Robots and Artificial Intelligence: A Body and a Mind That May Join or Part

In 1961 the first industrial robot was installed in a General Motors plant in New Jersey. Called Unimate, it moved die castings along the production line and welded them onto car bodies, so that people no longer had to do that hazardous work, according to Guinness World Records. There was nothing in that robot that we would today call artificial intelligence, and yet nobody doubts it was a robot. Here lies the root of the confusion.

What is a robot?

The international standard ISO 8373 (2021 edition) defines a robot as an actuated mechanism, programmable in two or more axes, with a degree of autonomy, moving within its environment to perform intended tasks. The definition speaks of movement, programming and autonomy, not of learning or understanding. The industry is enormous quite apart from AI: the International Federation of Robotics (IFR) counted about 542,000 new industrial robots installed worldwide in 2024, more than double the number ten years earlier, with an operational stock of about 4.66 million robots, and China accounted for 54% of installations.

What is artificial intelligence?

The field was born as a research programme when John McCarthy, Marvin Minsky, Nathaniel Rochester and Claude Shannon proposed, in a document dated August 1955, a summer study at Dartmouth College in 1956. It rested on the conjecture that every aspect of learning, or any other feature of intelligence, can in principle be described so precisely that a machine can be made to simulate it. AI is, at its core, a software field: algorithms and models that learn from data and reason, with no need for a body. The chat assistant that writes you a message has no arm and no wheels. For more on what AI is and is not, see the article "What Is Artificial Intelligence, Really? A Plain Explanation Without the Hype".

When do they meet?

The two fields meet when a mechanical body is given a system that learns. In March 2025 Google DeepMind announced Gemini Robotics, a vision-language-action model built on Gemini 2.0 and extended so that physical actions are among its outputs, allowing it to control robots directly. The company says it generalises to objects, instructions and environments it has not seen, and adjusts when the object to be handled moves or slips. That is the developer's own claim, not an independent evaluation. What matters here is that the combination works by joining a software mind to a body, and in any composite system we can still tell the two components apart: what is physical movement, and what is information processing.

Why does the difference matter?

Because each has its own risks. A robot that moves among people is governed by considerations of physical safety and liability for material damage, whereas a software system that decides on credit or shortlists a candidate for a job is governed by questions of bias, privacy and transparency (see the articles "Algorithmic Bias" and "Privacy in the Age of AI: How Your Secrets Are Inferred, and How to Protect Them"). And because anyone who confuses them may direct money and training to the wrong place: building capability in data and models calls for people, data and computing, while building robotic capability calls for factories, sensors and mechanical engineering. Both are legitimate investments, but they are not the same thing.

When you meet a new system, ask three questions. Where does its essence lie, in physical movement or in information processing? Does it learn and adapt, or only carry out what it was programmed to do? And what kind of risks does it carry, physical or cognitive? The answers return each system to its place, so that we neither fear a machine without a mind nor overlook a mind without a body that shapes our decisions where we cannot see.