NVIDIA CEO Jensen Huang: Geoffrey Hinton's Predictions Were Completely Wrong, Stop Predicting
NVIDIA CEO Jensen Huang specifically called out Geoffrey Hinton on the Ezra Klein show, stating that all of Hinton's predictions were wrong and that he should stop making predictions. In response to Hinton's claim that AI has about a 10% chance of causing societal collapse, Huang said that this 10% lacks scientific and research basis; what scientists say does not equate to science; these predictions are harmful.
Hinton had previously stated in 2016 that training radiologists should be stopped immediately, predicting that deep learning would surpass human image reading within five to ten years, comparing practitioners to coyotes that have jumped off a cliff without looking down. Ten years later, the number of practicing radiologists in the U.S. has increased by about 10%, and the case volume is expected to rise by about 25% from 2018 to early 2025; the average salary in 2025 is projected to be around $571,000, an increase of about 9% year-on-year; job vacancies once exceeded 4,300, with an average filling time of about 130 days. The shortage of radiologists in the UK once reached about 32%. Hinton later admitted that his prediction was premature and acknowledged that he had treated image reading as the entirety of the job.
Huang stated that doomsday theories scare young people away from college, making them think they won't find jobs; he argued that creating panic is not a public good. He believes that the track record of alarmists is poor and that warnings should be based on evidence and science. Hinton mentioned on the BBC that a 10% chance is not an unreasonable estimate, but he also admitted that no one truly knows how to provide reliable probabilities, adjusting the time frame from 30 to 50 years down to about 10 years.
In market mechanisms, the buyers are companies that need to expand chip production and hospitals that need to hire, while the sellers are researchers turning disaster probabilities into media products. The event was driven by a podcast confrontation, with funds remaining in computing infrastructure and medical imaging expansion, rather than halting the training of doctors; the beneficiaries are hardware vendors selling AI as an amplifier to hospitals, while the pressured parties are those trading extinction probabilities for regulation and attention.
Source: Public Information
ABAB AI Insight
Huang disassembled Hinton's authority into verifiable professional predictions: naming the industry, providing mechanisms, and setting a five to ten-year deadline. After the deadline, the numbers, salaries, and vacancies contradicted the suggestion to stop training. He separated "scientist's words" from "scientific research" to protect young people's investment in education and careers, while also safeguarding NVIDIA's demand-side narrative: AI increases scan volumes without eliminating the need for image readers.
The capital pathway shows radiology as NVIDIA's most convenient counterexample. Imaging algorithms have entered workflows, case volumes are rising, and hospitals need more staff, not fewer. Doom probabilities cannot be settled in laboratories but can alter the choices of 18-year-olds; Huang termed this change as harmful, equating the attention tax from safety narratives to reclaiming infrastructure narratives.
A similar structure is seen in the repeated delays of autonomous driving timelines, where taxi drivers have not disappeared, and despite the proliferation of translation software, there remains a shortage of conference interpreters: tasks are automated, and positions expand due to elastic demand. The AI employment debate is transitioning from a narrative of replacement to one of expansion, with pricing power shifting from extinction percentages to observable salaries and vacancies.
Structurally, this belongs to the misinterpretation of technological replacement as job elimination. The mechanism is that once a single task is taken over by a model, the system can perform that same task more times; elastic demand turns efficiency into increments rather than layoffs. Those who treat tasks as careers will misinterpret the labor market when the deadline arrives.
ABAB News · Laws of Cognition
- What scientists say does not equate to science.
- Judging a task as doomed does not equate to judging a profession as doomed.
- Creating panic does not automatically equate to public good.