Whether strong artificial intelligence can appear, the future of mankind is doomed to grey

This article starts with the public number "fallacies and fallacies" (Wxieshuo).

Recently, whether in Chinese or English, the discussion of artificial intelligence has suddenly become more, for artificial intelligence will not endanger mankind has become a hot topic. Much of the discussion focuses on the likelihood of so-called strong artificial intelligence, and in these discussions from Futurists, far beyond human intelligence, "strong artificial intelligence" will appear in our lifetimes, and by that time human destiny will become. I think people think too far. Before the emergence of strong artificial intelligence, the fate of mankind has been doomed to grey.

Today, these discussions, if able to travel through time, take back to the 80 's our peers, they are not unfamiliar, and will not be intimidated. Even though it is a time when no smartphone, no wireless network, even the commercial internet has not yet formed. But that era of people, almost think that by 2000, everything can be achieved, even strong artificial intelligence will also be achieved.

Looking back to the 80 's, even with today's standards, it is also an era of science and technology, from the beginning of the 60 's various scientific concepts and ideas, to the 80 generation of basic available. Not only today's artificial intelligence basic theory and algorithms in that era have a breakthrough, even today everyone is talking about electric cars, fuel cells ... are in that era of commercial products, today we say the so-called strong artificial intelligence, but also the 60, the 70 's product. The word Moore's Law, also appears in the 60 's, the 80 's mature. Not coincidentally, science fiction was quite flourishing in the 80 's, and there were cyberpunk schools with the main themes of cyberspace and the future. I have always felt that what these futurists are talking about today is no more than the depth of science fiction, especially the cyberpunk genre of science fiction, the near future that they predict is almost achieved.

The "Nerve Rover", published in 1984, is a landmark in the history of science fiction, creating the cyberpunk science fiction genre based on cyberspace, and even setting the foundation for this genre of science fiction, including cyberspace and giant monopolies to control the world, and of course, There is also the ultimate control of the artificial intelligence.

The future of the cyberpunk genre of science fiction is mostly gloomy, which is not the result of a process that produces a higher intelligence, a total enslavement of humans, but the technology and humanity itself.

People often do not realize that, now we have already lived in the scope of the program defined. This does not need the so-called strong artificial intelligence to do, distributed in different systems with a certain intelligence, or no intelligent procedures, has long been an important helper. In most cases, people trust in the outcome of the program more than the person itself.

Even people who know nothing about software technology have a direct feeling in some clues. For example, many years ago, when a variety of systems were not fully networked and computerized, a lot of things, as long as they can convince the terminal operators, he pen, the problem is solved. But to this day, the terminal office staff will often sympathize with you, tell you I know that you have a point, but I do not have the authority to change the system, so I have no way. Most of the time, we are completely trapped by the system, and we can only think of alternatives. These systems are often not intelligent, they are only tough to implement the set of rules, such systems can still regulate people's lives.

A slightly smarter example is the banking and insurance industry. They have more data and can calculate more results based on data. Whether to apply for a loan or to buy insurance, the result is often not a real person, at least not simply by a real person to make a decision, but through a pile of procedures "intelligent" calculation. As for the results of the program calculation is not exactly right, is not fully in line with your situation, the person you are dealing with often do not know that the system is a black box, it made a decision, people follow it.

Smarter, and more commonly, on the internet. Every internet company collects all the data it can reach with a near-greedy attitude. Which button the user clicked on, which page stayed longer, what they shared, what they praised, what was forwarded, what was commented, what was added to the cart, what was paid for ... Everything is recorded. A new generation of Internet companies are committed to providing each individual with customized products, and the pages you see may be completely different from your friends, and even the prices you see are not necessarily the same as what your friends see. These are based on your own behavior, the results calculated by the program, in most cases, there is no human involvement. Only from the experience, this is not a bad thing, after all, according to user custom content, you can make him more comfortable to use, save more time, people also prefer such products.

As it was a few decades ago, today people are talking about strong AI again, and once again think that it will happen in our lifetime, in addition to futurists preaching around, the real reason is that the depth of learning as the representative of unsupervised learning technology has begun to show excellent results. With this technique, the computer looks smart and can get a better result by feedback. While these technologies allow computers to do more, they are fundamentally different from the intelligence that humans have.

Strictly speaking, deep learning is not a new technology, it is still based on neural network algorithm. Or in the 80 's, the neural network algorithm has matured (in fact, the 50 generation), that is the high generation of artificial intelligence input, a large number of scientists, graduate students into this field, unfortunately the effect of long-term can not achieve real usable, finally only a few special scenarios can be successfully applied. By the 90 's, the field had lost most of its attention and funding. In the last 10 years, this algorithm continues to evolve into today's deep learning, and this time they succeed. The most important scientist in the field of deep learning, a team led by Professor Hinton of the University of Toronto, won a great success in the 2012 LSVRC competition, and they used deep learning algorithms far more than other contestants.

Why is it that a method almost abandoned in the 80 has returned to the center of the world? Because the 80 's can complete the theoretical basis, but the real access to practical needs of the data and computational capabilities are not available. Over the decades, with the development of the Internet, chip prices, parallel computing development, the previous difficulties are no longer a problem. In addition to the number of data, and more importantly, because of the popularity of smartphones and various kinds of sensors, we can get more kinds of data than the 80 's, so that we can try in more areas. Under the combined action of these technologies, the new artificial intelligence high investment era has started again.

The danger does begin, but the source of the danger is not the artificial intelligence algorithm, but the more we provide the data, the greater the chance of being limited and guided, and the degree to which the technology of artificial intelligence today can be made very uncomfortable. In the future, the available computing power will increase, and the type of data that people provide will continue to grow, especially with human-related biometric data that will allow humanity to give up more freedom, a trend that is almost impossible to reverse.

From what has happened now. Some sites will show different prices to various users, is not a secret, a few years ago, hotel reservation company Orbitz Experiment so that Apple users can only see more expensive room, said the effect is good, let them earn a lot of money. This is done based on very simple rules that divide users directly into two categories, Apple users and others, and then guide Apple consumers. If some smart, can be in accordance with the user purchase habits, to provide different users with different recommendations, which we are not unfamiliar, in the electric business site as long as there is an account, even if only a few things in the login status, will soon receive a variety of customized recommended mail.

If the computational power is sufficient, let the user "spend a little more money" this goal, but also can do a lot of fine. For example, before the user adds something to the cart and pays for it, can you find a product that is similar in price but has a higher profit margin and is recommended to the user? After the recommendation, can you give him some reasons to give up the goods before, to buy the reason-such as how many people think this is better, or how many of his friends praised it, or even, recently this thing is on sale, is the lowest price in history. After that, the user buys or does not buy, is a feedback, the system collects this kind of feedback, next time lets the recommendation plan become more suitable for this user. Each repetition of this process can help the AI system improve a little bit, and the data generated by one's behavior can be used to influence his friends.

This process can also do better, for example, the user added to the shopping cart but did not buy the goods, if the reduction of 20% will promote his immediate purchase? If not, the price to go back, will make him feel regret, then next time the price reduction of 15%, is not immediately bought? As we networked more data, the basis for analyzing user behavior is more pronounced. Compared to the past, people began to put more and more biological data networking, heartbeat, blood pressure ... With these parameters, the previous process can be changed, when the price reduction of 20%, the user's biological characteristics will not change, such as heart rate acceleration? If the price rises again, how will the biological characteristics change. Can the system analyze what is exciting and what is regrettable from these biological data changes? If the price after the user feel regret, the next price reduction of 15%, it is likely to lead to a purchase. This has been our training program, but the procedure is to train us. After this training, the next time the price reduction of 20%, the user may buy immediately, the user's behavior pattern changed, this is the result of the program training.

This is a relatively intuitive example, and we can speculate on some reasons and rules that can be expressed in language. In practical applications, the AI program is often a black box, which is designed to allow users to buy things at higher margins. As to how to do it, there will be different ways for different people. Even the owner of this black box, even the creator, can hardly exhaust all the possibilities and know exactly what it does to everyone. Of course, until today, these have not come true, but I believe that small attempts should already be in progress. There is no large-scale application, just because the computational capacity is not strong enough, the method of collecting data is not enough. Simply put, it is because it is not cost-effective enough, but I believe that it is not too far from the day it was realized.

More and more of our use of the network, but also provide more and more data to different companies, and the popularity of various sensor chips, but also let us provide more and more kinds of data. In many cases, users are not even aware that these companies are collecting our behavior. An average user may find it difficult to realize that when he opens a webpage to stay still, this is also a feedback. It's hard to imagine opening an app, which button it is, and it's often recorded. These user behavior data is the best material for "feeding" AI programs. The data that scientists in the 80 have struggled to obtain are everywhere today, and users do not find them valuable. Whether it is Google, or a small personal site, when you have opinions about it, there will always be someone to tell you that you do not spend money, people love how to do. This is certainly wrong, users pay data so precious things, how can it be called free? To put it in a few decades ago, a person's behavior patterns and preferences, often only his best friends and family can know, today not only by a variety of companies learned, but also used for commercial purposes. The law now protects only information that is simple enough to correspond to real people (such as identification numbers), which are considered private, and which companies are free to collect and use. Every company insists that it has a privacy policy, and even if we do not consider the possibility of data theft or leakage, it would be enough to say that acquisitions and mergers between different companies are sufficient to make the data scattered around the globe progressively more aggregated.

Today this is still not an obvious problem, although we can not leave the network, but most of the time, we are still in the real world. In the real world, a billboard stands on the roadside, all see is the same content, people do not have the means and online advertising, so that everyone see different ads.

But man is bound to enter the virtual world, as cyberpunk science fiction shows. With the development of virtual reality technology, the boundary between the real world and the virtual world will gradually blur away. Last year Facebook spent a lot of money buying Oculus, an immersive virtual reality company, and now it looks silly, with a big box on its head that people are unlikely to walk around with. Earlier Google glasses had a bigger ambition and wanted to be a device for everyday use. Their experience is far from practical. However, this direction is bound to have the actual available products produced. More recently, it is possible that the retina projection technique described in the rainbow end is the real world of the half virtual reality of various wearable devices. Not to forget that biochemical technology has been developing, neural networking that day early evening.

By that time, we would have had more data to produce and be collected. The reality and the virtual boundaries disappear completely, we still in reality will see through the virtual reality technology for each person custom content. There will be more behavior in the way that the AI program prescribes, and it will come from the heart to think that it is his independent idea. People will become the terminal of the program, contribute data for the program, lead the behavior according to the program, and rely on the program to live and work.

Even today, it is hard to say which ones are our views and which are affected. Many people have this feeling, to see a debate on the pros and cons of the two-party article, one will feel that this side of the right, one will feel that there is a reason. It's not surprising that people are so easily affected, and it's hard to believe what you really think. If the information you see every day is carefully calculated by the program, it is likely to act in a predetermined manner. Designing a model for each person in the world to guide his behavior is no longer a fantasy, technically achievable, and the only thing left is computational power and cost.

In today's mix of programs and people, AI programs already have the ability to regulate our lives, even though they are not truly intelligent and more likely to be beyond human intelligence. These programs are created for different purposes, sniffing out human habits, changing and directing human behavior, and their creators-those companies-are also controlled by them, and it is difficult to leave them. Moreover, even the largest corporate shareholders, in their own lives, are also influenced and guided by other AI programs. Human beings and artificial intelligence are inseparable.

Think of what I've said before about the people who say I can't do anything about the system, he's actually a program that extends in the real world. That is, he does exactly what the program does, and he doesn't have any other permissions, and his job is to interact with the real world and the program. This is because the procedure has not yet been able to directly affect the real world. If one day the robot is mature enough, he will be replaced by a robot. Look around, we already have a lot of jobs in this state. For example, courier, they follow the instructions issued by the program, the program will design the best route for him, they only need to go down, they can play a small space, if the drone mature, courier work can be replaced. Humans gather information and work according to the command of artificial intelligence, into the network and computer terminals, in many discussions, this is beyond the human artificial intelligence, but its implementation in this kind of people and the process of composition of the system has been able to achieve good, a lot of people in the work of the program under the control of the final performer.

In the 80 's, science fiction envisioned a variety of situations, but I didn't expect people to contribute their own data today. Whether it's potential user behavior, explicit photos, sounds, friends, Brand likes and dislikes, and even your own thoughts and emotions, with the help of social networking, people are almost unreserved in their contribution. People in this era worry that their ID numbers are known, but they don't feel private about the information. In fact, the ID card number is not privacy, as long as the need, there are countless ways to get it, but other information, in addition to you and your friends actively hand out, it is difficult to have a way to collect directly. We will soon enter the era of data monopolies, where companies with huge amounts of data can use the data for better profitability, better services, and continue to get more data. Companies that lack data may not even be able to create an available service in the future.

This is far more frightening than a capital monopoly, because capital can be solved by financing, data can only be gained through accumulation, the position of the monopolist is more difficult to shake. So, friends of investment, remember to invest in companies that can collect and generate a lot of data, and the era of data monopolies will come soon.

When it comes to the monopoly of data, all cyberpunk science fiction, there is a set, that is, the future enterprises will replace the government, become human social managers. In another novel, the Avalanche, the U.S. federal government has even fallen into tents scattered around the highway, with the biggest rights for private companies. But this is not the case in the real world that we are experiencing. After Snowden exposed the prism, it is suddenly found that the U.S. government is the largest data owner, specifically, the NSA or the Department of Homeland Security is the real data holders and monopolies, not only through administrative means from the various functions of the data obtained, but also in the hands of almost all companies to request more data, Even to steal data, they monitor people all over the world, even the traditional American allies. In the end, our social form is likely to be the monopoly of government and Enterprise, who, together with artificial intelligence programs of various purposes, regulate each individual's specific behavior, and most people become accustomed to it and know nothing about it. The panorama of the vision that the French thinker Foucault envisioned will be realized in our lifetime, in a form that is actively involved and imperceptible to the monitor.

Real-world threats, such as the rise of IS, can greatly accelerate this process, and everyone could ask themselves, are you willing to hand over more privacy, to monitor ubiquity, to capture the precursors of terrorist activity, or to be attacked? The threat of terrorism will make most people choose the former.

Worst of all, today we are in a time when we can't turn back, we can't stop using smartphones or social networks, and we can't afford to give the data to big companies. If you give it up, you may already be in the actual prison. Just a few years ago, people still used to look at the map out, if today take away a person's smartphone, put him in a strange place, his first reaction is at a loss. As the nerve Rover describes, the way to punish a person in an age that has already been able to connect through the neural network is to destroy his nerves so that he can no longer access the web, rather than destroy the body. Or, as the end of the Rainbow describes, if you can't use high technology, it's basically a wreck. From all the circumstances now, human beings will run without hindrance to this gloomy future under control, where all people live within the program's planning, whether they are the owners or users of these systems, and all trust in the outcome of the program is greater than their trust, and human beings are extremely dependent on the system. When this day comes, it is difficult to imagine what progress mankind will make, everything is safe, stable, and no accident. Like the human battery in the matrix, live in the real world of batteries. By this time, no matter how far the future machine can produce beyond human wisdom, human beings are no different from destruction. Humans have created a system that puts all human beings in invisible cages, a situation that is far more likely to be realized than the possibility of strong artificial intelligence.

Trying to stop all this is even more impossible, and even now trying to stop the US government's collection of data is likely to be considered treason. This was a time when the unexpected case of Snowden, with his authority, was used, but he took advantage of it to make a bad deal for the whole system. In the future, everyone will live under the rules of the system, and all the data is monitored and computed, and when the future Snowden begins to collect those documents, his abnormal behavior will be discovered by the system and suspended immediately. Perhaps the only hope for humanity rests with a variety of hackers looking for loopholes.

Finally, a black technology, Canadian D-wave even made the first commercially available quantum computer, of course, academia has been arguing that it is not a real quantum computer, because it can only be used to calculate simulated annealing algorithm. But even this algorithm is useful, Google bought a computer that is said to be used for image recognition. Image recognition and voice recognition are the technologies that blur the boundary between the real world and the virtual world. Today, the computer has been able to from the camera's vision, more accurate tracking a face, but limited to the computational capacity, not unscrupulous tracking everyone, the future, everything is possible.

Resource notes:

Title Figure screenshot from the movie "Silver Wing Killer" Blade Runner, although the original is 1968, but the film is 1982, is also the 80 works.

"Nerve Rover" William Gibson original 1984 published, now has a good Chinese translation, Denovo translation, strongly recommended.

"Rainbow End" Vernor original published in 2006, has a Chinese version. It has been argued that Vernor's early works, "real" and "The Nerve Rover," are cyberpunk, because the former is much earlier than the latter. But I still vote for the latter one. The reason is not written down here.

"Avalanche" Stephenson original published in 1992, has a Chinese version.

(Responsible editor: Mengyishan)