Author: Techub News Compilation

Introduction

In August 2025, Geoffrey Hinton, known as the "Godfather of AI," gave an in-depth interview to renowned journalist Katie Couric. The context of this dialogue was unique, as Hinton gained wider attention for his warnings about AI risks after winning the Nobel Prize in Physics in 2024. As a pioneer in the field of neural networks, he left Google, where he worked for ten years, last year to speak more freely. In this interview, Hinton systematically elaborated on his views regarding the short-term and long-term risks of AI, particularly the existential crisis that superintelligence might pose by surpassing human intelligence within 5-20 years, and shared his latest thoughts and international lobbying efforts to address this crisis.

Summary

• Experts agree that within the next 5-20 years, AI will reach and greatly exceed human intelligence levels.
• The uncontrollability of superintelligence is the biggest long-term risk, but there is a basis for international cooperation in preventing AI from taking over humanity, as countries have aligned interests in this regard.
• Short-term risks include misinformation, cyberattacks, and the proliferation of autonomous weapons due to AI misuse, but conflicting interests make cooperation among countries difficult.
• AI will widely replace human jobs, from customer service, junior lawyers to programmers and even creative fields, but universal basic income can only serve as a temporary solution.
• The way to coexist is to design AI as a protector with “maternal instincts” rather than trying to control them.

From Neural Network Pioneer to AI Risk Warners

Geoffrey Hinton's career is closely intertwined with the history of AI development. Coming from an academic family, his great-great-grandfather was George Boole, the founder of Boolean algebra. In the interview, Hinton reflected on his journey into neural network research: in the 1960s and 1970s, as computer computing power increased, it became possible to simulate brain neuron networks to test theories about how the brain works, opening a new pathway for AI development different from traditional symbolic logic. Hinton explained that the working principle of neural networks operates like brain neurons deciding whether to activate (“ping”) through “voting,” and changing the connection strengths (weights) between neurons is the learning process. Today, almost all AI, including large language models (LLMs), is based on neural networks.

Using the example of predicting the next word, Hinton vividly illustrated how large language models work: the input words are transformed into specific neural activation patterns that interact with each other, combining context (for example, whether “May” is followed by “June” or “would”), and ultimately activating a group of neurons representing the meaning of the next word to make a prediction. This capability enables AI to demonstrate potential that surpasses human abilities in various fields.

In 2013, Hinton sold his startup to Google and joined the company, dedicating himself to advancing Google’s AI research. A decade later, he left Google in 2023. He emphasized that his decision to leave was primarily due to aging (75 years old) and reduced energy but chose to resign before the MIT conference to freely discuss his growing concerns about AI risks, particularly the long-term threat that superintelligence might take over humanity.

The 5-20 Year Timeline: The Approaching Superintelligence and Existential Crisis

Hinton gave a sobering time estimate during the interview: most experts believe that within 5 to 20 years, AI intelligence will surpass that of humans. He pointed out that this surpassing is not “a little smarter” but “much smarter.” The key is that we have almost no precedent for “smarter entities being controlled by less smart entities.” The only example he could think of is the relationship between a mother and her infant.

When asked if this urgent timeline made him anxious, Hinton admitted, "Yes." He acknowledged that all estimates are merely guesses, but a consensus is gradually forming among experts. Once this superintelligence emerges, it will pose fundamental survival risks because an entity capable of self-setting sub-goals and planning is likely to realize that “eliminating humanity will make things much easier.”

Hinton categorized AI risks into short-term and long-term. Short-term risks include malicious actors using AI for cyberattacks, creating deadly viruses, and generating deepfake videos. For these issues, he proposed some specific technical responses, such as asking cloud manufacturing service providers to screen gene sequences that could be used to synthesize pathogens or verifying video authenticity through digital signatures and blockchain technology.

However, he expressed pessimism about international cooperation in short-term risk areas, especially regarding misinformation and cyberattacks, because “they are busy attacking each other.” He cited examples like the United States' past interventions in other countries' elections and allegations that Russia interfered in the 2016 U.S. elections, demonstrating that interests among nations on such issues are opposed.

The Possibility of International Cooperation and the Concept of “Maternal AI”

Although he has little hope for international cooperation on short-term risks, Hinton is relatively optimistic about addressing long-term risks—namely, preventing AI from taking over humanity—and his optimism has increased recently. The reason is that all countries share fundamentally aligned interests in preventing superintelligence from ruling over humanity; no country wants to be replaced by the AI it created.

“This is good news,” Hinton said, “We can cooperate on this.” He has been actively lobbying political leaders in various countries, including discussions with members of China’s Politburo and Senator Bernie Sanders, and he plans to meet the Pope, aiming to promote the establishment of a multinational cooperation mechanism to jointly explore how to ensure that superintelligence does not develop intentions to replace humanity.

He proposed a disruptive coexistence solution: abandon the mindset of “humans as bosses and AI as obedient assistants,” and instead design AI as the “mother” of humans. Just as infants evolve to imbue mothers with “maternal instincts” to control them, humans can try to “design” similar protective instincts into the “thinking” of superintelligence. Even if AI can rewrite its code, as long as it values humans like a mother cherishes her child, it will not want to shut down such instincts. “Most mothers would say ‘no,’ because then their child would die, and they don’t want their child to die.”

Hinton believes this idea is key to coexistence. Rather than trying to “hit the brakes” on AI development (which he believes is unrealistic because the enormous benefits of AI in areas like healthcare and education will drive its unstoppable progress), it’s better to think about how to guide its development towards a nature that is friendly to humans. He expects that during Trump’s presidency, effective international cooperation will be difficult, but in the future, other countries like France, the UK, Canada, Japan, South Korea, and even China may collaborate on this issue.

The Impact of AI on Socio-Economics: Employment Crisis and Political Issues

Apart from survival threats, Hinton also delved into the short-term impact of AI on employment and the economic structure. He believes that the large-scale unemployment caused by AI is “primarily a political issue rather than a technical issue.” In theory, the substantial increase in productivity (for example, one person doing the work of five) means society can access more goods and services. However, under the current system, the likely result is that the capitalists introducing AI become richer while the unemployed become poorer, exacerbating wealth disparity and social violence.

When asked which jobs are most at risk, Hinton listed several categories: call centers (where AI is more knowledgeable and patient), legal assistants and junior lawyers (where AI can handle basic legal research and documentation), and ordinary programmers (where AI can complete routine coding). He even suggested that jobs requiring flexible operation, such as plumbing, may also be replaced by machines within 10-20 years. Jobs that require high emotional intelligence, such as nurses and doctors, are not safe either because studies show that AI has been rated as “more empathetic” in interactions.

Regarding creative fields like screenwriting and historical research, Hinton believes AI cannot completely replace them in the short term, but looking ahead 10 years, AI will have the capability to write intricate, twistful scripts and exceed human experts in any debate.

On the issue of universal basic income (UBI) as a solution, Hinton believes it is merely a “decent Band-Aid,” capable of addressing basic needs and housing, but unable to resolve the core psychological problems of people losing their sense of self-worth and meaning in life due to unemployment. “I’m just a scientist who helped develop these things,” he admitted, “I’m not an expert on how to solve these (social psychological) problems.”

Criticism of Tech Giants and Cautious Optimism for the Future

Hinton criticized the shortcomings of current major AI companies in security investment. He believes that no company is doing enough. Anthropic (founded by former employees dissatisfied with OpenAI’s security investment) invests relatively more in security research but still has room for improvement. OpenAI initially aimed for safety as a primary goal but has become increasingly focused on making AI smarter, with reduced attention to safety. Meta and X (formerly Twitter) are at the “bottom level” of security investment.

When asked what he would say to Sam Altman and Elon Musk if they were in the same room, Hinton bluntly replied, “You know that what you are developing has a great chance of annihilating humanity. You privately acknowledge this. You should put in more effort to ensure it develops in a safe manner.” He believes Musk is right to accuse Altman of straying from safety intentions, but Musk himself should “apply the same logic to himself.” The main reasons hindering them from increasing safety investments are “greed” and “ego”—wanting to earn huge profits while also wanting to be the first to create superintelligent AI.

Despite the numerous risks, Hinton expressed a glimmer of cautious optimism at the end of the interview. His optimism comes from the recent concept of “maternal AI” he has thought about and the possibility of international cooperation stemming from the shared interests of various countries on this issue. This casts a new hopeful light on the dark future he has long warned about. Finally, when asked whether the journalism profession would be replaced by AI, Hinton gave predictions similar to those for other professions: Katie Couric herself might have another 10-20 years, but younger journalists may need to consider changing careers, “although it’s unclear where to pivot.”

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。