Thursday 8 October 2026
Getting Away With Not Thinking
We tend to think that intelligence is about inference – thinking and reasoning. What if some of that work is already done by an agent’s shape, before any brain gets involved?
Yesterday, a paper in Science Advances described a soft robot inspired by the feather star, a marine animal whose feathery arms allow it to move through water. The robot uses two pneumatic actuators to control four flexible wings: by changing the timing of those inputs, the researchers can make the robot flap upwards, move forwards or backwards, hover, or spin around its own axis. A conventional design would typically need at least six actuators to produce the same range of motion – here, much of the work is done by the material itself.1
The researchers call this »mechanical intelligence«. The phrase points to a deeper question than whether a robot can be intelligent: how much of what we call behaviour actually needs to be computed? There are at least three different ways a body can end up with such competence already built in: it can be designed, it can be discovered, or it can be developed.
In 1948, the cybernetician W. Grey Walter built two small autonomous machines, Elmer and Elsie. Their equipment was almost comically simple: some light sensors, a few motors, and a handful of electronic components. Yet the tortoise-like robots exhibited surprisingly complex behaviour as they moved towards light, responded to obstacles, and interacted with their environment.2 Walter had not programmed miniature animals. He had arranged a small set of mechanisms so that the behaviour emerged from their interaction. This is the first of the three ways: design. Intelligence is built in by arranging the mechanism well enough to get a useful behaviour out of it.
Decades later, artificial-intelligence researchers Rolf Pfeifer and Josh Bongard gave this intuition a more systematic name: »morphological computation«. The idea is that the body can perform part of the work that would otherwise have to be performed by a brain or controller.3 A flexible knee joint, for example, can absorb irregularities in the ground without a computer calculating exactly how far it should bend. This is a different claim from architect Frei Otto’s form-finding, where a soap film or a hanging chain settles into its shape once and stays there. Here, the shape keeps working – responding differently, moment to moment, as the ground or the water changes underneath it.
The nematode Caenorhabditis elegans is a very simple worm. Carrying only 302 neurons, it still navigates its environment, finds food, and responds to threats. In 2023, a team at Georgia Tech built a limbless robot to test how much of the worm’s undulating locomotion could be reproduced without complex neural-style control, and found that a surprising amount of it could, because the mechanics of a flexible body moving through a resistive medium take over part of the job that the nervous system would otherwise have to do.4
As author and researcher Max Bennett argues in A Brief History of Intelligence, this pushes the history of intelligence much further back than the appearance of brains that look anything like ours. Intelligence may not begin with the ability to construct an internal representation of the world. Maybe it begins with an organism whose body is simply capable of doing something useful in the world, like swimming towards food.5 But how does a body acquire such useful capabilities in the first place, rather than having them designed in by someone who already understands the mechanism? That’s the second way: discovery.
Artist Theo Jansen’s kinetic Strandbeests sculptures offer one answer. Jansen searched through different combinations of leg dimensions using a genetic algorithm, selecting those that produced increasingly effective walking mechanisms.6 Propelled by the wind, the resulting oversized beach creatures appear to walk on their own. While designed by a human, their particular geometry was discovered through an evolutionary process of variation and selection. The intelligence wasn’t written into a programme telling each leg what to do. It was found in the structure itself.
And there’s a third possibility: neither design nor search, but development. In an experiment with the slime mould Physarum polycephalum, researchers placed food sources at positions that corresponded to cities around Tokyo. The organism initially spread through the available space, then strengthened some connections while abandoning others. The resulting network was remarkably similar both visually and in terms of efficiency and robustness to the city’s existing railway network.7 The organism didn’t calculate a map of Tokyo. The solution emerged from growth, reinforcement, and withdrawal.
»Discovery, design, and development are less separate explanations than different routes to the same outcome: useful behaviour folded into a body«
But the categories blur on inspection: Jansen’s »discovery« was still written by a person, and the robots’ »design« by people who understood their materials. The slime mould’s »development« is itself a product of evolution’s much longer search. Discovery, design, and development are less separate explanations than different routes to the same outcome: useful behaviour folded into a body.
And this may be true of humans, too. When a hand catches a falling glass, the brain is certainly involved. But the movement isn’t a miniature physics simulation followed by a conscious instruction to every muscle. Shape, elasticity, reflexes, learned motor patterns, and the dynamics of the body absorb much of the problem.
We tend to think of intelligence as something that happens inside a head and then gets expressed through a body. Perhaps that gets the order backwards. Maybe the remarkable thing about intelligence isn’t how much a system can think, but how much it can get away with not thinking.
1 Haitao Qing, Yuanhang Zhu, Jiacheng Guo, Caizhi Zhou, Haoze Sun, Haibo Dong, Daniel Quinn & Jie Yin (2026) »Minimal-Actuation Feather Star–Inspired Soft Swimmers for Multimodal 3D Maneuverability«. Science Advances, 7 October 2026. https://doi.org/10.1126/sciadv.aeg9211
2 W. Grey Walter (1950) »An Imitation of Life«. Scientific American, May 1950. https://www.scientificamerican.com/article/an-imitation-of-life/
3 Rolf Pfeifer & Josh Bongard (2007) How the Body Shapes the Way We Think. Cambridge, MA: MIT Press.
4 Tianyu Wang, Christopher Pierce, Velin Kojouharov, Baxi Chong, Kelimar Diaz, Hang Lu & Daniel Goldman (2023) »Mechanical Intelligence Simplifies Control in Terrestrial Limbless Locomotion«. Science Robotics, 20 December 2023. https://www.science.org/doi/10.1126/scirobotics.adi2243
5 Max Bennett (2023) A Brief History of Intelligence. New York: HarperCollins.
6 Marcus Fairs (2014) »‘I try to make new forms of life,’ says Strandbeests creator Theo Jansen«. Dezeen, 12 December 2014. https://www.dezeen.com/2014/12/12/strandbeests-theo-jansen-interview-wind-powered-machines-new-species/
7 Atsushi Tero, Seiji Takagi, Tetsu Saigusa, Kentaro Yumiki, Ryo Kobayashi, Dan Bebber, Mark Fricker & Toshiyuki Nakagaki (2010) »Rules for Biologically Inspired Adaptive Network Design«. Science, 22 January 2010. https://www.science.org/doi/10.1126/science.1177894