What! Artificial is not intelligent,a review of “Artificial Unintelligence (Published in 2018–4)” book
Now, artificial intelligence has now become the mainstream of technological development, and it is affecting everyone’s life, as if overnight, everything has begun to get involved with artificial intelligence. However, is artificial intelligence really completely intelligent? This is not the case. The leader of artificial intelligence technology is human, and if someone participates, there must be subjective consciousness. If we do not solve human problems, then technology will not solve real social problems after all.
It has been more than two years in the field of artificial intelligence education. As artificial intelligence projects are gradually increasing, the work content is gradually increasing.
For example, in addition to making products, they are also responsible for the landing training of artificial intelligence classrooms. The larger training is the Wanzhou District project. The two intensive trainings included nearly 300 teachers in more than 50 primary and middle schools.
Before telling teachers about educational products, I usually talk about the value of artificial intelligence. Just as we learn the “Eight Honors and Eight Disgraces” and “Disciple Rules” from a young age, we must not only learn knowledge, but also learn the concept of value, and establish a correct outlook on life and values.
The same is true as an AI product manager. Before communicating with customers, let him establish the correct values and form a “critical” thinking to help him recognize the reality of “artificial unintelligence” and stop trying to solve the problem. .
Next, let’s use critical thinking to talk about the value of artificial intelligence.
What is my opinion?
I believe you have heard too many topics about artificial intelligence, such as: artificial intelligence will change the world, artificial intelligence will make most people unemployed and so on. In fact, the future in which data and algorithms control everything is very far away from us. We all overestimate artificial intelligence.
I recommend a book “Artificial Unintelligence (Published in 2018–4)” to everyone. The author of this book is Broussard, an artificial intelligence expert at Harvard University.
The reason why he wrote this book is not just to show everyone the current situation of artificial intelligence development, to tell everyone that artificial intelligence is not as good as expected, but because people rely too much on computer algorithms, it has brought a series of society Problem, so what he did is to lead everyone to establish correct artificial intelligence values.
How do I demonstrate my point of view?
Part 1: What kind of social problems will overestimating artificial intelligence produce?
Part 2: How difficult is autonomous driving technology?
Part 3: What is the fundamental contradiction in the social application of artificial intelligence at this stage?
1. Overestimating the social problems that artificial intelligence will produce
There is a sentiment called “technical chauvinism”, which believes that all social problems can be solved by technology, especially computer technology. The author believes that the so-called new technologies, including artificial intelligence, do not actually solve real social problems.
He cited an example in the book, the issue of the distribution of teaching materials in the United States.
What problems will arise in the distribution of teaching materials, because the education systems in the United States and China are different. Chinese textbooks are published uniformly by the state, and the prices are cheap. The American textbooks are published by private companies and the prices are high. For example, a textbook for middle school students costs $114.75.
The government’s education funds are often limited, and sometimes the textbook fees allocated to students may only be more than 30 US dollars. As a result, there is a big gap.
Moreover, it is not feasible to buy teaching materials by yourself, because without this channel, there are no pirated copies.
Because the textbooks for primary and secondary schools are published by the three major companies, they are directly supplied to schools, and the exam questions are also issued by these companies. In other words, you use the textbooks of this company, and then use the exam questions of these companies, one-stop service, and a business worth billions of dollars a year.
Of course, some problems that seem to be solved by technology can’t be solved. For example, sometimes the fees are in place and the books are bought, but they are placed in the school’s warehouse and not delivered to the students.
He found the district board of education, and the district board said that they have a central database with data on teaching materials ordered by schools.
Taking this data over, the results show that the teaching materials and the number of students ordered by many schools are not correct.
Because of the limited funding and insufficient manpower, some principals need to manage the teaching materials themselves, and some principals are unwilling to use the database system, so they make a table themselves, which is difficult to share with others, and there are a lot of problems in the end.
No matter how advanced data technology is, the initial data entry is also done by people. If the principal is not willing to enter data at all, the teacher does not know how many teaching materials his class needs. What use is the advanced technology?
The nature of education is a chaotic, dynamic system that varies from person to person, while the public education system is a unified, standard, and preferably unchanging system.
Gates once made standardized syllabus and teaching materials, but the level of students in different districts is very different. It is unreasonable to use one standard and the school has no freedom. Therefore, computer artificial intelligence cannot play a very good role in the issue of teaching materials.
Because computers and artificial intelligence are good at solving engineering problems, engineering problems require well-defined, and complex social problems cannot be well-defined.
2. How difficult is autonomous driving technology?
Artificial intelligence at this stage has a fundamental weakness, that is, it is highly dependent on data. It is a summary of past experience. They have no way to predict “unseen” things.
It is suitable for the most common, simplest, and constant application scenarios. Once you encounter a changing scene, you will face various problems, such as “self-driving cars.”
Level zero means that there is no automation at all, that is, people drive.
So far, no one has exceeded the level two. Some experts believe that level five autonomous driving will never reach it, because artificial intelligence cannot handle accidents.
For example, you see a disabled person in an electric wheelchair chasing a duck in the middle of the road. The duck runs in a circle, and she also runs in a circle.
Can this situation accurately predict the course of action of these people at once? Autopilot relies on lasers to hit all kinds of things and then reflect them back. However, if it is snowing or raining, the laser machine may hit snowflakes or raindrops and reflect on them, making a greater misjudgment of surrounding objects.
In addition, autonomous driving also involves an ethical issue. If your brakes fail and there is a group of elementary school students in front of you, and a tree, do you choose to hit a group of elementary school students or a tree?
We said that artificial intelligence relies on data accumulation. Google has been accumulating data, and algorithms are readily available. The real value is data. Whoever has the data will have a market for autonomous driving technology.
Artificial intelligence also has great limitations: first, it is not safe; second, it is immoral; third, it cannot promote business equality, it will only make powerful companies stronger.
3. What is the fundamental contradiction in the social application of artificial intelligence at this stage?
The United States has developed an algorithm for criminals, called COMPAS, specifically to judge the likelihood of a criminal committing a crime again. There is a scoring standard, from 1 to 10. The higher the score, the higher the chance of crime in the future.
The principle of scoring is also to use big data and to examine several key indicators based on past experience. The algorithm gives a person with 1 point, the crime rate is only 22%, and the person with 10 points has a crime rate of 81%.
However, some people have filed accusations of racial discrimination against this company. We all know that the predictions are probabilities. Even if a score of 5 or more is scored, the prisoner will not necessarily commit a crime again. If a prisoner clearly does not commit a crime again, he beat High score, then it is tantamount to being wronged.
For those who did not commit any crimes later, the probability of blacks getting high marks is 42%, while that of whites is only 22%, which means that the algorithm still discriminates against blacks.
In fact, this contradiction is common in life. For example, in Shanghai, where there is basically no patriarchal patriarchal city, girls’ test scores are better than ordinary boys.
Then, if this rule is to improve the overall performance of the school, as many girls as possible are admitted when entering the first year of high school, isn’t this discrimination against boys?
This is the essential flaw of all decision-making based on experience. No matter how powerful artificial intelligence is, as long as it is based on experience, as long as the prediction is less than 100% accurate, someone must be wronged.
The Ugly Duckling Theorem says that all classifications are subjective, and if there are classifications, there will be discrimination. Artificial intelligence provides great convenience for our decision-making, but society is still this society, mathematics is still the same mathematics, artificial intelligence cannot change the nature of the problem.
Of course, when we talk about technological progress, we usually hear good news. For example, drones can be used to deliver goods in remote areas to solve the problem of difficult delivery. In rural areas, farmers can watch TV to learn technology, solve the problem of limited educational resources, and so on.
Technology brings some convenience to people, but if you want to solve a social problem in an all-round way, it is a bit difficult. Even if it is a seemingly successful practice, there are not many unexpected situations behind it.
Therefore, the emergence of artificial intelligence technology does often make things more efficient, but it is people who write programs, people who enter data, and people who use the system. As long as someone participates, there is a subjective consciousness.
Therefore, technology cannot solve real social problems if it does not solve human problems.