Andrew Ng: Big AI labs are fear-mongering to shape regulation

Andrew Ng, co-founder of Google Brain and Coursera, argued in a long-form interview that leading AI companies are exaggerating AI risks to push for regulation that entrenches their market position. Ng claimed a "handful" of large labs have been "very loud voices of fear-mongering around AI," with a clear incentive: if a company spends billions training a model, it is inconvenient when someone else releases a comparable one for free. Regulation, he said, would force "high toll" payments on newcomers while incumbents absorb the cost more easily. He dismissed the "AI is like nuclear weapons" analogy as "an analogy that has no basis in fact," and accused companies of cherry-picking incidents and overstating data-center resource use.

These are assertions about industry motives rather than demonstrated facts, but they carry weight given Ng's history — his online machine learning course reached millions — and his current stakes: he runs several AI ventures, including a firm that handles sensitive data for banks.

On jobs, Ng rejected the "job apocalypse" scenario. Citing economists Erik Brynjolfsson and Andy McAfee's task-level analysis, he argued AI can do roughly 30–40% of many jobs, which increases the value of the remaining human work. Software engineering is the profession most affected today because AI writes code well, he said, yet engineering openings are up and good engineers are "busier than ever." His formulation: "people that use AI will replace people that don't use AI."

On education, Ng said "AI models are terrible for learning." He cited studies showing students score higher on homework when using AI but retain less later, calling the effect "cognitive offloading." He announced a new organization, Learn Vector, focused on individualized tutoring rather than the one-to-many course format he helped popularize in 2011. Universities, he added, are slow to adapt, still preparing students "for the jobs of 2022."

On AGI, Ng attributed the disagreement between NVIDIA's Jensen Huang — who says it has arrived — and himself to divergent definitions. Using his own — "AI that could do any intellectual task that a human can" — he listed gaps such as writing a PhD thesis or learning to drive in an unfamiliar environment with minutes of practice. He said OpenAI had an economic incentive to declare AGI earlier under its agreement with Microsoft, but that the deal has been renegotiated and the incentive is gone.

His practical advice was to learn AI, build fast, and talk to customers. He described the real constraint as a "product management bottleneck" — deciding what to build — and cautioned that building a company remains much harder than building an app: AI can produce code in hours, but success requires technical depth or deep customer insight accumulated over years.

Ng's position after a decade and a half of teaching AI is that the skills shift is real, and the onus, in his view, is on individuals to move past fear and adapt.