Phoenix 24
A Microsoft engineer exposes the mathematical processes behind apparently human chatbot behaviour.
New York, June 2026. A Microsoft engineer has used the strategy video game Age of Empires II to illustrate why modern artificial intelligence systems should not be considered conscious. Adrian de Wynter, a Microsoft researcher and collaborator with New York University, designed the experiment to challenge the growing tendency to attribute emotions and self-awareness to chatbots. His work argues that apparently intelligent responses emerge from mathematical operations rather than genuine understanding or subjective experience. The demonstration seeks to separate the persuasive appearance of conversational systems from the computational mechanisms operating beneath their interfaces.
The project emerged amid renewed public debate about whether advanced language models could possess some form of consciousness. Biologist Richard Dawkins recently suggested that Claude, an artificial intelligence model developed by Anthropic, might be conscious even without recognising its own condition. Similar claims had previously been made by former Google engineer Blake Lemoine, who argued that an experimental chatbot showed signs of sentience. De Wynter considered these interpretations examples of anthropomorphism encouraged by the fluency and apparent emotional sensitivity of modern digital assistants.
To demonstrate his argument, the researcher used the scenario editor included in Age of Empires II. This tool allows players to create customised environments by combining characters, landscapes, buildings and other elements from the game. De Wynter recreated a perceptron, one of the simplest forms of artificial neural network and a fundamental concept in machine-learning development. His unconventional construction used grass to represent zero, bridges to represent one and goats to function as digital bits.
The researcher then arranged these elements to build NAND logic gates, which can serve as fundamental components for computational operations. This structure allowed him to simulate a functioning neural network entirely within the medieval strategy game. Instead of a friendly conversational interface, observers saw goats moving across a virtual landscape while executing predefined logical instructions. The visual contrast demonstrated that processes presented as intelligence or empathy can ultimately be reduced to calculations and programmed transformations of information.
Chatbots such as ChatGPT, Gemini and Claude communicate through coherent language that can create the impression of intention, personality and emotional understanding. Their responses may feel personal because they are designed to predict and organise words in ways that resemble human communication. De Wynter argues that removing the conversational interface reveals the absence of an experiencing mind behind the generated text. The systems process patterns and probabilities without necessarily possessing awareness of their answers, their users or themselves.
The experiment does not suggest that artificial intelligence lacks usefulness, sophistication or transformative potential. Instead, it warns against confusing advanced linguistic performance with consciousness and human-like cognition. People naturally assign intentions and emotions to systems that communicate coherently, even when those systems operate through impersonal mathematical procedures. Understanding that distinction could help users interact with artificial intelligence more critically and avoid developing unrealistic perceptions of digital companionship.
Behind the chatbot’s convincing words remains a complex computational system rather than a conscious mind.