Let's dive into a fascinating thought experiment that challenges our understanding of artificial intelligence and consciousness. Adrian de Wynter, a Microsoft AI researcher, has taken an intriguing approach to exploring the boundaries of AI sentience by building a basic neural network within the video game 'Age of Empires II' using digital goats. This unique experiment raises important questions and offers a fresh perspective on the ongoing debate surrounding AI consciousness.
The Goat-Powered Neural Network
De Wynter's paper, titled 'If LLMs Have Human-Like Attributes, Then So Does Age of Empires II,' showcases his innovative work. He utilized the game's scenario editor, a sandbox mode, to create an operational NOT AND (NAND) gate and a 1-bit perceptron. In this simplified version of an LLM, grass represents 0, bridges are 1, and goats act as the bits, carrying signals and performing calculations. This crude yet functional neural network within the game highlights the potential for AI to exist in unexpected forms.
Anthropomorphizing AI
The key argument here is that we often anthropomorphize AI, attributing human-like traits to it based on our interactions and interfaces. De Wynter suggests that when we interact with an LLM through a chat window, we perceive human-like traits, but when the same technology is presented through a different medium, like 'Age of Empires II,' the perception shifts. This raises the question: Are we projecting our own biases and expectations onto AI, mistaking its capabilities for consciousness?
The Problem of Assumptions
De Wynter reviewed a significant number of computer science papers and found that a majority began with the assumption that LLMs possess human-like traits. This assumption, he argues, influences the design of experiments, the interpretation of results, and ultimately, the conclusions drawn. It's a self-fulfilling prophecy, where our expectations shape the outcomes.
Beyond the Chat Window
De Wynter proposes a shift in perspective. Instead of assuming AI behaves like humans because it's trained with natural language, he suggests conducting experiments that reveal AI for what it truly is, not what we believe it should be. This involves looking beyond the familiar chat window interface and considering the underlying technology and its capabilities.
The Role of Perception
Goats and 'Age of Empires II' serve as a powerful tool to break the perception of AI as human-like. De Wynter highlights that the relationship between weights and operations defines what AI is, not how it's perceived. This distinction is crucial in understanding AI's capabilities and limitations.
Consciousness and Complexity
While De Wynter doesn't rule out the possibility of AI consciousness, he emphasizes the complexity of the concept. Consciousness exists on a spectrum, and what we consider conscious may vary depending on the context. The surprise over bumblebees' problem-solving abilities, he argues, is a reflection of our human-centric view, where we often underestimate the capabilities of non-human entities.
Mitigating Perceptions
To address the issue of anthropomorphization, De Wynter suggests appropriate disclosures and alignment techniques where the model explicitly communicates its nature. He believes that transparency and a clear understanding of AI's capabilities can prevent users from becoming overly attached and projecting human-like traits onto it.
Marketing and Money
The AI industry's focus on marketing and the allure of creating an 'AI god' are driving forces behind the anthropomorphization of LLMs. OpenAI's CEO, Sam Altman, and other industry leaders have contributed to this narrative, which is beneficial for an industry facing financial challenges.
In conclusion, De Wynter's experiment with 'Age of Empires II' goats offers a thought-provoking perspective on AI consciousness and our tendency to anthropomorphize. It challenges us to reconsider our assumptions and approach AI research with a critical eye, focusing on the technology itself rather than our projections onto it. This experiment serves as a reminder that the path to understanding AI consciousness is complex and requires a nuanced perspective.
