April 19th news, according to the "Guardian" report, talking about the future of robots, the father of recurrent neural network, German computer scientist Jürgen Schmidhuber (Jürgen Schmidhuber) said in an interview, "Future robot Our attention will be like our concern for ants.†He also said that artificial intelligence will surpass human intelligence by 2050.
In a studio behind a warehouse in western Berlin, a group of international scientists are discussing the future of our robots. An engineer from a large European car manufacturer is presenting a cautiously optimistic report on the progress of driverless cars. He explained that robotic cars are learning to distinguish between cars and pedestrians, cyclists and other relatively vulnerable moving objects. Some robot cars are better at distinguishing different dog breeds than humans. "But of course, these are only small progress," he said.
Then, a man with a strong body and wearing a light gray suit came to power. “Not long after, the smartest and most important decision-makers may not be human.†He said, with a pitiful smile on his face that tells young people about growing up troubles, “I, we are not on the verge of another industrial revolution. But it is on the verge of a new type of life, it is more like a big bang."
Jürgen Schmid Hubert is known as the one who will be recognized as the father by the first robots with autonomy. The algorithm developed by the 54-year-old German scientist allows us to talk to our computer and also allows our smartphone to translate Mandarin into English, but he is not very keen on the future. The existence of robots will mainly serve humans. The saying.
On the contrary, he believes that machine intelligence will soon be more than enemy intelligence, and will surpass human intelligence. The heat-resistant robots designed and manufactured by the future autonomous replication robot factory will be closer to the sun's energy than the thin humans. They will eventually colonize the asteroid belts of the Milky Way. Schmid Hob is trying to build a brain for the robot.
After giving a speech, Schmid Hubert explained that in a laboratory in Lugano, the Swiss Alps, his company, Nnaisense, is already developing systems that can be operated like babies, and they have put some small experiments on those systems. So that they understand how the world works. He called this "real AI." The only problem is that they are still too slow -- there are currently only 1 billion neural connections, and the number of neural connections in the human cerebral cortex is about 100 trillion.
“However, our current trend is that our computers can accelerate ten times every five years. Unless this trend ends, it will take only 25 years to develop a recurrent neural network that rivals the human brain. Level intelligence is not a lot of time, such as the intelligence of crows or capuchin."
How many years is it worse? “I think it’s better to be in years than in ten years, but I don’t want to limit myself to a time range of 4 to 7 years.â€
When asked why he was so confident in his timetable, he mentioned the super-light engine. Our discussion suddenly jumped from the big bang to the Neolithic revolution, from the invention of gunpowder to the birth of the World Wide Web. Schmid Houb pointed out that the interval between major events in the history of the universe seems to be exponentially shortened - the interval between the two milestones is about a quarter of the previous interval. According to this rule, machine intelligence beyond human intelligence seems to occur in 2050.
"By 2050, time will not stop, but we will usher in an AI that is smarter than us. By then, it will be meaningless to be obsessed with studying the human biosphere. They will push history to the next stage. Going to places with abundant resources. After a few hundred years, they will establish colonies in the Milky Way."
He called this time pass "omega". The word Omega was first coined by the Jesuit pastor, Teilhard de Chardin, who was born in France in 1888. Schmid Hubert said he likes omega, "because it sounds a bit like 'oh my god'".
Controversy and criticism
The status of Schmid Hubert as the godfather of machine intelligence is not entirely indisputable. As a computer scientist, sometimes his speech sounds inconsistent with scientific principles and is shocking. During his speech in Berlin, he was able to hear complaints from people behind the audience. When Schmid Hubert outlined how the robot would eventually leave the Earth and “leave alone†to explore the universe, a neuroscientist from Brazil interjected: “This is what you have to say? The robot has fun. Algorithm? You are destroying scientific methods in front of all these scientific people! It’s terrible!"
When asked about those reactions, Schmid Hube once again showed the kind of pity. "My papers have been controversial for decades, so I'm used to these common arguments. But many neuroscientists don't really know what's going on in the AI ​​world."
But even in the AI ​​community, Schmid Hubert has many critics. Hearing his name, several professionals in the field of artificial intelligence have pointed out that his work is undoubtedly influential, "even more and more influential," but he is "a bit indignant." Many people feel that his optimism about the speed of technological progress is unfounded and may even be very dangerous. Some people say that Schmid Hobb is far from being the real prophet of the future of robots. He actually pushes artificial intelligence to a Segway-balanced car (the Segway was touted as a technological revolution similar to PC invention, but ultimately The fate of the dismal end.
To understand why Schmid Hubert is considered to be both a prophet and a joke, he must have a deep understanding of his resume. Born in Munich, Germany in 1963, he borrowed a large collection of popular science books and science fiction novels from a library near his home during his adolescence, and he became interested in robotics. His favorite books include Olaf Stapleton's Star Maker, ETA Hoffman's The Sandman, and Lem Stanislaw ) The novel.
He said that his most admired idol is Albert Einstein. “Sometimes, I would think that if I build something smarter than myself or even smarter than Einstein, then I might be more influential.†He received math and computer science at the Technical University of Munich. The degree, the university awarded his professorship when he was 30 years old.
LSTM long and short memory
In 1997, Schmid Hube and his student Sepp Hochreiter co-authored a paper on how artificial neural networks can be improved through memory functions, based on previously obtained information. Interpret the closed loop of the pattern of words or images. They call it long-term and short-term memory (hereinafter referred to as "LSTM").
At that time, AI was in a long period of “cold winterâ€: technology failed to reach the first wave of touting around artificial intelligence, and financing was difficult. In the 1960s, the industry's hope was that machines might be able to understand the intricate world from the top down. If there is any new hot spot in the industry right now, it is a seemingly simple idea: machines may introduce a relative foundation, but they can let them understand the algorithms of the entire complex world from the bottom up.
In 1997, Schmid Houb's paper on LSTM was rejected by the Massachusetts Institute of Technology, but from now on it seems to be one of the main concepts behind the rise of deep learning. In 2015, Google announced that it has successfully improved the error rate of its speech recognition software by nearly 50% with LSTM. LSTM is the system that drives Amazon Alexa, and Apple announced last year that it is using LSTM to improve the iPhone.
If Schmid Hubert was successful, the concept would be more recognized. In 2015, he complained in a rigorous article that three Canadian computer scientists who were honored as superstars in the AI ​​field in Silicon Valley - Jeffrey Sinton (Google), Jahn Leku (Facebook And Joshua Bengio (IBM) - "a large number of mutual quotes of their respective views", but "not attributed to the pioneers in this field."
During his speeches and interviews in Berlin, he repeatedly stressed that the current craze around computer learning is “outdated†and that LSTM has reached that point many years ago. He also felt that Silicon Valley was not that important. He felt that the region was too much dominated by “violent competitionâ€, so that the value produced was not as good as that of European research institutions.
Listening to Schmid Hubert's talk about the future of robots, you may have the feeling that he is constantly talking about high-tech utopias, just to ensure that he will not end up in the situation of Sisto Rodriguez. Strategy. Rodriguez is very influential, but he has not received much attention. Instead, he is a Silicon Valley person who enters the Temple of Honor.
Will robots enslave humans?
Given that Schmid Hobb has a strong interest in science fiction, has he never worried that robots that will become self-aware will one day enslave us and dominate us? He shook his head and said, "We will not be enslaved, at least because we are not suitable for slaves who can directly create robots that are much more advanced than us." As for the imprisoned in The Matrix Humans are used to drive the plot of AI, and he also believes that the future will not happen: "That is by far the most stupid plot. Why do you use human creatures when the power station that allows robots to stock produce much more power? Energy to drive robots?"
But if that happens, will the robot see it as a more efficient way to completely eliminate humans? “Like all scientists, a highly intelligent AI is fascinated by the origins of life and civilization. But this fascination will weaken over time, like many people who don’t know the origin of the world. Generally speaking, The best protection for us will be their lack of interest in us, because the biggest enemy of most species is their own. Their attention to us will be like our concern for ants."
This analogy does not seem to be as compelling as he hoped. Sometimes we can't help but step on the ants? Some people even use chemicals to poison ants that live in one place. "Of course, that's only a small part of the global ant colony. No one seems to want to destroy all the ants on the planet. On the contrary, most people are happy when they hear more ants on the planet than humans. Most of the ants are in the jungles of Brazil."
He quoted Harvard professor Edmund O Wilson's recent controversial claim, "We may be much smarter than ants, but the total weight of humans on Earth is comparable to the total weight of all ants. â€
Will it cause massive unemployment?
Let's not talk about science fiction. Talk about even more pressing concerns, such as the automated revolution that caused massive unemployment. In an article recently published in Nature, AI researcher Kate Crawford and Internet law professor Ryan Calo warned that using robots instead of humans The social chain reaction, a new wave of spurs around intelligent design is creating a dangerous blind spot.
In this regard, Schmid Hubert is not very worried. He said that when the two daughters began to be born in the millennium, he was very clear that the future of robots will come. “What advice do I give to my two daughters? I told them: Your father thinks everything will be fine, even if there may be ups and downs. Be prepared to continue to do new things. Be prepared to learn how to learn.â€
“Humans are always very good at creating jobs that don’t exist. Most people are already doing some extravagant work, just like your work and my work.†He said, “I predict which jobs will disappear and It's not difficult, but predicting what new jobs will emerge in the future is not easy. In the 1980s, who would have thought that in 30 years someone would be a professional video game player or a YouTube star to make millions of dollars?"
“Even the well-respected position in the medical profession will be affected. In 2012, robots began to use deep neural networks to win cancer screening competitions. Does this mean that doctors will be unemployed? Of course not. It just means that The same doctor was only able to treat one patient for a period of time and will be used to treat 10 patients. Many people will be able to enjoy medical research with less money. More human life will be saved in the future, humans The life expectancy will also be extended."
He happily suggested that the unemployment rate will be relatively low in countries with more robots such as Japan, Germany, South Korea and Switzerland.
Will the robot fall in love?
Whenever you try to understand Schmid Hubert's optimistic vision for the future of robots, you will fundamentally encounter a very simple situation. He said that when there are conflicts of interest between the two creatures, they have two ways to resolve the conflict: either cooperation or competition.
When asked about the future robot, besides the fun and fun, whether he will be able to fall in love, he gives a positive answer, saying that "love is obviously an extreme form of cooperation. The emotional life of the robot will It will be multiple and multi-love, not single: there will be various relationships between robots. They will be able to share their thoughts, which humans can’t do now, or only if they are facing each other. It takes a lot of time to do it. The robot will also have a fusion that does not exist between organisms."
If love is really just a strong form of cooperation, why does it make people feel irrational? Why do we have lust? Why do you feel heartbreak? Schmid Hubert replied, "We already have pain sensors, so the robot will feel pain when it hits something. We will eventually figure out the erotic problem. After all, they all belong to the same thing."
Will the enemies appear?
What if a company, such as Apple or Google, has a monopoly on super-intelligent robots that rule the world in the future? Schmid Hubert believes that the kind of enemy state "nothing can happen." In this scenario, cooperation will also be king. "The intelligent central algorithm will be very short. The algorithm that makes the system improve itself may have only 10 lines of virtual program code. We are currently missing only 5 lines of code."
"Maybe we will develop the 10 lines of code in my small company, but in the beginning, even the Swiss bank secrets can't be saved, the code won't stay in my company. They will be leaked. Maybe India An unknown person will think of the code and make it available to everyone."
If it sounds a bit blindly optimistic, it is also because of Schmid Hubert's own experience - LSTM was initially refuted, and he has always distrusted the "competitive" Silicon Valley - certainly let him know that competition can create losers Can also create winners. In terms of personal feelings, his optimism will make people put a warning, but I will feel more comfortable for the most advanced existence of the future by Jurgen Schmid Houber, if he is willing to clarify those Questioning.
At the end of the interview, he apologized: "I am sorry to make you talk to a teenager. But I have been talking about these things since the 1970s and 1980s. The only difference now is that people I started to take me seriously."
Nantong Boxin Electronic Technology Co., Ltd. , https://www.ntbosen.com