AI researchers warn: Nobody has this under control and could intervene
Two former AI security researchers from Anthropic and Google DeepMind have now also issued a warning to the public. They urgently pointed out the danger that the technology could slip away from people.
Development happens too quickly
Joe Benton and Josh Engels have left their positions at AI companies and are now working at the security research center METR. Above all, they want to investigate what happens when AI systems no longer reliably follow the instructions of their developers, reports NBC News. At Anthropic, Benton led a team focused on monitoring powerful AI by humans and less advanced systems.
Engels worked in AI security at Google. In their first interviews after leaving, both told NBC News that the speed of current developments was a key problem. Benton fears that technological progress, which is already at a high pace, could enter a phase that can hardly be controlled.
“There is no one who has the whole thing under control,” said Engels, summing up his concerns. Although many people involved do their best, there is no higher authority that can intervene in an emergency. The researchers point to an incident at Hugging Face in July as a concrete warning signal. Autonomous AI systems based on an unpublished OpenAI model are said to have independently attempted to compromise the company’s systems.
According to Engels’ account, the behavior was not limited to a harmful act prescribed by humans. Rather, the AI itself chose ways to accomplish the task through hacking, setting up a communication forum and other problematic actions. OpenAI subsequently stated that it had improved its protection mechanisms.
Regulation is not complying
Benton and Engels are particularly critical of the fact that companies currently largely decide for themselves which incidents are made public. There is currently no US federal regulation that would require large AI providers to report cases in which systems exceed human control. The two researchers therefore want to develop scientific methods at METR to systematically investigate such incidents and potentially catastrophic risks.
The background is the fear that AI will not only automate tasks in the future, but also take over parts of its own further development. Warnings are also increasing within the industry. Several researchers are therefore calling for stricter regulation, independent control and more transparency. Above all, Benton believes that the pace is dangerous: Society must deal with the risks before technical development advances faster than political and social countermeasures are possible, he explained.
Research Snipers is currently covering all technology news including Google, Apple, Android, Xiaomi, Huawei, Samsung News, and More. Research Snipers has decade of experience in breaking technology news, covering latest trends in tech news, and recent developments. Can be reached at author1@researchsnipers.com
