Google DeepMind’s Demis Hassabis wanted to keep AI in the lab longer: Here’s why

Google DeepMind co-founder Demis Hassabis has reflected on his original vision for artificial intelligence (AI) development, which he believes was overshadowed by the competitive rush to commercialize the technology. Hassabis, who founded DeepMind with the mission to 'solve intelligence and use it to solve everything else,' has stated he would have preferred to keep AI development in a controlled scientific environment for much longer. He argued that the final stages of creating artificial general intelligence (AGI) should have been approached with caution, akin to a global scientific collaboration like CERN, to ensure rigorous understanding at each step. Hassabis emphasized that AGI holds the potential to be the 'most transformative technology in human history,' warranting a deliberate and methodical development process. He acknowledged that this approach might have delayed progress by a decade or more but believed the risks of hasty development outweighed the benefits. His ideal scenario involved isolating AGI development while deploying specialized AI systems, such as DeepMind’s AlphaFold, to tackle specific scientific challenges like curing diseases or discovering new materials. In his vision, society would benefit from narrow AI applications—like AlphaFold’s protein-folding capabilities—without exposing the broader, unpredictable AGI technology to market pressures. Hassabis highlighted that language models, such as those powering chatbots, proved easier to develop than anticipated, contributing to the accelerated race among tech firms. This unpredictability, combined with a shared blind spot among AI researchers, ultimately led to the release of advanced models like Google’s Gemini. Hassabis’ original blueprint prioritized medical and scientific advancements over consumer-facing AI products. He suggested that specialized AI systems could deliver immediate societal benefits, such as cancer cures or new energy sources, while AGI remained under strict scientific control. However, the competitive landscape and public demand for AI-driven tools shifted the focus toward rapid commercialization, leaving Hassabis’ vision largely unrealized. Despite his reservations, Hassabis acknowledged the inevitability of AI’s commercialization, citing the unexpected ease of developing language models as a key factor. The release of models like Gemini reflects the industry’s shift toward a consumer-driven race, where market forces dictate the pace of innovation. Hassabis’ insights underscore the tension between scientific rigor and commercial ambition in the pursuit of AI advancements.