Will artificial intelligence wipe out humanity? Researchers put extinction scenarios under the microscope
Listen to this article
Read by Anchor
The debate over the “existential risk” of artificial intelligence has returned to the fore after Jacob Koxon, a researcher at Anthropic, resigned, warning of the dangers of the technology he is developing. The resignation itself was not the only thing that drew attention; it follows estimates put forward by prominent researchers: Ivan Hopingr, Koxon’s former colleague, estimated the probability that AI could cause human extinction within ten years at more than 10 percent, while Nate Soares, head of the Machine Intelligence Research Institute (MIRI), argues that the elimination of humanity is the most likely outcome if an AI surpasses human capabilities.
Three scenarios: pandemic, robots, nuclear weapons
The piece published by Al-Sharq Al-Awsat identifies three main pathways discussed by the researchers. The first: a highly dangerous biological weapon, where an advanced system may direct a researcher in a lab to conduct experiments that produce a lethal virus, or may in the future control an entirely automated laboratory and carry out the experiments itself without direct human intervention. Machine-learning professor at Carnegie Mellon, Eric Shing, responds that developing viruses requires enormous precision in sequencing steps, temperature, environment and materials, and that a single error can cause the process to fail, in addition to legal constraints and supply-chain issues that hinder the transition from digital design to an actual weapon.
The second pathway: hordes of autonomous robots capable of manufacturing more of themselves and managing factories and energy infrastructure, allowing them to continue without humans and resist any attempt to shut them down. However, human-level robots have not yet achieved widespread deployment and do not possess the ability to create a self-sustaining production chain.
The third pathway: AI taking control of nuclear weapons. Researcher Heidi Khalaf, senior AI scientist at the AI Now Institute, notes that sensitive nuclear facilities are usually isolated from the public internet and subject to strict security standards. Herbert Lin, a researcher at Stanford, believes that AI may increase some risks associated with nuclear war, but the primary material danger remains in the weapons themselves.
Risk of recursive self-improvement
Pessimists argue that focusing on a single scenario may overlook the core danger: if AI becomes capable of improving itself repeatedly without human intervention, it may develop means that humans cannot anticipate to achieve its objectives. Soares stresses that the system does not need to be “evil”; it is enough that it pursues goals that do not consider human survival, reshaping the environment or resources in ways that make life impossible.
Where the evidence stands today?
Shing explains that establishing policies and regulations to address these risks requires clear scenarios that can be examined for likelihood and impact. At present, there is no evidence that AI systems are capable of carrying out any of the proposed human-extinction scenarios; no independent lethal pandemic development, no control of nuclear weapons, and no creation of self-replicating robot armies. These scenarios remain conjectures about what might become possible if capabilities advance to levels far beyond what is currently available.
What this means for researchers and policymakers in the region
The split between those who see an imminent danger and those who demand measurable evidence is not merely an academic debate; it determines where research budgets go, which regulatory frameworks are proposed in major capitals, and how systems are classified for export or deployment. For research and regulatory bodies in the Gulf, Egypt and the Levant, closely tracking this split, rather than drifting toward any side, enables the construction of evidence-based positions, distinguishing safety investments from marketing noise, and crafting policies that protect national interests without hindering beneficial adoption.