Per reporter: Wen Qiao Per editor: Lan Suying
With the rise of ChatGPT, there are endless talks about the risks of AI. On May 30, local time, the non-profit organization "Center for AI Safety" released a joint open letter on its official website, saying that reducing the risk of extinction caused by AI (to humans) should become a global priority, on the same level as other large-scale risks affecting society such as pandemics and nuclear war.
In this context, there are more and more calls for regulation of the AI industry. "Now is a good time to start thinking about superintelligence regulation – the AI systems of the future will be more powerful than AGI." Altman et al. said.
In this vigorous wave of technology, how should humanity respond? How can governments around the world prevent AI from getting out of hand?
The Daily Economic News interviewed Jürgen Schmidhuber, the internationally renowned "father of deep learning", Francesca Rossi, global head of IBM AI ethics, and Yaser S.Abu-Mostafa, professor of electrical engineering and computer science at the California Institute of Technology, to discuss how humans should build a world where humans and machines coexist.
On June 1, 2023, people were looking at a server with a Grace hopper superchip CPU for generative artificial intelligence Visual China Graph
"It's not just an industrial revolution"
Since the birth of ChatGPT, humans have been amazed by the power of AI, and many even believe that the scenes staged in "Terminator" and "The Matrix" may become a reality in the future - silicon-based creatures will dominate carbon-based organisms.
Yuval Noah Harari, author of A Brief History of Humanity, also recently gave a talk hinting at the possibility that "AI could change the meaning of ecosystems, because in the 4 billion years that Earth has existed, ecosystems have only contained organic life forms." And now, or in the near future, we may see the first inorganic life forms, or at least the emergence of inorganic agents. ”
He said that in the sci-fi scenario, there are two prerequisites for AI to pose a significant threat to humans, that is, AI must become sentient, develop consciousness, feelings and emotions, and must be good at navigating the physical world. However, the current situation is that AI does not need consciousness to threaten the survival of human civilization.
In Harari's view, one of the most important aspects of the AI revolution we are currently in is that AI is mastering language at a level that exceeds the average human ability. By mastering language, artificial intelligence is gaining the master key to the door to all institutions, and it can be said that AI has cracked the operating system of human civilization. "By mastering language, AI can also build deep and intimate relationships with people, harnessing the power of intimacy to influence our views and worldviews." He said.
"In the long run, AI will change every aspect of our civilization. Everything will change, and this is not just another industrial revolution. Jürgen Schmidhuber, known as the "father of deep learning" and director of scientific affairs at the Swiss AI laboratory IDSIA, told the Daily Economic News.
"Every 5 years, the cost of computing decreases by a factor of 10. Based on this exponential trend, it is a preliminary prediction that there will be 'cheap' computers in the 21st century, with a thousand times the original computing power of all human brains combined. Soon, millions, billions, and trillions of such devices will emerge, and almost all agents will be able to surpass the human brain. This is something new that will eventually surpass humans and even beyond biology. He further said.
The LSTM (Long Short-Term Memory Network) proposed by Schmidhuber in 1997 is now widely used in Google Translate, Apple Siri, Amazon Alex and other applications, and is one of the most commercially commercialized technologies in the field of deep learning; His Transformers architecture with linearized self-attention was the basis for the later birth of ChatGPT.
Regarding the role of AI in human society, Yaser S. Abu-Mostafa, an international authoritative scholar of machine learning and now a professor of electrical engineering and computer science at Caltech, said, "AI can and should continue to be subordinate to humans, which will be of great benefit to us in the long run." In medicine, for example, AI is making revolutionary advances that will have a profoundly positive impact on our lives and our quality of life. Abu-Mustafa is also the founding project chair of the Neural Information Processing Systems Conference, which is the leading authority in the field of machine learning.
Schmidhuber is also convinced that "individuals and organizations will benefit from the rapid development of AI in many ways. ”
What role will AI play in the future?
In the broader discussion, most people's fear of AI comes more from the surging "AI instead of artificial" theory.
In March, OpenAI and OpenResearch, a nonprofit research experiment, and the University of Pennsylvania, published a paper that said about eighty percent of the U.S. workforce is likely to have at least 10 percent of their work tasks affected by the introduction of large language models, while about 19 percent of workers may see at least 50 percent affected. Goldman Sachs also said in a previous report that the AI revolution can replace 300 million white-collar jobs in Europe and the United States, and the most affected are legal and administrative jobs, while architects and engineers are also in danger of being eliminated.
It is worth noting that the efficiency of all walks of life has been greatly improved by also benefiting from the development of AI. Taking cross-border e-commerce as an example, some respondents told the "Daily Economic News" reporter that with the help of AI tools such as ChatGPT, it can complete things that previously cost tens of thousands of yuan in 10 minutes, and in March ~ April this year, its company's performance increased by more than 200% compared with the same period last year.
The researchers' data also illustrates this. According to a year-long study of customer service staff at a Fortune 500 software company by Stanford University and MIT, generative AI improved productivity by an average of 14 percent, while new employees with new skills benefited the most, increasing efficiency by 35 percent.
There are gradually more different voices on whether AI can replace human labor. Deutsche Bank's recent report said that history proves that AI ultimately creates rather than destroys jobs, and that more jobs are created under the AI wave than it replaces.
"I said in the '80s that it's easy to predict which jobs will disappear, but it's hard to predict which new jobs will be created. Countries with higher robots per capita have relatively low unemployment rates because many new jobs are created [with the help of new technologies]. Schmidhuber told reporters, "30 years ago, who could have predicted that so many people would now be video bloggers?" ”
The Economist's recent article titled "Your Job May Not Be Affected by AI" echoes this sentiment. In 2013, Oxford University published a widely cited paper suggesting that automation could wipe out 47 percent of U.S. jobs over the next decade or so. But what actually happened was a surprise: the average unemployment rate in developed countries has roughly halved over the past decade, the share of working-age people in employment is at an all-time high, and countries with the highest penetration of automation and robotics, such as Japan, Singapore and South Korea, have the lowest unemployment rates.
When talking about the role that AI will play in human society in the future, Francesca Rossi, IBM academician and global head of AI ethics, told the "Daily Economic News" reporter, "AI should be designed to enhance rather than replace humans." ”
"It's true that AI will have a temporary negative impact on a lot of work, and the impact will happen relatively quickly, so we need to have a safety net to minimize the pain of the transition." Abu-Mostafa said.
How can human-machine coexistence be achieved in the future?
So, how should the world build this safety net during this transition period?
According to Francesca Rossi, concerns about AI often stem from a lack of understanding of what's happening inside the "black box": "The development and application of AI systems must be rooted in ethics, which means prioritizing responsible training, cooperation and oversight." As AI technology continues to be popularized and used, it's normal for people to still be wary of the technology, but the reality is that the benefits of AI are clear. Only by putting ethics at the heart of AI development can society learn to trust it. ”
"When deploying AI, the first thing to consider is whether product safety and quality precautions currently applicable to various products can also be applied to AI products. In addition, an agency similar to the FDA is needed to approve AI products and ensure that they are safe, unbiased, and that their 'side effects' (such as privacy leaks) are within acceptable standards. Abu-Mostafa told the Daily Economic News.
Privacy risks are indeed a widely questioned issue in AI deployments. In April, Samsung was exposed to three leaks of company secrets less than 20 days after introducing ChatGPT.
Schmidhuber admitted to reporters that "privacy surveillance and loss may be an inevitable consequence of an increasingly complex society." The more complex the structure of a superorganism such as cities, countries, and corporations, the more comprehensive the information it gathers. Coronavirus tracing to reduce the spread of the virus is a recent example. ”
In the wave of revolution led by AI, globalized and rapidly developing data is the most valuable, and it is even regarded as "oil in the AI era". So how can we manage data more effectively and ensure data security?
Schmidhuber believes that data ownership can help solve this problem. "If the data is really 'new oil,' then it should have a price. For now, however, major surveillance platforms do not compensate users for the loss of data and privacy. But in the future, we may see attempts to create efficient data marketplaces where the true value of private data can be seen through the interplay between supply and demand. He told the "Daily Economic News" reporter.
For ordinary consumers, Abu-Mostafa believes that after AI products are rigorously tested for safety, effectiveness, fairness and other criteria, consumer trust will gradually be established.
With the widespread use of AI in the human world, how to achieve a balance between human-machine coexistence?
Schmidhuber told the Daily Economic News, "Whatever balance we reach at some point, it will have to change soon because AI will continue to learn to do new things that were previously only possible for humans." Therefore, the balance between automation and human engagement is constantly changing, and humans should also be ready to learn new things. Just like 200 years ago, in many countries, 60% of the population worked in agriculture. Today, this data is only 1%~2%. ”
Daily economic news