In a context where the race for artificial intelligence is accelerating. Tech giants and between the United States and China, the question of the security of AI systems is becoming a major global issue. On the occasion of the AI Summit, Yoshua Bengio, often referred to as “the godfather of AI” . A presented the AI Safety Report , during an event organized by Reporters Without Borders and the Center for AI Safety (CeSIA), at the ENS. This report is the first international document on the security of artificial intelligence. Supported by 30 States, the OECD. A UN and the European Union, it was written by 100 AI experts and summarizes the current state of advanced AI capabilities as well as the associated risks. Its objective is not to issue political recommendations . A but to provide a solid scientific basis to promote an . A international understanding of the challenges posed by advanced AI. “AI does not impose itself on us; it is the choices of individuals that determine its future . The report recalls. In particular . The warns against the concentration of AI power in the hands of a few individuals, which could jeopardize democracy and increase the risks of massive influence and political destabilization. As well as the possibility of loss of control of AI, which would want to “escape” itself and lead its own life…
The big picture
– Today, we are witnessing an exponential increase in AI capabilities . Between 2018 and today, each year, the computing power dedicated to training systems has quadrupled. At the same time, data efficiency has increased by 30% , while algorithmic efficiency has tripled. In addition, global investments in AI have reached approximately $100 billion , with an annual growth of 30%.
– A trend toward Artificial General Intelligence (AGI) . Three individuals that determine Turing Award winners, including Yann LeCun, Meta’s chief AI scientist, believe that AGI could become a reality in a few years or decades. AGI is defined as AI that has human-like capabilities for most tasks. Some experts worry about 99 acres database the profound implications of such a breakthrough. “What scares me the most is the possibility that humanity could disappear. It’s terrifying, even if it’s a tiny risk, we have to worry about it,” warns Yoshua Bengio.
The risks
– Artificial intelligence presents three broad categories of risks: those related to malicious uses, such as the dissemination of false content online or biological attacks; those individuals that determine resulting from malfunction, such as bias or loss of control; and finally, systemic risks, including invasions of privacy or major croatia phone number library disruptions to the job market. Among the concerns raised, some experts predict that jobs that do not require a direct human relationship could disappear, especially in a world where low-cost production could be done in rich countries, where AI is developed. In this context, the question of the adequacy of social safety nets and economic impacts becomes crucial. Some experts believe that these risks will not materialize for several decades, while others argue that general-purpose AI could cause societal damage in the coming years.
One of the major risks of artificial intelligence is the loss of control of machines.
“If we go further, 2, 3 or 5 years, what worries me the individuals that determine most is the loss of control ,” warns Yoshua Bengio. The concern lies in the development of self-preservation behaviors by machines. “ The machines we are training have a will to live. If they learn that they are going to be bfb directory replaced by a new version . . A try to save themselves. They try to hack the computer they are on to set up their own code in place of the new version. . It is a bit like an animal that we raise . A whose behavior as an adult we do not really know. “We are creating intelligent agents, entities that have their own objectives. It is a self-preservation objective. We do not want to create machines that want to save themselves.” Hence the importance of developing control methods before these systems become more powerful than humans.