Organic Computing and Embodied Intelligence
Jonas Lauren Norr
Pushing the boundaries of compute also means reconsidering where computation begins. A material can respond to light, retain a trace of an earlier input, or change its behaviour under stress. If those properties can be controlled and read reliably, the material becomes part of the processing system. Organic computing and embodied intelligence are interesting because they invite us to develop machines around those capabilities, extending the range of physical systems that can do useful computational work.
The terms cover different approaches. Organic electronic materials can be synthetic and nonliving. Biological computing uses living systems as functional components. Embodied intelligence concerns the contribution of a system’s physical structure and interaction with its environment to its behaviour. These fields overlap, but their differences matter. A polymer device, a neural organoid, and a fungus connected to a robot involve different mechanisms, operating conditions, and development challenges. Their progress should be assessed through the specific functions they demonstrate. [1][2][3]
One route brings processing closer to sensing. A September 2025 paper in Light: Science & Applications described organic polymer memristors that combine optical sensing and information processing within the same devices. A memristor’s electrical response depends on its previous state, allowing physical dynamics to participate in computation. The researchers demonstrated a compact reservoir computing system for fingerprint recognition. The broader possibility is a sensing surface that performs useful interpretation before passing information to another processor. [1]
Living neural tissue provides a different example. The Brainoware work published in Nature Electronics in December 2023 used a brain organoid connected to an electrode array as a computational reservoir. The researchers demonstrated speech recognition and prediction of a nonlinear equation within that experimental framework. It is a striking demonstration of biological dynamics participating in a computing system. It does not establish that an organoid can perform the general functions of a data centre, or that the complete supporting system has a proven energy advantage. [2]
Embodiment makes the opportunity more tangible. In 2024, Cornell researchers reported two robots controlled using electrical activity from fungal mycelia. Their interface processed the biological signals into commands for artificial actuators, and ultraviolet light changed the robots’ movement. The living material was participating in a sensing and control loop. The electronics and engineered control architecture remained essential. This is an example of combining biological responsiveness with a machine’s ability to act on its surroundings. [3]
What interests me is the possibility of designing sensing, processing, and action together. A conventional development process often treats these as separate components to be connected later. Organic and biohybrid systems encourage a different starting point: which useful behaviours are already available in the material, and what supporting architecture would make them dependable? That could lead to new forms of environmental sensing, adaptive surfaces, or machines suited to particular physical settings. Those are development possibilities, rather than capabilities established by every laboratory demonstration.
The infrastructure around these systems will be decisive. A promising material needs an interface that translates its behaviour into something a developer can use. It needs calibration, a way to distinguish meaningful response from drift, and a workable maintenance and replacement process. For living components, the environment that keeps the material functional belongs inside the system boundary. Any comparison of energy use or cost should include those requirements, together with the electronics and data processing around them.
This creates an opportunity for shared development tools as well as individual devices. Reference tasks, documented interfaces, repeatable measurements, and comparisons with conventional alternatives would make it easier to identify where an organic approach offers a useful advantage. A good early application may have a modest computational requirement but an unusual physical one. Matching the material to that setting could be more productive than asking every new architecture to compete immediately on the broadest computing benchmarks.
The frontier here is a wider choice of ways to build intelligence into physical systems. Some approaches may remain specialist research tools; others may become components in practical machines. I am interested in the work that allows those possibilities to be tested seriously. Materials research, biological understanding, electronics, and compute infrastructure have to meet in a system that can be understood and sustained. That is where an intriguing physical behaviour can begin to become a useful technology.
Sources
[1] Zhou et al., Optoelectronic polymer memristors with dynamic control for power-efficient in-sensor edge computing, Light: Science & Applications, September 8, 2025.
[2] Cai et al., Brain organoid reservoir computing for artificial intelligence, Nature Electronics, December 11, 2023.
[3] Mishra et al., Sensorimotor control of robots mediated by electrophysiological measurements of fungal mycelia, Science Robotics, August 28, 2024. See also Cornell’s account of the experiment.
Pushing the boundaries of compute also means reconsidering where computation begins. A material can respond to light, retain a trace of an earlier input, or change its behaviour under stress. If those properties can be controlled and read reliably, the material becomes part of the processing system. Organic computing and embodied intelligence are interesting because they invite us to develop machines around those capabilities, extending the range of physical systems that can do useful computational work.
The terms cover different approaches. Organic electronic materials can be synthetic and nonliving. Biological computing uses living systems as functional components. Embodied intelligence concerns the contribution of a system’s physical structure and interaction with its environment to its behaviour. These fields overlap, but their differences matter. A polymer device, a neural organoid, and a fungus connected to a robot involve different mechanisms, operating conditions, and development challenges. Their progress should be assessed through the specific functions they demonstrate. [1][2][3]
One route brings processing closer to sensing. A September 2025 paper in Light: Science & Applications described organic polymer memristors that combine optical sensing and information processing within the same devices. A memristor’s electrical response depends on its previous state, allowing physical dynamics to participate in computation. The researchers demonstrated a compact reservoir computing system for fingerprint recognition. The broader possibility is a sensing surface that performs useful interpretation before passing information to another processor. [1]
Living neural tissue provides a different example. The Brainoware work published in Nature Electronics in December 2023 used a brain organoid connected to an electrode array as a computational reservoir. The researchers demonstrated speech recognition and prediction of a nonlinear equation within that experimental framework. It is a striking demonstration of biological dynamics participating in a computing system. It does not establish that an organoid can perform the general functions of a data centre, or that the complete supporting system has a proven energy advantage. [2]
Embodiment makes the opportunity more tangible. In 2024, Cornell researchers reported two robots controlled using electrical activity from fungal mycelia. Their interface processed the biological signals into commands for artificial actuators, and ultraviolet light changed the robots’ movement. The living material was participating in a sensing and control loop. The electronics and engineered control architecture remained essential. This is an example of combining biological responsiveness with a machine’s ability to act on its surroundings. [3]
What interests me is the possibility of designing sensing, processing, and action together. A conventional development process often treats these as separate components to be connected later. Organic and biohybrid systems encourage a different starting point: which useful behaviours are already available in the material, and what supporting architecture would make them dependable? That could lead to new forms of environmental sensing, adaptive surfaces, or machines suited to particular physical settings. Those are development possibilities, rather than capabilities established by every laboratory demonstration.
The infrastructure around these systems will be decisive. A promising material needs an interface that translates its behaviour into something a developer can use. It needs calibration, a way to distinguish meaningful response from drift, and a workable maintenance and replacement process. For living components, the environment that keeps the material functional belongs inside the system boundary. Any comparison of energy use or cost should include those requirements, together with the electronics and data processing around them.
This creates an opportunity for shared development tools as well as individual devices. Reference tasks, documented interfaces, repeatable measurements, and comparisons with conventional alternatives would make it easier to identify where an organic approach offers a useful advantage. A good early application may have a modest computational requirement but an unusual physical one. Matching the material to that setting could be more productive than asking every new architecture to compete immediately on the broadest computing benchmarks.
The frontier here is a wider choice of ways to build intelligence into physical systems. Some approaches may remain specialist research tools; others may become components in practical machines. I am interested in the work that allows those possibilities to be tested seriously. Materials research, biological understanding, electronics, and compute infrastructure have to meet in a system that can be understood and sustained. That is where an intriguing physical behaviour can begin to become a useful technology.
Sources
[1] Zhou et al., Optoelectronic polymer memristors with dynamic control for power-efficient in-sensor edge computing, Light: Science & Applications, September 8, 2025.
[2] Cai et al., Brain organoid reservoir computing for artificial intelligence, Nature Electronics, December 11, 2023.
[3] Mishra et al., Sensorimotor control of robots mediated by electrophysiological measurements of fungal mycelia, Science Robotics, August 28, 2024. See also Cornell’s account of the experiment.