Due Diligence

“I think robotics has a long way to go… I think the ability, the dexterity of human-level manipulation is something we have to wait a lot longer to get. ” – Dr. Fei-Fei Li – Stanford Professor – world-renowned authority in artificial intelligence

While AI has made dramatic progress in perception and reasoning, the physical manipulation and dexterity seen in human hands is far from being matched by machines.

Context of the Quote: The State and Limitations of Robotics

Dr. Li’s comment was made against the backdrop of accelerating investment and hype in artificial intelligence and robotics. While AI systems now master complex games, interpret medical scans, and facilitate large-scale automation, the field of robotics—especially with respect to dexterous manipulation and embodied interaction in the real world—remains restricted by hardware limitations, incomplete world models, and a lack of general adaptability.

• Human dexterity involves fine motor control, real-time feedback, and a deep understanding of spatial and causal relationships. As Dr. Li emphasises, current robots struggle with tasks that are mundane for humans: folding laundry, pouring liquids, assembling diverse objects, or improvising repairs in unpredictable environments.
• Even state-of-the-art robot arms and hands, controlled by advanced machine learning, manage select tasks in highly structured settings. Scaling to unconstrained, everyday environments has proven exceedingly difficult.
• The launch of benchmarks such as the BEHAVIOR Challenge by Stanford, led by Dr. Li’s group, is a direct response to these limitations. The challenge simulates 1,000 everyday tasks across varied household environments, aiming to catalyse progress by publicly measuring how far the field is from truly general-purpose, dexterous robots.

Dr. Fei-Fei Li: Biography and Impact

Dr. Fei-Fei Li is a world-renowned authority in artificial intelligence, best known for foundational contributions to computer vision and the promotion of “human-centred AI”. Her career spans:

• Academic Leadership: Professor of Computer Science at Stanford University; founding co-director of the Stanford Institute for Human-Centered AI (HAI).
• ImageNet: Li created the ImageNet dataset, which transformed machine perception by enabling deep neural networks to outperform previous benchmarks and catalysed the modern AI revolution. This advance shaped progress in visual recognition, autonomous systems, and accessibility technologies.
• Human-Centred Focus: Dr. Li is recognised for steering the field towards responsible, inclusive, and ethical AI, ensuring research aligns with societal needs and multidisciplinary perspectives.
• Spatial Intelligence and Embodied AI: A core strand of her current work is in spatial intelligence—teaching machines to understand, reason about, and interact with the physical world with flexibility and safety. Her venture World Labs is pioneering this next frontier, aiming to bridge the gap from words to worlds.
• Recognition: She was awarded the Queen Elizabeth Prize for Engineering in 2025—alongside fellow AI visionaries—honouring transformative contributions to computing, perception, and human-centred innovation.
• Advocacy: Her advocacy spans diversity, education, and AI governance. She actively pushes for multidisciplinary, transparent approaches to technology that are supportive of human flourishing.

Theoretical Foundations and Leading Figures in Robotic Dexterity

The quest for human-level dexterity in machines draws on several fields—robotics, neuroscience, machine learning—and builds on the insights of leading theorists:

The Challenge: Why Robotics Remains “a Long Way to Go”

• Embodiment: Unlike pure software, robots operate under real-world physical constraints. Variability in object geometry, materials, lighting, and external force must be mastered for consistent human-like manipulation.
• Generalisation: A robot that succeeds at one task often fails catastrophically at another, even if superficially similar. Human hands, with sensory feedback and innate flexibility, effortlessly adapt.
• World Modelling: Spatial intelligence—anticipating the consequences of actions, integrating visual, tactile, and proprioceptive data—is still largely unsolved. As Dr. Li notes, machines must “understand, navigate, and interact” with complex, dynamic environments.
• Benchmarks and Community Efforts: The BEHAVIOR Challenge and open-source simulators aim to provide transparent, rigorous measurement and accelerate community progress, but there is consensus that true general dexterity is likely years—if not decades—away.

Conclusion: Where Theory Meets Practice

While AI and robotics have delivered astonishing advances in perception, narrowly focused automation, and simulation, the dexterity, adaptability, and common-sense reasoning required for robust, human-level robotic manipulation remain an unsolved grand challenge. Dr. Fei-Fei Li’s work and leadership define the state of the art—and set the aspirational vision for the next wave: embodied, spatially conscious AI, built with a profound respect for the complexity of human life and capability. Those who follow in her footsteps, across academia and industry, measure their progress not against hype or isolated demonstrations, but against the demanding reality of everyday human tasks.