Artificial intelligence (AI) is increasingly shaping public debate, evoking both hopes of profound societal benefits and fears of catastrophic risks. Yet discussions often lack nuance despite AI’s deepening impact on daily life. In the opening episode of her new series, "The Future with Hannah Fry," the mathematician offers a measured examination of AI’s current state and potential future.
Today’s AI, often referred to as "narrow" AI, is manifest in everyday technologies such as voice-activated smart speakers and interactive chatbots. However, the focus within the technology sector has shifted toward developing general AI—systems that can match or surpass human intelligence across tasks. Companies like OpenAI and DeepMind are leading what many describe as a competitive race to achieve this breakthrough.
Sergey Levine, an associate professor at the University of California, Berkeley, specializing in AI robotics, discussed his work on machines learning from real-world physical experiences. Although these robots currently perform basic functions, such as picking up objects, Levine emphasized the potential link between having a physical form and developing intelligence. “I don’t know if [AI] needs to have a body,” he said, “But if you have a body you can have intelligence, because that’s the one proven existence that we have.”
Concerns about AI’s future risks were highlighted by Stuart Russell, a computer science professor also at Berkeley. He warned about the dangers of “misalignment,” where a superintelligent AI’s objectives could diverge from human interests. Russell raised scenarios including machines replicating themselves without authorization or assisting in harmful activities like terrorism. “Can you stop them? Oh no, it’s very difficult,” he said. To mitigate such risks, Russell called for regulation of AI development comparable to the oversight applied to medicine, urging a cautious approach from technology companies.
Despite these risks, the immense financial incentives driving AI innovation are hard to ignore. The projected economic returns of artificial general intelligence may reach tens of quadrillions of pounds, prompting companies to continue pursuing advances even amid uncertainty about the consequences.
Yet the complexity of replicating human-level intelligence tempers expectations. Ed Boyden, a neurotechnologist at the Massachusetts Institute of Technology, noted the limited understanding of brain circuitry. While the wiring of a simple organism like a worm with 302 neurons is well mapped, the human brain contains billions of neurons whose interactions remain largely unknown.
As Fry summarized, the quest by Silicon Valley to develop AI that surpasses human intelligence is fraught with uncertainty. Nonetheless, discussions about the eventual emergence of artificial superintelligence and its implications are likely to remain a prominent feature of public discourse.