The coffee test, often called the Wozniak test, is a proposed benchmark for general intelligence put forward by Apple co-founder Steve Wozniak in a 2010 Fast Company interview. Wozniak asked whether a computer could make a cup of coffee, and answered with a concrete challenge: a machine would have demonstrated real general intelligence if it could enter an average, unfamiliar home, find the kitchen, and brew a cup of coffee the way a person would - without any prior knowledge of that specific house.
The test is interesting precisely because it is the opposite of the Turing Test in spirit. The Turing Test measures conversational ability behind a screen, where a system can succeed through language alone. The coffee test demands embodied intelligence: a robot has to perceive a novel physical environment, locate unlabeled objects and appliances it has never seen in that arrangement, improvise around the layout and equipment of a strange kitchen, and carry out a multi-step physical task. It bundles perception, common-sense reasoning, planning, and dexterous manipulation into one everyday goal that humans find trivial and machines find extraordinarily hard.
The coffee test has become a popular informal marker for artificial general intelligence, frequently cited alongside other proposed criteria as a reminder that fluent text generation is not the same as competence in the messy physical world. Despite enormous progress in language models, no general-purpose robot can reliably pass it, which is part of why the test endures as a vivid illustration of the gap between digital and embodied intelligence.
Why a general reader should care: it is a clear, intuitive way to see what current AI still cannot do - operate competently in the unstructured physical world - and a useful counterweight to claims that human-level AI has already arrived.