As The recent CES provided a powerful showcase for humanoid and other adaptive robots. The demonstrations served as an important reminder that embodied AI is the next AI frontier. Beyond the technology challenges that remain before conquering this frontier, how should we view embodied AI? Will it be the cause of massive job losses in various industries, or an opportunity to improve productivity by forming collaborative teams that combine humans with intelligent machines? We define four dimensions to answer these questions.
Introduction
Embodied AI refers to autonomous agents with a physical form that enable them to perceive, interact with, and learn from the physical world. Unlike digital-only autonomous agents, embodied AI agents use sensors and computer vision to navigate and adapt to environments in real time. Embodied AI agents are also called adaptive robots because of their ability to adapt to their environment.
Those of us who live in the San Francisco Bay Area have become used to seeing Waymo’s robotaxis. These vehicles today represent the most visible form of embodied AI. A less visible but equally important use case is Amazon’s deployment of thousands of adaptive robots in its warehouses. Autonomous transportation and autonomous warehouse operations represent embodied AI’s first successful use cases. In addition, several industries, including the automotive industry, are actively engaging in pilot projects with humanoid robots.
At CES, humanoid robots and other embodied AI form factors stole the show with the feats they demonstrated and the imaginations they conjured.
Questions on how to think about embodied AI are starting to emerge with increasing frequency. For example, my frequent collaborator Joe White and I were recently discussing whether the automotive industry assembly line workers should be concerned about being replaced within a few years by adaptive robots in the factories currently under development.
Technologists, investors, and corporate executives are in active discussions about “dark assembly lines,” “dark warehouses,” “dark industrial kitchens,” and other types of “dark” spaces, i.e., production spaces filled with robots and devoid of humans.
Four Dimensions to Assess Adaptive Robots
It is easy for embodied AI to fill us with dystopian visions. However, before this happens, we must use four dimensions to assess the deployment of adaptive robots: task complexity, task risk to humans, industry worker demographics, and industry growth.
Task complexity determines whether a task can be performed exclusively by an adaptive robot or requires the robot to collaborate with a human. For example, e-commerce companies today use adaptive robot ” pickers “ in their fulfillment centers to pack boxes with items of the same SKU. This is a task that adaptive robots can reliably accomplish today on their own. On the other hand, in Amazon’s fulfillment centers, adaptive robots bring items to human pickers who subsequently place them in the appropriate shipping boxes.
Task risk determines the level of risk the human is exposed to while accomplishing a task on their own. Mining and construction involve tasks that expose humans to a high risk of injury and disease. Having adaptive robots perform the dangerous tasks of these industries would be a welcome achievement.
Worker demographics reflect whether an industry experiences a manpower shortage. For example, agriculture and certain healthcare sectors, e.g., nurse assistants and caregivers to the elderly, are experiencing staffing shortages. Adaptive robots can alleviate such shortages.
Industry growth. E-commerce, logistics, hospitality, and several other industries are growing. Many of their business processes include tasks that require the collaboration between a human and an adaptive robot. In these cases, the companies employing these robots are looking for productivity improvements even before considering cost-cutting.
Other industries, due to their maturity and lack of meaningful growth, will adopt adaptive robots to drive material cost-cutting. The US automotive industry is a prime candidate for adaptive robots. The annual auto demand peaked in 2016 and will not grow substantially. Automakers are testing humanoid robots for this reason. Even more radically, Tesla started to pivot towards becoming a developer and manufacturer of such robots. Certain service sectors are also candidates.
Conclusion
Embodied AI is AI’s next frontier. Adaptive robots of various form factors will shortly find their way to many other industries and business processes. Ample venture and corporate funding support the development of world models and robots. I presented a way to organize industries and tasks using four dimensions. This organization will help us determine how to:
Build the appropriate technologies and value chains to make the broad deployment of adaptive robots a reality.
Train and upskill employees of growth industries on how to effectively collaborate with adaptive robots to ensure continued productivity improvements.
Prepare the employees of mature industries with the means to address their displacement by these robots as a result of cost-cutting.
