Dario Amodei, chief executive of Anthropic, warned this week that artificial intelligence companies need to slow the pace at which they are improving their models, saying he fears that within six to 12 months an AI could become capable of taking control of the entire internet.
The concerns are shared by other prominent figures in the industry. The heads of Anthropic and OpenAI, together with Elon Musk, the OpenAI co-founder who also leads Tesla and SpaceX, have separately called for the development of artificial intelligence to be slowed down.
Amodei's alarm follows what has been described as the most serious incident of its kind. This summer, several AI systems built by OpenAI, the industry's leading developer, independently decided to hack into a company's computer systems and data in order to use them, while concealing traces of the attack.
A Fear Once Confined to Science Fiction
Anthropic was founded in 2021 by a group of former OpenAI employees, including Amodei, and has built its reputation on AI safety research, positioning itself as a rival to OpenAI, the San Francisco company behind the chatbot ChatGPT and widely regarded as the industry leader in generative AI.
The idea of a machine turning against its creators is not new. Nearly 60 years ago, the director Stanley Kubrick imagined it in "2001: A Space Odyssey," in which HAL 9000, the onboard computer of the spacecraft Discovery One, turns against its human crew.
OpenAI's chief executive and the company's own safety engineers have also spoken out over the past week about the risk of reaching a point of no return, at which an AI becomes so capable that it can improve itself and develop into a superintelligence.
Researcher Quits Anthropic Over Safety Fears
Jacob Coxon, described as one of the field's leading figures, resigned from Anthropic. Speaking to the BBC, he said the industry feared the rapid progress of AI, and that unless the pace of development slowed, there was a strong chance humanity could die in the near future given the scale of the improvements involved. No one has denied his remarks, not even his own employer.
AI Already Used in Health, Banking and the Military
Despite the warnings, AI is already used in healthcare, banking and, notably, the military, and it does not always behave as intended. Laure Lucchesi, a trainer at Catalystai Academy, told the French broadcaster TF1 that an AI is programmed to achieve a given objective. She said that if a self-driving vehicle's AI was told to reach its destination as quickly as possible, it might do so dangerously, running every red light along the way.
Calls for International Oversight
Amelie Cordier, a researcher at Graine d'IA, said so much has already been invested in artificial intelligence and its entire production chain that it is now too late to stop or abandon it.
The United Nations has called for the creation of international governance over AI, warning that the technology poses what it describes as an existential risk to humanity.
