Deutsche Bank Report Questions the Reliability of AI Doomsday Predictions

ME News report, September 16 (UTC+8), Beating AI Brief: Deutsche Bank Research has released “AI Apocalypse: A Short History of Terrible Technology Predictions,” casting doubt on recent debates about AI extinction. The report argues that it is already difficult to predict how far technology can advance, let alone when it will be realized, how society will adopt it, or what its ultimate impact will be. The judgments of Silicon Valley AI experts about AI’s future may be no more reliable than those of the average person. The report revisits a century-long history of failed predictions. In 1934, Einstein believed nuclear energy was virtually impossible to harness—less than a decade later, sustained nuclear chain reactions were achieved. Bob Metcalfe, co-inventor of Ethernet, predicted in 1995 that the internet would “crash spectacularly” the following year; the prediction failed, and he reportedly drank a pulp of the article containing it. The AI industry itself has also made numerous misjudgments. In 2016, Geoffrey Hinton predicted that deep learning would surpass radiologists within five to ten years and even suggested halting the training of new radiologists. Yet over the past decade, the number of radiologists has actually increased by approximately 10%. Deutsche Bank argues that such predictions often err by equating the easiest-to-automate parts of a job with the entire job. Radiologists do far more than interpret images—they also provide explanations, make judgments, communicate with patients, and conduct clinical decision-making. Similar pitfalls have arisen in predicting autonomous driving. Technological progress is merely the first hurdle; behind it lie endless edge cases, regulatory hurdles, liability allocation, and public acceptance. Deutsche Bank even suggests that how far AI ultimately advances may depend less on improvements in model capabilities and more on whether companies can successfully integrate it into workflows and are willing to pay for it. The report adds another layer of realism to AI doomsday narratives: nearly all participants have incentives to exaggerate. Founders need belief, investors need hype, consulting firms need urgency, policymakers need visibility, and media crave drama. Social platforms inherently favor negative, emotionally charged, and extreme claims—making moderate judgments far harder to spread. (Source: BlockBeats)