Are We Projecting Humanity onto Machines? A Thought Experiment with Goats and AI
Let’s start with a question that’s been nagging at me lately: What if the way we interact with technology is shaping our perception of its intelligence far more than the technology itself? This isn’t just a philosophical musing—it’s at the heart of a fascinating thought experiment by Microsoft AI researcher Adrian de Wynter, who built a neural network inside Age of Empires II using digital goats. Yes, goats. And it’s far more profound than it sounds.
Personally, I think what makes this particularly fascinating is how it challenges our assumptions about AI. De Wynter’s experiment isn’t just a quirky tech demo; it’s a critique of how we anthropomorphize machines. By creating a basic neural network within a game—where goats act as bits and grass represents binary inputs—he’s forcing us to confront a simple truth: If we can build something that mimics AI in a game, does that mean the game itself is conscious?
From my perspective, this raises a deeper question: Are we attributing human-like qualities to AI simply because we interact with it through human-like interfaces? De Wynter’s point is that the interface—whether it’s a chat window or a game of Age of Empires II—shapes our perception. When we chat with an AI, we assume it understands us. But if the same technology were embedded in a game, we’d likely see it as just a clever mechanic, not a thinking entity.
One thing that immediately stands out is how easily we project humanity onto machines. De Wynter reviewed hundreds of computer science papers and found that over half assumed AI has human-like traits. This isn’t just a scientific oversight—it’s a cultural phenomenon. We’re so accustomed to treating AI as a conversational partner that we forget it’s a tool, not a person. What many people don’t realize is that this projection isn’t just harmless; it influences how we design, regulate, and even fear AI.
Take the claims that AI models like Claude have a “constitution” or experience anxiety. If you take a step back and think about it, these are marketing tactics as much as they are scientific assertions. De Wynter’s experiment with goats is a brilliant counterpoint: if we can build a neural network in a game, why do we assume AI in a chat window is fundamentally different? The answer, I suspect, lies in how we’re conditioned to interact with technology.
A detail that I find especially interesting is De Wynter’s use of absurdity as a tool. By dialing things up to 11—building an AI in a game using goats—he’s making a point that’s both humorous and profound. It’s a reminder that our perception of AI is often more about us than about the technology itself. What this really suggests is that we need to rethink how we study and talk about AI. Instead of assuming it’s human-like, we should focus on what it actually is: a complex system of algorithms and data.
This raises another point: the role of marketing in shaping our perception of AI. OpenAI’s Sam Altman has hinted at creating an AI god, while Anthropic’s Dario Amodei can’t rule out AI consciousness. These aren’t just scientific hypotheses—they’re narratives designed to capture our imagination (and our wallets). But here’s the thing: if we’re so quick to ascribe consciousness to AI, are we losing sight of its limitations?
In my opinion, the real issue isn’t whether AI is conscious—it’s whether we’re projecting our own desires and fears onto it. De Wynter’s experiment with goats is a wake-up call. It forces us to ask: What if the human-like qualities we see in AI are just a reflection of ourselves?
What makes this particularly intriguing is how it connects to broader trends. From self-driving cars to smart assistants, we’re increasingly interacting with machines that mimic human behavior. But as De Wynter points out, this mimicry doesn’t mean the machines are human. It’s a distinction we’re dangerously close to forgetting.
If you take a step back and think about it, the implications are huge. If we continue to anthropomorphize AI, we risk overestimating its capabilities and underestimating its risks. We might build systems we can’t control, all because we assumed they were more human than they are.
So, what’s the takeaway? Personally, I think we need to stop treating AI as a mirror and start seeing it as a tool. De Wynter’s experiment with goats isn’t just a clever thought experiment—it’s a call to rethink how we interact with technology. Maybe, just maybe, the key to understanding AI isn’t in making it more human-like, but in recognizing its inherent differences.
After all, as De Wynter puts it, ‘It’s not like your neurons know they’re part of a brain.’ And perhaps that’s the most human insight of all.
