Daniela Rus

In the [YOU] on AI Field Guide

The cycle that began with [YOU] on AI wrestles with the fate of human agency in an age when a powerful cognitive amplifier becomes available to everyone. Rus offers something rare in that conversation: a concrete, engineering-grounded vision of collaboration in which the human is not displaced but extended—given superpowers—by machines that bring endurance and precision while people bring creativity and meaning. Her insistence that robots are tools, neither good nor bad in themselves, is an insistence that humans remain the agents: the technology is something we wield rather than something that wields us. Her refusal of both utopian and apocalyptic framing is not political centrism but a builder’s discipline, the discipline of someone who has watched robots fail in ways no simulation predicted and who therefore refuses to let impressive demos stand in for genuine capability.

Her calibration performs a crucial service in the discourse the cycle is trying to correct. The same fluency that makes a language model seem more capable than it is makes it easy to conflate describing the world with understanding it. Rus’s physical-intelligence lens cuts through this conflation: the machine that writes a flawless paragraph about catching a ball cannot catch one, and the gap between the two is not trivial. Until the machine can navigate an uncontrolled environment with the reliability of a living organism, something essential about intelligence remains unbuilt—and pretending otherwise misdirects both resources and expectations.

She also supplies the cycle’s most direct defense of the irreducible human core. Her claim that no machine will produce a genuinely new artistic movement, no artificial Shakespeare or Tolstoy, is not nostalgia for human specialness; it is a structural argument about what the machines are doing. They recombine and interpolate within the space of what has been. The deepest human creativity opens a space that did not exist before. The chip provides the means. The heart provides the why. And the why, she argues, cannot be delegated.

Origin

Born in Cluj-Napoca, Romania, the daughter of a computer scientist and a physicist, Rus came to the United States and earned her doctorate at Cornell University under John Hopcroft, one of the towering figures of theoretical computer science, with a dissertation on dexterous manipulation—how a machine might skillfully handle objects with its hands. The choice was revealing: from the beginning she was drawn not to computation in the abstract but to computation that reaches into the physical world. She began her academic career at Dartmouth before joining MIT in 2004, becoming director of CSAIL in 2012.

The intellectual biography is a succession of wagers that the hard problem lives in the physical: modular and self-reconfiguring robots that can rearrange their components to suit the task, soft robots made of compliant materials that offload intelligence into their bodies, the soft robotic fish SoFi that swam among real fish in a Pacific reef, the Roboat autonomous vessels that navigate the canals of Amsterdam. In 2020 her lab demonstrated that a liquid neural network with nineteen neurons could steer a car—the beginning of a body of work arguing that biological inspiration and elegant architecture could achieve with a handful of adaptive neurons what brute scale achieves with billions.

In 2023 she co-founded Liquid AI to carry the research into the world: foundation models built from first principles, compact enough to run on edge devices rather than in data centers, with an interpretability that comes from small size rather than from post-hoc explanation. She is a MacArthur Fellow and the author of The Heart and the Chip (2024), which argues that robots are tools designed to give people superpowers—not to replace them.

Key Ideas

Physical Intelligence. The hardest problem in artificial intelligence is not the symbolic manipulation that language models do but the embodied competence that a toddler takes for granted: moving through the world, handling objects, adapting in real time to surprises, acting in a domain that is dynamic, uncertain, and unforgiving. Physical intelligence requires not just a powerful mind but the right body, the right materials, intelligence distributed through flesh as well as computation. The fly navigates with a brain smaller than a grain of rice; the worm finds food and avoids harm with three hundred and two neurons. Nature has discovered architectures for embodied intelligence that are compact, efficient, and robust in ways our brittle, power-hungry systems are not.

Liquid Neural Networks. Inspired by Caenorhabditis elegans—the one-millimeter worm whose 302-neuron nervous system has been fully mapped—Rus and her collaborators invented a class of neural network whose internal parameters continue adapting to incoming data even after training ends. Where a conventional network is frozen at deployment, a liquid network stays fluid, handling novelty and distributional shift with robustness that brittle large models lack. Nineteen of them can steer a car. A handful can control a drone that finds an object in a winter forest after training in summer, because the network learns the causal structure of the task rather than its surface statistics.

Soft Robotics and the Intelligence of Bodies. Rus pioneered soft robotics—machines made of compliant materials that passively adapt to the objects they touch and the forces they encounter, the way a human hand conforms to an irregular object without requiring a precise model of its geometry. The body does some of the thinking. A soft gripper does not need to compute the shape of what it grasps; its material properties absorb the complexity. This is the principle of morphological computation: not all of a machine’s intelligence has to be in its brain. Some of it can be in its flesh.

The Heart and the Chip. Rus’s framing of human-machine collaboration rejects the zero-sum model in which the machine’s gains are the human’s losses. The chip is good at what the heart is not—precision, endurance, tireless processing. The heart is good at what the chip cannot do—creativity, judgment, the capacity to decide what matters and why. The error is to see these as competitors rather than complements. Her vision of robots as tools that give people superpowers is a vision of amplified rather than diminished humanity, in which the technology extends human reach rather than substituting for human purpose.

Democratizing Physical Intelligence. Rus has worked persistently to make the building of robots cheaper and more accessible—through flat-sheet fabrication that anyone can cut and fold, through 3D-printed designs that integrate mechanical and electronic components, through the compact embedded models of Liquid AI that run on phones and edge devices rather than data centers only well-capitalized organizations can afford. When the ability to build intelligent machines spreads widely, many more problems can be approached by the people who actually face them. The same argument she has made in robotics for decades is now unfolding across all of AI.