Physical Artificial Intelligence
Physical Artificial Intelligence (PAI)
"PAI refers to the theory and practice of creating physical systems capable of performing tasks that are typically associated with intelligent organisms. The PAI methodology inherently combines the considerations of materials, design, and manufacturing. Robots developed using PAI may autonomously perform tasks and maintain homeostasis in unstructured environments by exploiting the physical and computational features of their bodies along with the computational abilities of their "brains". Akin to biological organisms, PAI robotics can either replace digital AI or synergistically complement it by interfacing to a "brain". Many small robots, or those with limited computational functionality, would not include a dedicated central "brain," and their body's computation will guide their performance. Similar to the principles of natural diversity, PAI synthesis refers to robotic systems of any functionality, shape, size, and habitat, with particular emphasis on integrating chemical, biological, and material based functionalities."
"A number of disciplines are closely intertwined in the ambitious venture of creating PAI. Mainly, these are materials science, mechanical engineering, computer science, chemistry, and biology. Leading the paradigm shift from robots as an assembly of hard devices to PAI-comprised robots requires a combination of skills in these disciplines. The domains of each discipline indicate the great plurality and diversity of the desirable skills, leaving space for even more. However, such a broad range of skills for a single person is difficult to obtain today, and no systematic educational approaches exist to develop them."
"The following multi-level methodology can be proposed for implementing PAI in the current higher education programs. We suggest that the methodology needs to encompass the following three components: 1) institutional support, 2) appreciation of the complexity and challenges of working in interdisciplinary research, and 3) existence of venues for research recognition and dissemination."
Physical Artificial Intelligence (PAI) refers to the theory and practice of synthesizing lifelike robots. This is achieved by using skills from the disciplines of materials science, mechanical engineering, computer science, chemistry and biology to evolve a fully-functional robot (in the illustration: robotic bee). The process incorporates the considerations of materials, structure, sensing and actuation, and results in a fully autonomous multi-functional robot. Adopted from "Skills for Physical Artificial Intelligence", Aslan Miriyev, Mirko Kovac, Nature Machine Intelligence, 2, 658-660, 2020.
Physical Artificial Intelligence (PAI) is comprised five main disciplines: materials science, mechanical engineering, computer science, biology and chemistry. Adopted from "Skills for Physical Artificial Intelligence", Aslan Miriyev, Mirko Kovac, Nature Machine Intelligence, 2, 658-660, 2020.
Nature Machine Intelligence comment paper: Miriyev, A., Kovač, M. Skills for physical artificial intelligence. Nat Mach Intell 2, 658–660 (2020). https://doi.org/10.1038/s42256-020-00258-y
Paper full text pdf
Nature Machine Intelligence editorial on the topic of PAI: Materializing artificial intelligence. Nat Mach Intell 2, 653 (2020). https://doi.org/10.1038/s42256-020-00262-2
Imperial College London press release on PAI
Cover of Nature Machine Intelligence, Issue of November 2020.
Based on the "Hands of creation" image, showing human hands creating PAI by using skills from 5 disciplines. Image credit: Aslan Miriyev Mirko Kovac / Empa and Imperial College London; Cover design: Karen Moore.