The Goats of Consciousness: Why We Need to Rethink AI Sentience
The debate around AI sentience has reached a fever pitch, with some claiming large language models (LLMs) possess human-like consciousness. But what if I told you that by this logic, Age of Empires II might also be sentient? This isn’t just a cheeky thought experiment—it’s a profound critique of how we anthropomorphize technology. Let me explain.
The Absurdity of Goats as Neurons
Microsoft AI researcher Adrian de Wynter recently made waves by building a basic neural network within Age of Empires II using digital goats. Yes, goats. In his paper, ‘If LLMs Have Human-Like Attributes, Then So Does Age of Empires II’, de Wynter demonstrates that the same underlying technology powering LLMs can be replicated in a 23-year-old video game. Grass represents 0, bridges represent 1, and goats act as signal carriers. It’s absurd, but that’s the point.
What makes this particularly fascinating is how it exposes our tendency to project human traits onto machines. We don’t think of goats in Age of Empires II as conscious beings, even though they’re part of a functioning neural network. Yet, when it comes to LLMs, we’re quick to ascribe emotions, morality, and even anxiety to them. Why the double standard?
In my opinion, it’s because of the interface. Chatbots mimic human conversation, so we naturally anthropomorphize them. But as de Wynter points out, the underlying technology is the same—whether it’s goats in a game or tokens in a model. This raises a deeper question: Are we mistaking interface design for consciousness?
The Anthropomorphism Trap
De Wynter’s work highlights a troubling trend in AI research. After reviewing 315 computer science papers, he found that 57% assumed LLMs possess human-like traits from the outset. This isn’t just sloppy science—it’s a fundamental misunderstanding of what LLMs are. They’re not human-like; they’re something entirely new.
One thing that immediately stands out is how marketing exacerbates this issue. Companies like OpenAI and Anthropic often frame their models as god-like entities or beings with “constitutions.” It’s great for selling products, but it distorts public perception. Personally, I think this is dangerous. If we start treating LLMs as conscious beings, we risk overlooking their limitations and ethical implications.
Consciousness: A Spectrum, Not a Binary
What many people don’t realize is that consciousness isn’t a yes-or-no question. It’s a spectrum. We confidently say humans are conscious, but what about dogs? Ants? Plants? LLMs? De Wynter argues that we’re too quick to apply human-centric definitions to something we don’t fully understand.
If you take a step back and think about it, our surprise at bumblebees solving problems reveals more about our biases than about the bees. We’re amazed because we underestimated them, not because they suddenly became human-like. The same goes for LLMs. They’re not human, but they’re not just “blobs of weights” either. They’re something in between—and that’s okay.
Breaking the Illusion
De Wynter’s goats serve as a brilliant metaphor for breaking the illusion of AI consciousness. By stripping away the chat interface, he forces us to see LLMs for what they are: complex systems, not sentient beings. This isn’t to say LLMs can’t have emergent properties—they absolutely can. But those properties aren’t the same as human consciousness.
A detail that I find especially interesting is how users get attached to LLMs when they appear warm or personable. We don’t get attached to our toasters, but we do to movie characters. LLMs occupy this strange middle ground, and companies exploit that. What this really suggests is that we need better alignment techniques and transparency in how these models are presented.
The Way Forward
So, where do we go from here? De Wynter’s work isn’t just a critique—it’s a call to action. We need to stop assuming LLMs are human-like and start studying them on their own terms. This means designing experiments that focus on their capabilities, not their perceived consciousness.
In my opinion, the AI industry also needs to tone down the hype. Selling LLMs as god-like entities isn’t just misleading—it’s irresponsible. As de Wynter puts it, these models are products, not beings. Let’s treat them as such.
Final Thoughts
The goats of Age of Empires II may not be conscious, but they’ve taught us a valuable lesson. Consciousness isn’t about the interface or the marketing—it’s about the underlying nature of a being. LLMs are remarkable tools, but they’re not human. And that’s perfectly fine. Let’s stop projecting our own traits onto them and start appreciating them for what they truly are: a new frontier in technology, not a mirror to our souls.
