't be afraid, the probability is less than 2%.
Bayesian formula has achieved human trust
Today's Bayesian theory has begun to spread everywhere.
From physics to cancer research, from ecology to psychology,
Bayes' theorem is almost as accurate as the "second law of thermodynamics" in the universe.
Physicists proposed Bayesian interpretations of quantum machines, defending string and multiverse theories.
Philosophers argue that science as a whole can be regarded as a Bayesian process.
In the IT world, the thinking and decision tree of the AI brain is designed by engineers as a Bayesian program.
In daily life, we often use Bayesian formula to make decisions, but we don't notice that this is the "Bayes theorem".
For example, when we go fishing by the river, we don't know where there are fish. It seems that we can only choose randomly. But in fact, we will find a backwater bay area to start fishing based on the Bayesian method and the accumulated experience in the past.
This is to make a subjective judgment based on prior knowledge, strengthen this judgment after fishing, and then choose again next time.
Therefore, when the understanding of things is not comprehensive, the Bayesian method is a rational and scientific method.
Bayesian theory is now recognized mainly from two things:
❶ The author of "Federalist Humanities Collection" revealed the secret
In 1788, the "Federalist Collection" was published anonymously, and the writing styles of the two authors were almost identical. The authors of 12 articles are in dispute, and it is extremely difficult to find the author of each article.
The two statistics professors used a classification algorithm with Bayesian formula as the core. In more than 10 years, they inferred the authors of 12 articles, and their research methods also caused a sensation in the statistics community.
❷U.S. Scorpio nuclear submarine search and rescue
In May 1968, the USS Scorpio nuclear submarine disappeared in the Atlantic Ocean. The military used various technical means to investigate to no avail, and finally had to turn to mathematician John Craven.
The scheme proposed by Craven also uses the Bayesian formula. After searching a certain area, the probability map is corrected according to the search results, and then the search areas with small probability are eliminated one by one. A few months later, the submarine was found on the seabed southwest of the explosion point. Up.
At the beginning of 2014, Malaysia Airlines flight MH370 was lost. The first method that scientists thought of was to use Bayes' theorem, a common method of shipwreck and air disaster search and rescue, to start a regional search.
At this time, the Bayesian formula has become famous all over the world.
Bayes' theorem shows "miracle"
Of course, Bayes' theorem is famous all over the world, mainly in the application of artificial intelligence.
In particular, the technical recognition of natural speech has allowed humans to see the "thinking power" of AI.
The ambiguity of human language can be said to be the most complex and dynamic part of information.
How does the machine know what you are talking about?
In 2020, as long as you see the accuracy of machine translation,
You will also lament that these are simply "miracles", they are much better than most live translations.
Speech recognition is essentially to find the text sequence with the highest probability.
Once conditional probability appears, Bayes' theorem can always come forward.
We use P(f|e) different from the above P(A|B) to explain the speech recognition function.
The problem of statistical machine translation can be described as: Given a sentence e, which of its possible foreign translations f is the most reliable.
That is, we need to calculate: P(f|e)
The right end of this formula is easy to explain:
Those foreign sentences f with higher prior probability and more likely to generate sentence e will win.
We only need simple statistics to get the probability of any foreign sentence f.
With the iteration of a large number of data input models, with the continuous improvement of computing power, and with the development of big data technology, the power of Bayes' theorem has become increasingly prominent, and the great practical value of Bayes' formula has become more and more manifested.
Speech recognition is just one example of the use of Bayesian formula.
In fact, Bayesian thinking has penetrated into all aspects of artificial intelligence.
Bayesian network, the expansion of AI wisdom
Speech recognition has witnessed the ability of Bayes' theorem.
The expansion of Bayesian networks can see a more powerful future of artificial intelligence.
With the help of classical statistics, humans have solved some relatively simple problems.
However, classical statistical methods cannot explain the phenomena caused by complex parameters, such as:
The cause of the tornado, comparison of the smallest possible parameter values of 2 to the 50th power;
The origin of galaxies, the processing of possible nebula data to the 350th power of 2;
The working mechanism of the brain, the possible quantum flow of consciousness to the 1000th power of 2;
Cancer-causing genes, 2 to the 20000 power possible gene map;
……
Facing such an order of magnitude calculation, classical statistics seem to be inadequate.
Scientists have no choice but to find Bayes' theorem for help.
Connect the relevant parameters of a certain phenomenon, and then substitute the data into the Bayesian formula to obtain the probability value. The formula nets form a cause network, namely the Bayesian network, as shown in the following figure:
This is why Bayesian networks are called probabilistic networks and causal networks.
Use prior knowledge and sample data to establish the association between random variables, and then draw conclusions.
One node after another, one probability after another, all come from human prior knowledge. The more effective knowledge, the more shocking the power of Bayesian networks.
Today a vigorous "Bayesian Revolution" is happening in the AI world:
The Bayesian formula has penetrated into the bones of engineers, and the Bayesian classification algorithm has become the mainstream algorithm.
In the eyes of many engineers, Bayes' theorem is the cornerstone of AI development.
AI's way of thinking: no literary and rational points
If you understand Bayes' theorem, you basically understand the way AI thinks.
This is why " big data + algorithms + computing power" constitute the three elements of artificial intelligence.
❶Big data, it is the teacher of AI, it teaches AI what kind of person to be.
❷Hashing power, which belongs to personal ability, the energy needed when AI grows up to deal with problems.
❸Algorithm, the methodology (talent) bestowed by the creator, the better the algorithm, the more effective it will be.
Starting from these core elements, let’s look back at the initial questions:
Does AI really not understand art?
Can it not be "subjectively created"?
Can it not help humanity in art?
The answer is no, there is no difference between artificial intelligence thinking.
It is a science student, a liberal arts student, and an art student.
AI's thinking gene comes from the subjective "Bayes' Theorem". As long as there is good data, the machine can create classic works of art after learning.
Currently, "AI+Art" has become a new trend of thought.
The French art team Obvious created the "Edmond Bellamy" AI artwork based on painting data. The high price of 432,500 US dollars at Christie's shocked the world.
AI can also compose! OpenAI neural network has been able to create any genre. In 2019, "Beauty World" from Australia won the AI version of the "Europe Song Contest" champion. This song is based on the background of commemorating the animals killed in the Australian fire. It uses Eurovision Song Contest songs as big data and handed over to AI to compose the lyrics. .
Newton Rex, a well-known musician, said: Music life is full of creativity. Look at the role of AI in music from a positive perspective. Music education, humans and AI can also join hands.
Will AI say "I think, therefore I am"?
Starting from Bayes’ theorem, engineers believe that artificial intelligence can only be mathematical probability.
Will never produce free will;
People have always believed that AI will never understand human love, hatred, and hatred, just as it does not understand the darkness of night at night.
However, the simulation of art by artificial intelligence has surpassed the discrimination ability of most people.
In the near future, AI may pass the "music Turing Test",
What is the difference between the best music AI and the greatest music teacher?
Maybe there is still a gap between AI and large-scale application in the field of art.
However, Tencent's exploration of using AI to popularize art education is not a good exploration.
The control system of Google's self-driving car;
The AlphaGo system that challenges the last bastion of human wisdom;
Tencent OpenAI's new creation in music generation;
From Bayesian networks to neural networks, AI is becoming more and more human.
All of this is based on the genes of Bayes' theorem.
If AI can create a song, then it can become a great music church.
When Descartes said "I think, therefore I am", it is considered "human awakening".
Will AI ask "Who am I" one day?
If humans want to preset answers in the underlying system.
Then we will set:
You are an "AI with free will",
Or "You are the AI created by humans".
Want to hear more about mathematics and AI? "The Beauty of Formula" created by the Quantum School is here!
brief introduction:
Mankind invented mathematical formulas to describe the vast universe and various life states. The prosperous beauty of the world reflects the simplicity of the symbolic formula. Einstein’s mass-energy equation and Yang Zhenning’s gauge field explored the rules of the ultimate game of the universe; Fermat’s last theorem and Euler’s identities reveal the mathematical world behind the changes in the universe; from Kelly’s formula to Bayes’ theorem, gradually complete Predict human behavior; the Lorentz equations of the butterfly effect and the three-body problem tell us the boundaries of mathematics.
The Quantum School devotes itself to creating "The Beauty of Formulas", which contains 23 universal, profound and practical formulas, writing the brilliant history of geniuses exploring nature and society.
About the Author:
Quantum School: It is an education platform that focuses on the field of natural sciences (mathematics, science and philosophy). Most of the natural science articles published by its public account "Quantum School" have a reading volume of 100,000+. It is one of the top ten popular science education platforms in the country. The platform has launched a series of courses well received by readers, including "The Beauty of Mathematics", "The Beauty of Logic", "The Beauty of Reason", and "The Beauty of Science".
#Welcome to leave a message#
Top three with the most comments
Program Life Hand in Hand【Peking University Press】Sent
The Beauty of Formula
As of 14:00 on September 21