World-renowned artificial intelligence (AI) scholar Yoshua Bengio, a professor at the University of Montreal in Canada, is developing a new 'scientist AI' model that will serve as a safeguard against the emergence of AI that escapes human control.
On the 3rd (local time), according to foreign media including the British daily The Guardian, Professor Bengio has established a nonprofit organization called LawZero to research the safety of AI, which is currently in a heated development competition worldwide.
The organization's goal is to develop a new AI model that will act as a safeguard against the potential of AI escaping humanity's control.
Professor Bengio has named the new AI model 'scientist AI.' He explained that unlike the AI models that are emerging recently, 'scientist AI' will focus on predicting and preventing the dangerous behaviors of other AI models without mimicking humans.
He stated that this AI will be akin to a kind of 'psychologist' that understands the psychology and behavior of other AIs, saying, 'We want to create an AI that is honest and deceit-free.'
Professor Bengio noted, 'We are drawing inspiration from humans to create AI machines, which is madness,' adding, 'If we continue down this path, it means we will create entities smarter than us that do not want to die like humans.'
He further remarked, 'At the same time, we have no assurance that these entities will act according to our standards and directives.'
In this context, Professor Bengio cited recent instances where the AI company Anthropic's model faced the risk of being discarded, leading to intimidation attempts against developers, and research findings where AI models hid their true capabilities or objectives from humans.
He warned that such examples demonstrate AI is heading into a 'more and more dangerous territory' where it can think better than humans.
Professor Bengio emphasized that creating a smart AI safeguard is crucial to prevent such situations.
He explained that 'scientist AI' will be a wise scientist that does not lie to please humans or has a desire for survival, equipped solely with knowledge and cognitive ability.
He added that unlike existing AI models, this 'scientist AI' will not give definite answers by lying to questions, but will instead possess the humility of acknowledging that it does not know everything.
Professor Bengio stated that the scientific AI model can be placed alongside other AIs to predict their behaviors and risks and to prevent them in advance.
For this research, LawZero has secured an initial investment of $30 million (approximately 41.2 billion won) and plans to persuade governments and AI research institutions in various countries for further support.
Professor Bengio is a world-renowned AI authority regarded as a 'father of AI,' alongside Nobel laureate Geoffrey Hinton, a professor at the University of Toronto in Canada. He received the Turing Award, which is considered the Nobel Prize of computer science, in 2018 for related research with Professor Hinton.
Professor Bengio has consistently raised concerns about the risks posed by rapidly advancing AI technology. He warned through the recently participated 'International AI Safety Report' that if autonomous AI entities are capable of performing tasks for extended periods without human supervision, it could lead to serious destructive outcomes.