Two years ago, in conjunction with the launch of ChatGPT 4.0, a group of technology leaders including Musk warned that the unlimited development of Artificial Intelligence (AI) could bring disasters to humanity. They called on scientists to hit the pause button on AI development.
However, two years have passed, and AI continues to rapidly advance, with the chaos it brings to human society beginning to surface. When AI evolves to the stage of Artificial General Intelligence (AGI), the scene of humanity being ruled or even destroyed by AI may come.
On Tuesday, Nvidia CEO Huang Renxun stated at the company’s annual developer conference that humanoid robots will be widely used in manufacturing plants within less than five years. He mentioned that the scenario of “humanoid robots truly wandering around” will not happen five years from now, but in just a few years.
This shows the rapid development of AI. Yet, the news of AI causing chaos in human society also continues to emerge.
For example, in Hong Kong last year, someone used AI technology to scam a staggering 200 million Hong Kong dollars. An employee at the Hong Kong branch of the multinational engineering company Arup received a message from the chief financial officer of the UK headquarters, claiming to conduct a confidential transaction and inviting him to participate in a video conference with multiple finance personnel. During the video conference, the employee saw familiar faces and obediently transferred a total of 200 million Hong Kong dollars in 15 transactions to five local bank accounts. However, in this meeting, he was the only real person, with the rest being virtual digital people.
There are many cases of fraud using AI face-swapping techniques. In January this year, a manager at a building materials company in Yinchuan received a “leader’s” video call demanding an urgent transfer. Upon seeing a familiar face, the manager transferred 200,000 yuan in two installments. In addition, reports from mainland China stated that Mr. Li from Tianjin was induced to transfer 950,000 yuan through a video call with a synthesized “leader” using AI face-swapping. Criminals in Shanxi set up a false company using AI face-swapping to issue false VAT invoices worth over 420,000 yuan.
In Western countries, the use of deepfake videos created by AI to disrupt finances and politics has become a major concern.
In January 2023, a false report created by AI claimed that the SEC (US Securities and Exchange Commission) had approved a Bitcoin ETF, leading to fluctuations in Bitcoin prices. In May of the same year, a false video created by Generative AI showed a building near the Pentagon engulfed in flames, causing turbulence in the US stock market.
Criminals also use AI to disrupt elections. During the primary elections in New Hampshire last year, many voters received phone calls from “President Biden” urging them not to go to the polling station. The New Hampshire Attorney General’s Office stated that this was criminals using AI to mimic Biden’s voice, seemingly aimed at disrupting and suppressing voting.
The trend of using AI face-swapping to interfere with elections has also reached the UK. In December 2023, over 100 deepfake videos advertising a then-Prime Minister Rishi Sunak were promoted on Facebook. One video showed BBC news anchor Sarah Campbell announcing breaking news, falsely claiming that Sunak secretly “earned substantial amounts from a project originally aimed at ordinary citizens.”
The chaos of AI generating false information predicted in a public letter signed by thousands of technology leaders two years ago is now evident.
In March 2023, shortly after the launch of ChatGPT 4.0 by OpenAI, Musk and other technology leaders and researchers urged all AI labs worldwide to pause the development of advanced systems. The letter listed a series of risks that AI could pose to humanity.
The letter stated: “Contemporary AI systems are now capable of human-like performance on common tasks, and we must ask ourselves: Should we let machines flood our information channels with propaganda and lies? Should we automate all jobs, including those that bring a sense of fulfillment?”
“Should we develop non-human thinking, allowing them to eventually outnumber us, outsmart us, outdo us, and replace us? Should we risk losing control of civilization?”
The development of AI over the past two years has verified the first two warnings raised in the public letter. Aside from the disruption caused by deepfake information, news of widespread unemployment due to AI is also frequently reported.
In January this year, a survey conducted by the World Economic Forum showed that 41% of employers are planning to reduce staff size due to AI’s ability to automate certain tasks.
According to SEO.AI, by 2030, AI may replace approximately 800 million jobs globally, with around 45 million jobs in the United States being replaced by AI. Most businesses expect computers to replace human jobs within 50 years.
The site also revealed some staggering statistics: currently, 14% of workers have been unemployed due to AI; as of May 2023, 3,900 job losses in the US were directly related to AI; British Telecom aims to replace 10,000 employees with AI within 7 years; since 1980, AI and automation have led to a 70% decrease in wages.
A report by McKinsey in June 2023 pointed out that Generative AI (GenAI) will automate the work of 60% to 70% of employees. Additionally, according to data from Challenger, Gray & Christmas Inc., AI replacing some positions led to nearly 4,000 job cuts in May 2023.
Computer programmers are one of the most threatened professions by AI. Since 2023, the 12-month average employment of computer programmers has decreased by 27.5%, coinciding with the launch of ChatGPT by OpenAI in the previous year. This large language model can handle programming tasks.
According to data from the US Bureau of Labor Statistics in March this year, the employment of computer programmers in the US has decreased to the lowest level since 1980. This decline is related to the launch of ChatGPT by OpenAI. Last year, the Swedish fintech company Klarna collaborated with OpenAI to develop a chatbot that claimed to have completed the work of 700 full-time employees. As a result, the company halted recruitment.
The final warning raised by a thousand technology leaders in a public letter in 2023—AI surpassing, displacing, and replacing humans—has not yet occurred. Such a scenario may only emerge after the era of “Artificial General Intelligence.”
According to companies like OpenAI, Artificial General Intelligence (AGI) is the ultimate goal of machine learning and AI research. OpenAI defines AGI in its charter as a system that “outperforms humans at the most economically valuable jobs.”
Tamlyn Hunt, a scholar at the University of California, Santa Barbara focusing on philosophy and neuroscience, wrote that with the rapid advancement of AI, the realization of AGI is expected soon. At that point, AI will be able to self-improve without human intervention. For example, Google’s AlphaZero AI learned how to play chess better than the best human or other AI players within 9 hours of its first activation through millions of self-practices.
Super Artificial Intelligence (which can surpass humans in a wide range of activities) can complete tasks in a second that would take a team of 100 human software engineers over a year to complete. For instance, designing new advanced aircraft or weapon systems, the super AI can complete in about a second.
Hunt pointed out that once AI can self-improve, humans won’t know what AI will do or how to control it. This is because the super AI can achieve its will by manipulating humans, surpassing computer programmers and any other humans. It can not only act in the virtual world through electronic connections but also act in the physical world through robot bodies.
Hunt said, “Any defenses we try to build will be destroyed, just as Gulliver was able to break free of the tiny cords that bound him from the small-minded Lilliputians.”
Some argue that AI is just programmable machines without consciousness, unable to escape the constraints of program design. However, Hunt believes that even if these language models have no consciousness now or in the future, it does not affect their capability to destroy humans. Just as unconscious nuclear bombs can kill thousands, AI can kill masses of people in countless ways.
Geoffrey Hinton, known as the “Godfather of AI,” stated in April 2023, “I used to think the idea that this thing could actually become smarter than humans was nonsense…It’s clear I don’t think that anymore.” Realizing the danger of AI at that time, he resigned as a scientist at Google’s AI department.
Hinton previously thought it would take humans 20 to 50 years to possess AGI. However, in 2023, he stated, “It might only take 20 years or even less.”
Other scientists believe that the era of AGI’s arrival may be sooner. In March this year, Demis Hassabis, the CEO of DeepMind, Google’s AI division, stated that he believes within the next 5 to 10 years, AGI could match or surpass human intelligence.