Every generation believes its technologies are about to destabilize the world. The printing press initially was accused of eroding authority and unleashing uncontrollable voices. Railways were said to compress space unnaturally, disturbing body and mind. The telephone threatened the centrality of the public square; television, the vitality of the street. More recently, the smartphone has been blamed for turning us inward, replacing the wandering gaze of the flâneur with the glow of a portable screen.Few technologies, however, have generated the same degree of anxiety as the latest wave of generative artificial intelligence, from ChatGPT to Claude. Leading voices have described GenAI as an “existential risk” that could spell the end of humanity. Even Pope Leo XIV dedicated his first 40,000-word encyclical to this subject this May.Yet we believe there is a field from which we can examine GenAI’s impact with less anxiety: design. Few people know that our profession has been experimenting with generative systems for about a decade, offering early insight into what these technologies might do when deployed at scale across society.Generative AI as we know it today, based on generative pre-trained transformers (GPTs) and large language models (LLMs), is a new tool but has older roots. Starting around 2017, digital artists such as Refik Anadol demonstrated the capacity of generative adversarial networks to produce new images derived from curated collections of existing artworks.By the time LLMs arrived, many in the design community immediately recognized the underlying principle: GenAI makes new things out of old things. In short, it’s an extraordinary imitation machine — where imitation means assimilation and transformation rather than simple copying.In architecture, for example, one can take an existing building and transform it into different visual languages: A modernist house can be reinterpreted through a rural vernacular; a historical style can be blended with contemporary forms; one cultural vocabulary can be translated into another. GenAI doesn’t understand architecture in the human sense. Rather, it identifies patterns and uses a “transformer” to produce things that resemble what it previously encountered.The same principle applies today to systems such as ChatGPT or Claude, which are essentially translation machines. They can turn German into Greek, Portuguese into Piedmontese. A poorly written draft into a well-structured preprint, or a long text into a concise summary. When asked to do the opposite — for example, to generate a detailed report from a short prompt — they often hallucinate. This limitation has changed little recently and is unlikely to disappear soon, as it seems inherent to the way GenAI is built.One can go further and translate from human language into machine language. This is the realm of “vibe coding,” where software is generated automatically. You speak to the system, and the system translates your intentions into programming. In all these cases, the power of GenAI comes from its ability to absorb enormous quantities of human production, then merge and transform it into new forms that are similar but different from the precedent from which they derive.This should not be dismissed as trivial. Imitation always has been central to intelligence. It was the foundation of the famous “imitation game” proposed by British computer scientist Alan Turing in 1950, which later became known as the Turing Test. This experiment evaluates a machine’s skill at imitating human conversation convincingly enough that a human judge cannot reliably distinguish it from another human. It remains a key benchmark for evaluating machine intelligence.Imitation is also how humans learn. Children pick up languages by imitating sounds before they understand grammar. Traditional craftspeople learn their trades by observing masters more than they do by reading instructions. Across cultures and centuries, imitation has been one of the fundamental mechanisms through which knowledge is transmitted. Neuroscientists even discovered a special class of neurons, called mirror neurons, associated with the ability to imitate observed actions in species such as monkeys and humans.Still, imitation is not invention. This is why architects often use GenAI at the beginning of a design process — much as they once used reference books or Google image searches — to explore possibilities and discover unexpected directions. The most important step comes afterward: moving beyond the patterns already present in the data. In a profession that defines itself through invention, GenAI increasingly is seen less as a creative genius and more as a “stochastic parrot” — a term introduced in 2021 by leading AI researchers to describe a technology that learns to repeat the most likely sequences found in whatever data it’s been shown.This distinction does not mean that GenAI will have limited economic consequences — quite the contrary. Consider translation, editing human text, finding bugs in computer code (where ChatGPT and Claude have demonstrated market-disrupting capabilities) or vibe coding: the commercial impact for each could be huge. The same pattern likely will emerge across many other sectors. Tasks based primarily on imitation increasingly will be automated.However, if we look at design itself, we also can see limitations. After 10 years, there has not been a significant transformation in employment. The creative inputs at the beginning of the process — and, importantly, the final evaluation and judgment — have remained firmly in human hands. Neither has yet been taken over by AI.We do not know precisely to what extent the lessons from design will extend across a broader spectrum of disciplines. Moreover, making coding almost a commodity most likely will accelerate the development of new AI systems, which in turn will acquire new capabilities. Even so, if our lesson holds at least partially, there is no need to surrender to either utopian fantasies or dystopian fears.With Pope Leo’s recent reflections on technology in mind, one might say that AI does not replace the human; it challenges us to recognize what is most human. And the most human ability of all is not imitation, but invention: the capacity to see what has not yet been seen.Carlo Ratti is a Professor at the Politecnico di Milano and Massachusetts Institute of Technology, where he directs the Senseable City Lab. He is the co-founder of the design and innovation office CRA–Carlo Ratti Associati and the co-author of “Atlas of the Senseable City.”Mario Carpo is the Reyner Banham Professor of architectural history and theory at the Bartlett School of Architecture in London. His latest book is “Beyond Digital: Design and Automation at the End of Modernity.”
