Geoffrey Hinton, a godfather of modern AI, is surfacing skepticism about whether the…

The question of whether transformative technologies should have been developed belongs to a long tradition of technological ethics and philosophy. When prominent figures in a field voice doubts about their own life's work, it opens a window into how the most informed people think about unintended consequences, path dependence, and the limits of foresight. Deep learning's foundational moment The modern artificial intelligence renaissance traces to a specific technical achievement: the revival of neural networks as a viable approach to machine learning. For decades, neural networks had been considered a dead end by the mainstream research community. What changed was not the core mathematics, but three converging factors: dramatically increased computational power, access to vast datasets, and algorithmic refinements that made training deeper networks practical. This convergence occurred gradually over the 2010s and created the conditions for systems that could learn patterns from data at previously unimaginable scales. Why expert reversals feel unusual When technical founders question whether their field should exist, it registers as news because we expect consistency. Scientists and engineers typically advance their work, promote their methods, and build careers within their chosen domain. Complete epistemological reversal is rare at any career stage, rarer still among architects whose decades of work laid the intellectual foundation for an entire field. Such reversals often signal not fickleness but genuine concern emerging from proximity to the subject matter. Someone who built the tools understands their capabilities and limitations in ways outsiders cannot. The problem of foresight at scale A core tension in technology development is that intentionality and outcomes often diverge. Researchers working on narrow technical problems face genuine uncertainty about second, third, and fourth order effects once their work enters the world. They optimize for what they can measure in the lab: accuracy, speed, scalability. They cannot fully predict how institutions will deploy their tools, how incentives will shape adoption, or what behaviors will emerge at scale. This gap between isolated research goals and distributed real world impact is not unique to AI, but the power and reach of AI systems means the consequences can be proportionally larger. The perpetual tradeoff between capability and risk Every technology carries this tension. Electricity enabled both hospitals and weapons factories. The internet connected billions and created new modes of surveillance and control. Agriculture increased human population but created new dependencies and vulnerabilities. The relevant question is rarely whether a technology should exist in some absolute sense, but rather how its risks can be managed, who bears responsibility for problems, and what governance frameworks are adequate for the scale of impact. Acknowledging that a technology might have been bet