Next generation of AI
Smart robots need intelligent materials
Artificial intelligence (AI) is nowadays considered a fundamental asset for developing products and technologies across industries, giving companies who master it a competitive edge. What is still lacking, however, is a thorough discussion about the implications for society to better understand opportunities but also potential drawbacks. In an interview Mirko Kovac and Aslan Miriyev from Empa's Center of Robotics share their views about where AI and robotics research is headed.
You are about to host a virtual international workshop termed "Material Intelligence" – how can a material acquire intelligence? How do we have to envision a material that you would deem "intelligent"?
Mirko Kovac: The general definitions of "intelligence" typically refer to the interaction between an "agent" and its environment, such as an ability to learn, understand, react and/or adapt to situations and/or changes in the agent's vicinity. An agent can be any organism, say, a fish, a cat or a human being. In our case, the agent is a robot. With regards to materials, the intelligence is in their ability to react to external stimuli. But the challenge is in taking this from a passive, more or less predefined reaction to stimuli to an intelligent adaptation to a changing environment. This means that, for instance, a material, which responds to heat by changing its shape, will do so while adapting to the external space conditions or limitations. We cannot change the laws of physics, but we can help the material to adapt using clever design. So, when we talk about material intelligence we rather think of a responsive system of matter and design.
Aslan Miriyev: A significant benefit here would be reversibility and high repeatability of the intelligent behavior. Such systems may be used in morphological computation that utilizes shapes and material properties as a means of computing. Thus, in the Workshop, we discuss this path from functional materials as a passive substance reacting to stimuli to a system serving as a computing mean.
What is the link between intelligent materials and artificial intelligence (AI)? And where or how would robotics fit into the picture?
Miriyev: In the context of the previous question, material systems comprise the physical artificial intelligence of robots' bodies, similarly to the way biological tissues comprise the intelligence of our bodies. Synthetic and bio-hybrid materials are the key to collaborative robots that will co-exist with humans in future symbiotic human-robot ecosystems.
In recent years, AI research has moved away from a mere algorithm-based number-crunching approach to a more physical take on intelligence, called Physical AI. Where is this kind of research headed?
Miriyev: Indeed, the field of AI has been intensively growing in the past decades. Now, it has become clear that AI is not only the digital AI, which we all know, for example, from its ability to recognize, based on camera images, a pack of chips among a mess of many items on a table. AI is also about physical interaction with the world, with humans, the environment, nature, plants, animals.
Kovac: We envision the research going in the direction of developing the physical side of AI, and its merging with the digital counterpart. Recently, we described these ideas in a paper published in the journal Nature Machine Intelligence. We define Physical AI (or, PAI) as "the theory and practice of creating physical systems capable of performing tasks that are typically associated with intelligent organisms."
Which area or industry could benefit most from this type of AI? Do you have any specific examples?
Kovac: The areas of impact of Physical AI include healthcare, elderly care, infrastructure digitalization, disaster management, public safety, security, services sector, education, and industrial automation. The entire Industry 5.0 concept is based on real human-robot interaction. So far, most of the successively commercialized soft robotic solutions were in the pick-and-place, wearable, prosthetic, and minimally-invasive medical applications.
Miriyev: The current use of Physical AI is still very limited compared to the challenges of the world around us and to the opportunities that lie ahead. There is almost infinite room for the application of robots capable of co-living and co-working with people.
You are working on autonomous drones, or robots, if you wish, that are inspired by biological principles. How would you define their role, their function in interacting with humankind?
Kovac: In 2013 I published a paper entitled “The Bioinspiration Design Paradigm: A Perspective for Soft Robotics”. The principle may be divided into three stages: inspiration, abstraction, and implementation. In the inspiration phase, a scientist observes a natural phenomenon. In the subsequent abstraction phase, a scientist tries to find the underlying physical principles of the observed phenomenon. After that comes the implementation phase, in which scientist engineer the abstracted bioinspired design principles into a synthetic system. Specifically, we are now inspired by the process of creation of new living organisms as a new paradigm for creating artificial systems.
Miriyev: In Physical AI synthesis, knowledge from multiple disciplines contributes to the simultaneous creation of structures, sensing, actuation, and computation. In the coming Workshop we intend to discuss the paths to such advanced robot synthesis via the utilization of various aspects of material intelligence.
Given the number of dystopian Sci Fi movies/novels, you must face a certain degree of skepticism when you talk about achieving a symbiosis of man and machine. What are the biggest challenges in this arena and how do you address them?
Kovac: In the chase after fictional dramas, we often forget that real life's calls are much more obvious than plots of sci-fi scenarios. Elderly people really do need reliable assistance well beyond telepresence humanoids.
Miriyev: Society needs lifelike robotic assistants that adapt to them in the manner, speed and language, but at the same time adapt to the environment in the associated speed and language to provide all necessary help to their elderly human companions. This should be done without compromising the quality of humans' lives and/or their privacy and data. We need to think about creating useful items for people, for their everyday life.
So where are we in, say, 20 years from now with regard to AI, robotics and the like?
Kovac: Robotics becomes a fabric of our physical world and a digital tissue of our infrastructures and cities. Robotics will disappear as a term in many ways, and we will all become very used to engaging with synthetic intelligence in our everyday life. The frontier for the decades to come will be in bio-hybrid systems in a combination of robotics and living materials. This is the topic we also explore at Empa's Center of Robotics.
Join the Workshop "Material Intelligence: From Functional Materials to Material Computing" on 12 April 2021.
Aslan Miriyev (left) and Mirko Kovac are among the hosts of the workshop.
A Miriyev, M Kovac; Skills for Physical Artificial Intelligence; Nature Machine Intelligence, 2 (2020); Image: Empa / Imperial College London
PAI: Physical Artificial Intelligence: On the way to lifelike robots; media release, 11 Nov 2020
Forest dwelling drones: Drones to monitor ecological changes; media release, 3 Nov 2020
Mirko Kovac: Vertical take-off; media release, 14 Jan 2020