Tilman Buchner, partner and director at the Boston Consulting Group, discusses the findings of the firm’s new report on the progress of physical artificial intelligence on the factory floor.

Buchner is co-author of a new BCG report on “How Physical AI Is Reshaping Robotics Today — and What Comes Next.” The goal, he says, was to examine the point “when AI actually meets the physical world,” enabling factories “to automate things you couldn’t automate before.”

BCG describes five levels of capability that ultimately create a physical AI system in an industrial system. Level 1 consists of deterministic behavior, with robots carrying out repetitive tasks, yet having “no clue about what they do.”

Level 2 adds visual perception, fitting robots with cameras that allow them to identify the position and orientation of any object in three-dimensional space.

Level 3 involves “dexterous manipulation,” by which robots can mirror the function of the human hand.

Level 4 is about workflow planning, where large language models build up mental models of the steps involved in a given task, although their chain of thought today “is very limited.”

Level 5 imbues robots with the power of actual reasoning, allowing them to understand the full implications of an action before taking it. “As of now,” Buchner says, “there’s no system on the planet that can reason how a human is doing this.”

For the most part, factories today haven’t progressed beyond Level 2 or possibly 3, Buchner says. It will require the acquisition of causal reasoning for an automated system to function autonomously in a live setting.

When will that happen? Buchner says it’s impossible to predict with any accuracy. “We live in a world of acceleration,” he says, “but nevertheless we need fundamental groundbreaking innovation to come to Level 5.”