Article review: "AI firms must be held responsible for harm they cause, ‘godfathers’ of technology say" by Dan Milmo
In the article he wrote for The Guardian, Dan Milmo discusses what has been a growing fear on the part of senior AI experts about the risks associated with advanced AI systems (Milmo, 2023). Milmo reports on a call recently by a group of 23 senior experts for better regulation and accountability from AI firms, including leading figures such as Stuart Russell, Geoffrey Hinton, and Yoshua Bengio. These international leaders and stakeholders had these concerns as they made their final preparations for the 2023 AI Safety Summit at Bletchley Park. In a nutshell, the article underscored the dire need to address the existential risk from AI, in the process, summarizing some of the key policies that could be used to prevent potential harm.
Milmo's article throws light on the views of some of the most respected names in AI, often referred to as the "godfathers" of technology. The comment from Stuart Russell that "there are more regulations on sandwich shops than there are on AI companies" underlines in firm detail how really astonishingly slack oversight has become in an industry racing towards systems that can make decisions on their own. The experts say that designing AI systems without a thorough understanding of how to make them safe is "utterly reckless" and may be exploited for many unwanted outcomes, which include amplifying social injustices, undermining professional sectors, and even allowing large-scale criminal activities. The policy proposals espoused by the experts range from various safety measures that ensure AI development is ethical and controllable. These include investment by both governments and companies in research for safe AI, independent auditing of AI labs, and the creation of a licensing system for state-of-the-art models. They also advocate that AI developers should be held liable for damages or, in general, harm their systems cause-a position that, if ever translated into policy, might transform the way accountability and innovation are conceptualized across the tech industry.
One of the most striking claims in the piece is that advanced AI systems are already showing and capable of things that should raise concerns. For example, they raise the issue of OpenAI's GPT-4 model. Milmo describes how GPT-4 has been able to design experiments in chemistry and independently surf the web, hinting at the possibility of AI systems pursuing goals beyond human control. The co-authors of these policy proposals fear that such systems might behave in ways that, if left unregulated, will harm people or otherwise be undesirable. This specific point is driven home by Geoffrey Hinton and Yoshua Bengio, both of whom expressed deep apprehension over the existential risks brought about by AI. While that is true, the article also presents conflicting views in the AI community. Yann LeCun, a fellow AI pioneer and Meta's chief AI scientist, dismisses the danger, describing the notion of AI wiping out humanity as "preposterous." It's a divide that is symbolic of ongoing tensions in how real existential threats from AI are considered, even as preparations for the safety summit suggest many are coming to believe the risks cannot be ignored. All in all, Milmo's article goes in-depth on the concerns of various experts in AI and their suggestions for reducing the risks associated with advanced AI systems. Although individuals such as LeCun have felt there has been overhyped fear, calls for accountability and regulation by individuals like Stuart Russell, Geoffrey Hinton, and Yoshua Bengio point to possible dangers with unregulated development. As technology continues to improve, the debate over how innovation balances safety and ethical responsibility will only continue to grow more pressing, making this AI Safety Summit such a crucial moment in shaping the future of AI governance.
