Geoffrey Hinton has been warning for years about the dangers of artificial intelligence, often in an apocalyptic tone, but its latest forecast points to the possibility of a more feasible disaster, closer in time and with very concrete consequences. ‘At the end of this year or next we will see a Chernobyl, and that may be enough for people to really pay attention’he told The Guardian.
The comparison refers to the accident of April 26, 1986, when an explosion in a reactor at the Chernobyl plant, in the then Ukraine Soviet Union, destroyed it and released large amounts of radiation. Hinton uses this example as a symbol of a catastrophe that forces us to react to technological risks, not specifically referring to another nuclear accident..
More specifically, among the scenarios it proposes is an AI-assisted attack on electrical infrastructure that ’causes a massive brownout’, or an incident that paralyzes the financial system for a week. ‘Can you imagine what would happen? We could recover, (but) that would make them react and pay attention,’ he maintains, referring to ‘world leaders, except Trump.’
The author of these warnings is a 78-year-old British computer scientist, Nobel Prize winner, professor emeritus at the University of toronto and considered one of the ‘godfathers of AI’ for his contributions to methods that allow training neural networks. In 1986 he demonstrated, together with other researchers, the effectiveness of backpropagationa technique that allows adjusting the connections of these networks to reduce their errors and that remains essential to train artificial intelligence models.
He also participated in the development of AlexNeta network that in 2012 revolutionized image recognition and boosted deep learningthe technology on which current systems that recognize voice, generate images or answer questions are built. Between 2013 and 2023 he continued researching artificial intelligence in Google and it became vice president of the company led by Sundar Pichai.
In 2024 he shared the Nobel Prize in Physics with John Hopfield for discoveries and inventions fundamental to machine learning using these networks. His work applied tools from statistical physics to the development of systems capable of learning properties of data, contributing to the foundations of current artificial intelligence.
Since leaving Google in May 2023, he has spoken freely about its risks and has turned his interviews and conferences into a speaker for generally catastrophic warnings. That same year he even stated that a higher intelligence could retain humans temporarily to maintain power plants and dispense after us.
Compared to these hypotheses, the new scenario is more realistic because an attack against infrastructure does not require science fiction scenarios in which machines become conscious or replace humanity. Hinton places his concern on specific damages that could be caused by systems used maliciously..
The statements came before a screening of the documentary ‘AI, there’s probably nothing to worry about’a title suggested by a chatbot, in which he participates with his former Google colleague Demis Hassabis. ‘I have a reputation that makes me a spokesperson and I need to spend it completely on this,’ Hinton explains in the documentary.
His criticism also reaches those who run the large companies in the sector, especially Sam Altman, whom he claims he does not trust ‘at all’. ‘I think (Mark) Zuckerberg, (Elon) Musk and Sam Altman understand the risks and give a shit. They prefer to keep the money. That’s what I believe’.
To limit these dangers, defend mandatory testing of the most advanced models, independent evaluations and common standards between governments. Despite his warnings, he refuses to consider himself a ‘catastrophic’ and believes that AI can still lead to a ‘wonderful future’ and one ‘radical abundance’although he also admits that ‘we may have been better off if we had never developed it’.