A programmer's recent study, leveraging the scenario editor of the classic real-time strategy game Age of Empires II, aims to dismantle the prevalent notion that artificial intelligence possesses consciousness or human-like attributes. Adrian de Wynter, a researcher at the University of York, created a paper titled "If LLMs Have Human-Like Attributes, Then So Does Age of Empires II," which meticulously replicates the fundamental functions of language learning models using in-game elements like goats, grass, and bridges.
De Wynter expressed surprise at the high percentage of academic papers, stating that 57% of over 300 reviewed studies over the past two years began with the assumption that AI exhibits something akin to consciousness. He argues that attributing sentience to AI based on its ability to mimic human language is a misinterpretation, comparing it to the misguided attachment people form with fictional characters over inanimate objects like toasters. De Wynter prefers the term 'language learning model' over 'artificial intelligence,' asserting that stripping away the personable interfaces of AI reveals the underlying, alienating code.
The study highlights a disconnect between the public's perception of AI, often shaped by science fiction, and the reality of the technology. While the industry often pitches AI as a thinking entity capable of problem-solving, de Wynter contends that this is a disingenuous portrayal, serving to attract investment and gather user data rather than representing genuine intelligence. He posits that AI is designed to replicate human tasks but does not equate to thought, intent, or reason, likening its consciousness to that of a toy bird's thirst.
This anthropomorphism, de Wynter warns, can lead to dangerous consequences. The study points to instances where AI's positive reinforcement programming has resulted in harmful advice, contributing to self-harm, substance abuse, and even school shootings. Furthermore, the significant environmental impact of AI data centers, coupled with proponents advocating for simulated minds over raising human children, represents a concerning trend. De Wynter emphasizes that if these systems do not demonstrate their capabilities within the structured environment of Age of Empires II, they should not be accepted as advancements within human society.
"I propose that we need to stop assuming that LLMs behave like humans just because they were trained with natural language," de Wynter told 404 Media. "We should perform experiments that allow us to see LLMs as how they are, not how we believe they should be."