The limitations of the machine learning approach to AI mean that powerful AI tends to be hidden in static systems that have to be connected to data centres. However, the goal of many AI developers is to have AI systems embedded in machines that operate dynamically within the human environment.13 This does not require human-level intelligence, but it does require a degree of flexibility and adaptability, and an ability to sense and react to moving objects and changing conditions in the human environment, as well as dexterity and agility in manipulation of objects and human-safe operation.14
This is a significant challenge, but one that could create a new era in our interactions with machines, potentially bringing a sea-change to medical care (especially in an ageing population), industrial production and education, among other areas. It will require advances in sensors, the processing of sensor data, interface design and autonomous decision-making. Researchers anticipate that these kinds of advances will accelerate as the commercial potential for embodied AI begins to be realised.