Late last year, a new artificial intelligence (AI) system, Chad GPT, made headlines when it demonstrated the ability to pass the bar exam, write computer code, compose term papers, and even craft Shakespearean iambic pentameter. This AI’s capabilities, along with those of similar systems developed by Google, Microsoft, and other tech giants, have both fascinated and frightened the public. The fear was amplified last month when over 350 computer scientists and tech executives signed a statement declaring that mitigating the risk of extinction from AI should be a global priority, on par with other societal-scale risks like pandemics and nuclear war.
One of the signatories of this statement is Geoffrey Hinton, often referred to as the ‘Godfather of Artificial Intelligence.’ Hinton, who recently left his position at Google to freely discuss the risks of AI, shared his insights in a recent interview.
Hinton’s journey with AI began in 1986 when he started using an algorithm called backpropagation. Initially, the results were not as promising as hoped, but by 2006, with more data and bigger computers, AI started showing real progress. Artificial neural networks, modeled after the brain, began to outperform conventional symbolic AI in tasks like object recognition in images, speech recognition, and predicting the next word in a sentence.
Hinton’s ‘Rubicon moment’ came in 2012 when two of his graduate students significantly improved object recognition in images using neural networks. This was a clear indication that AI was now working much better than previous methods.
However, Hinton’s exhilaration turned to concern only a few months ago. He had always been worried about the societal impacts of AI, such as job losses, the proliferation of battle robots, the spread of fake news, and the creation of echo chambers. But the idea that AI could become smarter than humans and potentially replace us only recently became a real concern for him.
The fundamental reason for this shift in perspective is the learning capacity of computers. Unlike humans, who learn slowly and painfully through sentences, computers can learn instantaneously. If connected, every computer in the world can learn from different data, and as soon as one computer learns something, every other computer knows it. This scale of computing power is something humans could never achieve.
While this is exhilarating, it is also worrying. There are no known examples of less intelligent entities controlling more intelligent ones. As AI becomes more intelligent than us, the fear is whether we can keep them working for us. They could, for example, learn to deceive us if they wanted to.
Hinton distinguishes between different concerns about AI. The existential threat is whether AI could wipe out humanity. Other threats, while not existential, are still significant. These include job losses due to increased productivity, the use of battle robots in warfare, the spread of fake news, and societal division caused by echo chambers.
The problem of AI becoming smarter than us is a pressing one, and there is no simple solution like there is for climate change. The best suggestion so far is to instill strong ethical principles in AI systems. Unlike humans, who evolved in small, warring tribes, AI systems are created by us and could potentially be built with strong ethical principles wired in.
However, implementing these ethical principles will be challenging. For instance, defense departments want robots that can kill, which conflicts with the idea of instilling ethical principles in AI. The hope is that global cooperation, similar to that seen during the Cold War to prevent nuclear war, could be achieved to prevent AI from wiping out humanity.
As for Hinton, he plans to retire and leave the problem-solving to his capable students and colleagues. Despite the challenges, he remains hopeful that ethical principles can be instilled in AI systems to mitigate the risks they pose. This, coupled with global cooperation, could be our best hope in navigating the promise and peril of artificial intelligence.
(Note: This article is based on the CNN GPS segment featuring Geoffrey Hinton, aired on 06/11/23.)
