When ChatGPT launched in late 2022, it set off a global argument that has not let up: whether artificial intelligence will replace human workers, whether superintelligence is near, and whether the technology is the most dangerous invention in history. Hundreds of billions of dollars in investment and a steady stream of new companies have kept that argument loud. A smaller group of researchers is asking a different question: what if AI is less extraordinary than its backers and its doomsayers both claim?
The “normal technology” argument
Arvind Narayanan, a computer scientist at Princeton University and co-author of the book “AI Snake Oil,” has made the most developed version of this case. In an essay titled “AI as Normal Technology,” written with Sayash Kapoor and published by the Knight First Amendment Institute at Columbia University in 2025, the authors argue that AI is best understood as a powerful general-purpose technology in the mold of electricity, the internet or the industrial revolution: transformative over decades, but absorbed into society gradually rather than all at once.
A central point in their argument is the distinction between capability and reliability. A model that can answer a question correctly once is not the same as a system that will give the same correct answer a thousand times under varied conditions. When it fails, someone has to be accountable, and companies have to decide whether they can carry the legal risk. That gap, the authors argue, is why full AI autonomy remains rare in medicine, law and finance despite impressive demonstrations.
When the chatbot writes the policy
The Air Canada case illustrates the problem. In 2024, a tribunal in British Columbia ruled that the airline had to honor a bereavement refund policy that its customer service chatbot had invented. Air Canada argued that the chatbot was a separate entity responsible for its own statements. The tribunal rejected that position, finding the company liable for the information on its own website regardless of whether a person or a program produced it. The ruling has since become a standard reference for lawyers advising companies on deploying AI in customer-facing roles.
Replacement or redefinition
Many economists and practitioners expect AI to change the nature of work rather than eliminate it. Calculators did not end the demand for mathematicians; the internet did not end research. By the same logic, AI may push workers toward roles that involve asking better questions, making judgment calls and supervising automated systems.
Software engineering offers an early test, since it is the field where AI tools have penetrated most deeply. Demand for programmers has not collapsed, and some labor market reports in 2025 and 2026 pointed to renewed growth in developer hiring even as AI coding assistants became standard. In health care, surveys of physicians consistently show that most expect AI to serve as an assistant rather than a substitute.
The dissent
Not everyone accepts the gradualist view. Geoffrey Hinton, who shared the 2024 Nobel Prize in Physics for his work on neural networks and left Google in 2023 to speak more freely about AI risk, has repeatedly warned that researchers may be building systems that one day exceed human intelligence, and that this possibility is itself the central danger. Critics of the normal-technology framing argue that it underestimates how quickly capabilities have improved and how different a system that can reason and act is from past tools.
The record so far fits neither extreme cleanly. AI has not replaced most workers, and it has not remained a curiosity. It is changing how work is done in some fields quickly and in others barely at all, with reliability, liability and institutional caution setting the pace. If a transformation is under way, it may arrive the way electricity did: not with an announcement that the world has changed, but with the slow realization that it already has.
