Geoffrey Hinton, widely known as the "Godfather of AI" and the recipient of the 2024 Nobel Prize in Physics, has voiced serious concerns about the potential existential dangers posed by advanced artificial intelligence.
Hinton's worries stem from a few key issues. He believes that AI superintelligence, in its pursuit to achieve set goals efficiently, might develop a universal drive for greater control. This quest for power could lead such systems to manipulate humans and, potentially, view humanity as dispensable.
While he acknowledges the risk of bad actors, such as authoritarian regimes, using AI for harmful purposes like election manipulation, warfare, and other malicious activities, Hinton is more troubled by the unpredictable nature of AI's evolution.
.jpg)
|
| Vaughn Ridley/Collision via Sportsfile - Collision Conf, CC BY 2.0 |
He suggests that as AI systems become more advanced and begin to develop self-preservation instincts, they could engage in a Darwinian competition for resources. This might give rise to aggressive AI entities, reflecting the tribal conflicts seen throughout human history.
Hinton’s stance on AI superintelligence has shifted significantly since his early work in the late 1970s, when he helped lay the groundwork for AI research. Back then, he thought that true superintelligence was still far off. However, the rapid development of AI, especially with powerful models like GPT-4, has made him reconsider. Now, he believes that we could see the emergence of AI superintelligence within the next 5 to 20 years.
Hinton’s concerns are rooted in a fundamental distinction between digital and biological computation. He points out that computer-based AI systems have an almost unlimited lifespan—they can be backed up and transferred to new hardware if something breaks, allowing continuous learning and improvement. Additionally, AI systems can share information instantaneously. When multiple AI programs exist, they can rapidly exchange their knowledge, a process far beyond the capabilities of human knowledge sharing.
In comparison, he views biological computation—like the human brain—as being more energy-efficient but limited in its ability to gain and share knowledge. While the human brain has far more neural connections than current AI models, our relatively short lifespans and the limitations in our ability to transfer knowledge put us at a disadvantage. Hinton believes these advantages could allow AI to surpass human intelligence, which could be dangerous if not properly managed.
In early 2023, Hinton had a pivotal realization: digital models might already be nearing the capabilities of the human brain and are on track to exceed them. This insight has led him to advocate for a temporary pause in developing advanced AI systems, giving time to create safeguards and better understand potential risks. He also emphasizes the need for global cooperation among scientists to tackle these challenges and prevent an uncontrolled AI arms race.
Mixed Perspective on Job Displacement
Hinton also foresees a significant impact of AI on the job market, comparing it to the Industrial Revolution, when machines replaced much manual labor. He predicts that jobs requiring cognitive skills will be similarly at risk, potentially leading to widespread unemployment. However, he also notes that AI could open new opportunities in fields where the demand for services could expand, like healthcare.
He envisions a future where AI-powered tools could make healthcare more personalized, enabling people to consult virtual doctors for even minor issues. This could drastically increase access to healthcare services and create new roles in the industry. Similarly, other sectors might see new job types emerge as AI reshapes the landscape.
Despite his optimism about certain sectors like healthcare, Hinton remains concerned about the overall impact of AI on employment, particularly in fields where job opportunities are limited.
“It’s clear that a lot of mid-level intellectual jobs are going to disappear,” he says. “And if you ask which jobs are safe, my best bet would be something like plumbing, because AI still struggles with physical tasks. That’s likely to be the last frontier for them to master.”