At a recent symposium, journalist Karen Hao and scholar Paola Ricaurte challenged the prevailing "bigger is better" narrative in artificial intelligence development. Hao argued that the current trajectory—defined by hyper-scale data centers and massive datasets aimed at achieving artificial general intelligence—is both unnecessary and unsustainable. She pointed to the significant environmental costs, including staggering energy and water consumption, as well as the human toll on global gig workers who manually label data for these behemoth models.

Instead of the current focus on scale, Hao proposed a shift toward task-specific models that leverage the computational strengths of AI for well-defined problems. She cited AlphaFold, the Nobel-winning protein-folding tool, as a successful precedent. By training on highly curated, smaller datasets, AlphaFold achieves transformative scientific breakthroughs without requiring the massive infrastructure associated with general-purpose models.

Professor Paola Ricaurte echoed these sentiments, stressing that technology must be purpose-driven and responsive to the communities it serves. The speakers urged the audience to actively participate in shaping AI’s future, noting that the technology’s path is not yet fixed. This vision promotes a more "bicycling" approach to AI—tools that are efficient, accessible, and designed for everyday utility—rather than the "rocket" approach of resource-heavy general models.

At a recent symposium, journalist Karen Hao and scholar Paola Ricaurte challenged the prevailing "bigger is better" narrative in artificial intelligence development. Hao argued that the current trajectory—defined by hyper-scale data centers and massive datasets aimed at achieving artificial general intelligence—is both unnecessary and unsustainable. She pointed to the significant environmental costs, including staggering energy and water consumption, as well as the human toll on global gig workers who manually label data for these behemoth models.

Instead of the current focus on scale, Hao proposed a shift toward task-specific models that leverage the computational strengths of AI for well-defined problems. She cited AlphaFold, the Nobel-winning protein-folding tool, as a successful precedent. By training on highly curated, smaller datasets, AlphaFold achieves transformative scientific breakthroughs without requiring the massive infrastructure associated with general-purpose models.

Professor Paola Ricaurte echoed these sentiments, stressing that technology must be purpose-driven and responsive to the communities it serves. The speakers urged the audience to actively participate in shaping AI’s future, noting that the technology’s path is not yet fixed. This vision promotes a more "bicycling" approach to AI—tools that are efficient, accessible, and designed for everyday utility—rather than the "rocket" approach of resource-heavy general models.