Right Way to AI: Benefits Without Overdependence

Recent advances in AI, high-performance computing, and edge computing are driving transformative changes in computer science, turning once-theoretical possibilities into reality. At the same time, they are raising new challenges over data security, privacy, and ethics.

Joseph Sifakis, renowned computer scientist and 2007 Turing Award recipient, shared his perspectives on these developments. The Greek-born researcher holds positions at France's Verimag Laboratory and China's Southern University of Science and Technology. He is also a member of the Chinese Academy of Sciences.

Sifakis earned computing's highest honor - often called the "Nobel Prize of Computing" - alongside Edmund Clarke and Allen Emerson for developing model checking. This breakthrough verification technology now underpins critical systems across hardware, software, and network protocols.

The Turing Award recognizes extraordinary contributions to computer science, with typically just one recipient selected annually through rigorous evaluation. Few exceptions are made for collaborative breakthroughs.

On AI's developmental stage

"China could leverage its vast industrial ecosystem to lead the pursuit of autonomous AI," Sifakis said during the China Development Forum in March 2024.

As technology advances, people are debating what AI can do and how it will affect the future. Some speculate the imminent arrival of a "singularity point" where AI surpasses human intelligence or even dominates humanity.

Sifakis dismisses such notions as baseless. While acknowledging significant AI progress in recent years, he maintains that the technology remains in its "infancy," far from achieving artificial general intelligence (AGI).

AGI, often framed as a form of "strong AI," is, in his view, more of a buzzword coined by tech companies. "Defining true AGI is challenging because we don't yet know where AI is headed," he said.

He remarked that while discussions about AI are ubiquitous, most lack clarity on what intelligence truly entails. Definitions of intelligence predate AI's emergence. The Oxford Dictionary defines it as "the ability to learn, understand, and think rationally about the world."

Sifakis said current AI capabilities remain limited. "AI provides building blocks for intelligent systems, but we lack the principles and techniques to synthesize them effectively."

He predicts that accelerating integration between information technology and AI will lead to autonomous systems. "When we can reliably build systems that replace humans in performing tasks, we'll know machine intelligence has approached human levels. That leap from narrow AI to AGI will mark a milestone where autonomous systems surpass machine learning frameworks."

Regarding claims of AI dominating humanity, Sifakis remains skeptical. "AI currently has neither the will nor the capacity to rule humans. The real issue is determining machines' role in our lives. Overdependence on AI for decisions risks eroding critical thinking — that's the true danger."

He criticized tech firms and media for mythologizing AI's progress: "For now, AI is just machines that stop working when unplugged."

AI vs. human intelligence

"Human intelligence excels in integrating and coordinating multiple tasks, not just singular skills," Sifakis said. While machines may outperform humans in specific tasks (e.g., computational intelligence), humans remain superior in comprehension, imagination, adaptability, and innovation within complex scenarios.

"Intelligence spans multiple dimensions and stages; it cannot be reduced to a single metric," he added.

For instance, when a chatbot is instructed to generate content, the machine merely calculates word probabilities without understanding its actions. "Linking data-driven knowledge with symbolic reasoning to achieve true comprehension remains AI's greatest future challenge." Humans face cognitive limits in extracting insights from high-dimensional data, while machines struggle to create and apply symbolic knowledge. "The ideal path forward is combining AI and human intelligence for optimal outcomes."

AI risks

Sifakis acknowledges that technological advances streamline services, reduce decision-making burdens, and enhance the quality of life. Yet they also diminish the need for autonomous problem-solving, rendering certain skills obsolete.

"Early humans invented levers to move stones, reducing reliance on brute strength. But losing physical power pales in comparison to losing the ability to think and produce knowledge," he cautioned.

He also mentioned petitions urging caution about AI risks, including one calling for a six-month pause on large-model development. "This isn't opposition to AI, but a recognition that technology is outpacing regulation. We need time to assess impacts and establish frameworks." He singled out the trend of students using ChatGPT for homework as particularly "dangerous." "Education trains minds. Over-reliance on AI stunts independent thinking, focus, and creativity. Governments should ban such tools in K-12 education."

Unchecked AI dependence, he warned, could render people "slaves of slaves" and people could progressively lose skills and agency.

For young people, he has this to say: "Young people should strive to master diverse technical and scientific knowledge across disciplines. Lifelong learning is vital; building a broad foundation early is wise."

This article was edited based on the original Chinese version written by ZHANG Xinxin from Cover News based in Chengdu, Sichuan province.

Source: Science and Technology Daily