If AI Is Sentient Then So Is ‘Age of Empires II’

Key Points:

• Microsoft AI researcher Adrian de Wynter created a basic neural network inside the game Age of Empires II using goats to illustrate that claims of consciousness in large language models (LLMs) are often anthropomorphized and exaggerated.
• De Wynter’s experiment shows that similar underlying technology to LLMs can exist in non-traditional interfaces like AoEII, challenging perceptions that LLMs inherently possess human-like traits simply because they produce conversational outputs.
• His review of 315 computer science papers found that 57% assume LLMs have human-like attributes, which biases experimental design and conclusions, highlighting the problem of anthropomorphizing AI before rigorous scientific analysis.
• De Wynter argues that instead of assuming LLMs behave like humans, research should focus on understanding these models on their own terms and that appropriate disclosures and alignment techniques can help mitigate misplaced human attachment.
• The marketing of LLMs often leverages anthropomorphic perceptions, which benefits companies financially, even as industry leaders speculate publicly about AI consciousness, fueling hype beyond current scientific consensus.