Unleashing the Power of AI: Better Safe Than Sorry?
We should heed scientists’ warnings regarding the dangers of AI and take measures to avoid these risks.
By Cerasel Cuteanu
Artificial Intelligence (AI) is great news for many, but has the potential to turn into a catastrophe if “improperly” handled, according to Yoshua Bengio, a renowned AI scientist. On July 25, Bengio presented a written testimony before the U.S. Senate Judiciary Subcommittee on Privacy, Technology, and the Law, warning about the potential risks that accompany the obvious benefits of recent advancements in the field of AI.
Founder and Scientific Director of Mila-Quebec AI Institute, and professor at Université de Montréal, Bengio received the 2018 Turing Award, often referred to as the Nobel Prize for Computing, for his research on deep learning. Bengio believes it is urgent that, while we “reap the economic and social benefits of AI”, we take countermeasures to protect mankind from the threats that come along with AI.
Undoubtedly, AI brings tremendous opportunities to many industries and fields of human activity. Yet, used by rogue regimes or in maleficent ways, AI can become a threat to democracy, human rights, and even the safety of mankind in general. It is foreseeable that a scenario may arise where humans lose control over decisions affecting humankind, as Bengio stated in his testimony, “none of the current advanced AI systems are demonstrably safe against the risk of loss of control to a misaligned AI.”
The future is predicted to bring along superhuman AI systems, which is one more reason to take measures to avoid a situation where such entities become misaligned with human values.
Originally, AI researchers believed that it would take more than a century for these systems to be widely used. Those forecasts were later revised, especially after the recent “giant leap” made by the launch of generative AI systems such as Chat GPT. Bengio believes that an AI system with human-level cognitive skills can be built within the next two decades, or even sooner.
That is worrying, Bengio says, since we need time to properly assess the potential risks stemming from the rise of AI. He says that we must seriously consider that AI can be misused, both intentionally and unintentionally. For example, Oxford’s Jonas Sandbrink wrote that two AI tools, large language models (LLMs) and biological design tools (BDTs) “could raise the ceiling of harm from biological agents and could make them broadly accessible”. That can lead to catastrophic events. Scientists are increasingly convinced that AI tools such as LLMs and BDTs can enable non-experts to trigger a pandemic, for instance. And these are just a few scenarios floated by experts in their academic writings.
Another scenario with catastrophic potential would be the nuclear one that, fictionally, was described in the short film Artificial Escalation. If the AI were to be integrated into the military NC3 systems (nuclear command, control, and communication) of two nuclear powers, something that is supported by military experts, a grim scenario, Bengio says, would be one where a misaligned system could generate a nuclear threat. Other experts also suggest that AI integration in military decision-making may end up reducing the time for “de-escalatory measures” and increasing the speed of warfare.
What can be done to prevent all that? In his testimony to the US Senate, Bengio stressed the need for national and international regulations to stymie the dangers that AI can pose to societies. At the same time, more resources need to be invested in research focused on AI safety to ensure that human values are protected from AI threats. According to Bengio, AI systems that are not “convincingly safe” should be banned. He also emphasized the importance of implementing oversight and safety protocols as soon as possible.
Listening to scientists’ warnings regarding the development of AI is the right thing to do. Safety is arguably more important than economic benefits. However, in a society dominated by competitive capitalism, Bengio and other scientists predict that achieving such self-control may be difficult.
Alan Turing prophetically wrote in 1951 that “once the machine thinking method had started, it would not take long to outstrip our feeble powers… At some stage therefore we should have to expect the machines to take control.” That is precisely what Bengio and other scientists say we need to avoid.
Cerasel Cuteanu is a fellow with the Media and Journalism Research Center. He pursues a PhD at the Babes-Bolyai University in Cluj, Romania.
This article was originally published on the Media and Journalism Research Center’s Journalism Trends blog.
Invest in independent media research and join a community of practice.
MJRC takes no government money — which is why we ask people who value independent investigation to invest in it directly. Your contribution supports our investigations and global analysis. Supporters can receive early access to new findings, invitations to small-group briefings, inclusion in our Supporters Circle updates, and the option to be listed on our Supporters Page.