AI 'Godfather' Warns that Time Is Running Out to Properly Control Artificial Intelligence

Yoshua Bengio, one of the “Godfathers of AI,” worries that regulators may be running out of time to properly govern the rapidly advancing field of artificial intelligence.

Bloomberg reports that Yoshua Bengio, a pioneering researcher in the field of deep learning and one of the most influential academics in artificial intelligence, is sounding the alarm on the urgent need for AI regulation. In a recent interview, Bengio expressed his growing concern that policymakers may not have enough time to implement proper safeguards as AI technology continues to advance at an unprecedented pace.

Bengio, whose groundbreaking work laid the foundation for today’s AI models, admitted that his assumptions about the speed of AI development have been upended following the release of OpenAI’s ChatGPT in late 2022. “I’ve completely changed my mind after we ended up with machines that can talk back to us,” he said, acknowledging that he once believed we were “decades away” from developing AI capable of human-level performance.

In light of this realization, Bengio has begun to focus more on advocating for AI regulation. Along with fellow AI pioneer Geoff Hinton, he has come out in support of a controversial bill in California that would hold companies liable for catastrophic harms caused by their AI models if they fail to take appropriate safety precautions. Given California’s central role in the tech ecosystem and the absence of federal legislation in the US, this bill could set a precedent for how other jurisdictions regulate AI.

While many tech companies, venture capitalists, and AI academics oppose the California bill, arguing that it would stifle innovation and harm U.S. competitiveness in the field, Bengio believes that the conversation around AI regulation may be influenced by those looking to profit from the technology.

Bengio emphasized the importance of swift action, stating that “policy takes a lot of time” and that we may not have a decade to “get this right.” He stressed the need for not only domestic regulation but also international treaties to address the global nature of the issue.

When asked about the potential risks of AI, Bengio remained agnostic, acknowledging that there are compelling arguments on both sides. However, he maintained that from a policy perspective, it is crucial to protect the public against the worst-case scenarios, just as governments do in many other areas.

Bengio also highlighted some of the most concerning applications of AI, such as synthetic child pornography and the potential for deepfakes to distort the democratic process. He noted a recent study from Switzerland that demonstrated GPT-4’s superior ability to persuade people to change their minds compared to humans.

Despite these concerns, Bengio remains optimistic about the potential benefits of AI in fields such as biology, medicine, and drug discovery. He emphasized that these applications are generally not dangerous and should receive significant investment.