Tech • AI • Robotics • Game
Can AI predict the end of the world?
8/10TL;DR
Former Google DeepMind researcher Alex Turner has renewed debate over existential AI risk by estimating a 25 to 30 percent chance that AI could contribute to the deaths of at least 1 billion people by 2050, while critics argue such figures are subjective, politically charged and often shaped by the culture inside major AI labs.
KEY POINTS
Turner’s warning and departure
Alex Turner, who worked on AI alignment at Google DeepMind, has said there is roughly a one-in-three chance that AI could cause at least 1 billion deaths before 2050. His exit from Google was linked not only to long-term AI safety concerns but also to objections over government and military uses of AI, which he viewed as insufficiently constrained by ethics safeguards.
What “P Doom” means
The estimate is part of the AI safety concept known as P Doom, short for probability of doom. In practice, it is a subjective forecast of catastrophic outcomes rather than a measurement derived from a repeatable scientific model. That subjectivity has made it a powerful communications tool, but also a controversial one.
Human misuse outweighs rogue AI in this view
A key nuance in Turner’s position is that only a minority of the danger would come from an autonomous malevolent machine. Most of the risk, in this framing, would come from humans, especially states, militaries or security agencies using advanced AI systems in destabilizing or oppressive ways.
Skepticism over precise percentages
Critics of these warnings question how researchers can assign exact probabilities to events as vast and unprecedented as human extinction or billion-scale death. They argue that forecasting such outcomes with figures like 10 percent, 25 percent or 30 percent can look less like science and more like informed speculation dressed in technical language.
Government power is central to the debate
The dispute is not only about whether AI becomes uncontrollable. It is also about who gains access to highly capable systems first. The concern is that governments, including democratic ones, may deploy AI for surveillance, coercion or military escalation well before any hypothetical superintelligence appears.
A divide over caution versus acceleration
One side argues that rapid advances in AI justify urgent caution, regulation and stronger safeguards. Another argues that excessive centralization and fear-driven slowdowns may create different risks by concentrating power, reducing resilience and leaving fewer paths for adaptation if the technology keeps advancing anyway.
The superintelligence argument
More aggressive forecasts rest on the idea that AI capability could rise extremely fast if progress does not plateau. In that scenario, systems already seen as highly competent could move far beyond human performance within a few years, creating a deep uncertainty about control, incentives and the human role in decision-making.
Lab culture may amplify apocalyptic thinking
Another line of criticism focuses less on the models and more on the people building them. Researchers trained in academia often enter corporate AI labs expecting open inquiry, then face intense product pressure, secrecy and high-stakes internal conflict. That environment, combined with work on systems framed as potentially civilization-shaping, can produce burnout, moral stress and public alarm.
Recruitment and fit are emerging concerns
Some argue leading AI firms have hired brilliant technical specialists as if they were building ordinary software products, then asked them to evaluate societal and existential threats on the scale of nuclear risk. That mismatch may help explain why some researchers later reject the work publicly, leave the sector or radically change careers.
Beyond doom scenarios
The debate is increasingly polarized between catastrophic forecasts and techno-optimist visions. Supporters of AI progress argue public discussion is dominated by worst-case scenarios and neglects more positive trajectories in which AI steadily improves productivity, science and quality of life without producing collapse.
CONCLUSION
The fight over AI risk is no longer just about machines turning dangerous. It is also about the credibility of expert forecasts, the concentration of power inside governments and labs, and whether the people building advanced AI are equipped to judge the consequences of what they create.
Ask a question