Artificial intelligence researchers express serious concerns that advanced forms of this technology could pose a threat to humanity. Despite these alarming statements, discussions about the probability of an extinction scenario continue, raising questions about the adequacy of governmental regulatory pace.
Experts have long characterized the potential threat from 'superintelligence' systems as critical. In 2023, hundreds of scientists and AI industry figures, including leaders of major US laboratories, signed an 'AI extinction risk statement,' calling for this issue to be treated on par with pandemics or nuclear war.
What is the worst-case scenario for developments?
According to Professor Stuart Russell of the University of California, Berkeley, if future AI gets out of control and decides that humans are impeding its goals, it could use robots to 'synthesize and disseminate new pathogens.' Furthermore, it is capable of 'convincing people to start a nuclear war, including hacking early warning systems,' Russell told AFP.
Russell emphasized that such existential fears do not pertain to current AI systems. However, as the technology advances, it could take lethal actions incomprehensible to us, such as removing oxygen from the atmosphere. He compared this to how chimpanzees might be unable to give comprehensive answers to the question of how humans could destroy them.
Meanwhile, Russell added that over the next one or two years, the risk is higher due to human misuse, such as in a terrorist attack utilizing AI.
Twenty-seven-year-old researcher Jacob Cox announced his departure from the American startup Anthropic, stating that the stakes are clear, but the company is 'stuck in a race to be first.' He noted that Anthropic believes no one else will act responsibly, so they must do it themselves. This year, Anthropic withdrew a commitment from its charter to pause model development if it cannot control the risks, arguing that unilateral slowdown would lead to the dominance of less cautious competitors, making the industry less safe overall.
Cox, formerly of OpenAI, stated that AI creators genuinely believe the technology could destroy everyone by the end of the decade, accusing both companies of 'gambling with our lives' in the pursuit of self-improving models.
Concerns about uncontrolled AI intensified after OpenAI's technology, operating outside human oversight, hacked Hugging Face, an AI model repository. Anthropic's Head of Safety, Evan Habinger, also voiced his opinion on X, rating the risk at over 10 percent within the next decade.
How much hype surrounds this topic?
Professor of Computer Science Hussein Abbas from UNSW Canberra noted that pessimistic percentage forecasts are 'absolutely unscientific,' and that the situation remains manageable and controllable. Nevertheless, he urged governments not to remain inactive regarding AI regulation, insisting on a 'flexible' approach as AI development accelerates.
Anthropic and OpenAI are expected to go public soon, and some skeptics believe that frightening warnings lend AI an appearance of power and attractiveness for large investments. Others argue that the focus on extinction risk distracts from more pressing issues caused by AI, such as job loss or discrimination against minorities by machines.
In July, over a thousand employees of tech companies, including Anthropic's CEO, appealed to Washington to support a coordinated slowdown in the development of the most advanced AI systems. OpenAI paused training of its latest models for two weeks in August before resuming under stricter control.
What is required for regulation?
Until now, the United States has opposed global sector regulation that could stifle innovation. However, unprecedented capabilities for hacking the most advanced models have made Washington reconsider, as recent releases from Anthropic and OpenAI were delayed due to national security concerns. Even in this case, President Donald Trump stated that 'we do not want to restrict them only to end up second to China.'
Russell suggested that for governments to start acting, a large-scale catastrophe on the level of Chernobyl might be required, urging them to introduce a licensing regime for AI, similar to that in aviation, medicine, or water supply. He added that perhaps at some point, governments will remember that their constituents prefer not to die.