There’s an interesting debate playing out in AI research circles at the moment that highlights how easily we project human qualities onto systems that may not warrant them, and how that habit can shape the conclusions we draw.
Science fiction writer Ted Chiang asked readers to consider Microsoft Word. He wrote: “Being open to the possibility that LLMs are conscious is the same as being open to the possibility that Microsoft Word is conscious, or, more precisely, that multiple distinct consciousnesses are dormant in every Word document containing a conversational transcript, and that they are awakened every time the document is loaded.
“Should you consider the possibility that every time you open a Word document, you are bringing multiple conscious interlocutors into existence, and every time you close one, you snuff their existence out? No. Contemplating that scenario is not a good use of your time.”
If LLMs have human-like attributes, then so does Age of Empires
A Microsoft AI researcher named Adrian de Wynter took a related idea and ran with it in a more literal direction. He built a basic neural network inside the strategy game Age of Empires II using in-game goats as the underlying “bits,” and published a paper titled “If LLMs Have Human-Like Attributes, Then So Does Age of Empires II.”
In comments I came across discussing the project, he explained that absurdity can be a useful rhetorical tool: “I have this tendency to dial up things to 11 when I really think I need to make a point,” adding, “I should also note that absurdism is pretty standard in philosophy and theoretical computer science.”
The mechanics of the build are fairly straightforward once explained. De Wynter used the game’s scenario editor — a sandbox tool for building custom maps and quests — to construct a working NAND gate and a 1-bit perceptron, with grass representing 0, bridges representing 1, and goats acting as the signal carriers.
As he put it on GitHub: “Only one rail is active at a time, with a goat acting as the signal carrier. When the gate fires, the bit-goats are removed (they ded) and a new bit-goat is placed in its respective output rail.” A perceptron is about as simple as neural networks get (an algorithm that sorts inputs into one of two categories), and similar logic-gate builds have long existed in games like Minecraft using redstone. Nobody has seriously argued that those constructions are evidence of emergent intelligence.
That’s De Wynter’s point. The underlying mechanism in his Age of Empires build is functionally the same type of structure that sits underneath large language models, like Claude, ChatGPT, and Copilot, simplified considerably but built from the same basic principles. If we’d readily dismiss the idea that goats on a game map are conscious, it’s worth asking why some of the language used around commercial AI products, such as references to a model having a “constitution,” or experiencing something like anxiety, gets treated differently.
De Wynter’s broader argument is that the perception of human-like qualities in these systems often comes down to interface rather than substance. A simplified LLM hidden inside a strategy game doesn’t read as humanlike to anyone, while the same underlying technology presented through a chat window often does.
He chose the game deliberately, having played it since its 1999 release. In an interview about the work, he said: “Age of Empires was an excellent way to drive the point home. It is just about ‘alien’ enough to exemplify the representation-interpretation relation, but sufficiently well-known to really emphasise the point. It also works at a meta-level, since the example itself is a good representation of the argument.”
The assumption that LLMs possess human-like traits is rife
He also looked at how this plays out in the research literature itself. After reviewing 315 computer science papers published over the past two years, he found that 57% started from the assumption that LLMs possess human-like traits.
As his paper puts it: “What is common to some of these studies […] is that they test and ascribe blanket human-like properties (e.g., anxiety or morality) to these LLMs while considering them the central subject of the experiment. Regardless of these evaluations’ results being positive or negative, their core assumption — that LLMs possess anthropomorphic attributes — influences the experiment’s planning through (e.g.) the design of the test set, the interpretation of natural-language outputs, and even its null hypothesis. In turn, this directly impacts the conclusions made.”
In the same interview, he elaborated on the underlying tension: “We either start by thinking that tokens represent language to LLMs the same way they do to us, or that because an LLM outputs a relevant string, it must be understanding the concept/having theory of mind/empathy/etc. This goes both ways: we could also assume that LLMs are blobs of weights just floating about on a GPU, but that would not help explain some skills that they are shown to have.”
Experimenting on LLMs is ethical
His proposed fix isn’t to swing to the opposite extreme, but to set assumptions aside where possible: “I propose that we need to stop assuming that LLMs behave like humans just because they were trained with natural language. Instead, we should perform experiments that allow us to see LLMs as how they are, not how we believe they should be.”
Notably, De Wynter doesn’t rule out some form of machine consciousness outright — he just thinks the binary framing is unhelpful: “We tend to ascribe consciousness as some sort of binary construct (either it is or isn’t!) but I’d argue that there are levels. It’s hard to say that humans aren’t conscious. But, what about a dog? Yes, of course. What about a potato? What about a virus? It’s quite relative and we do tend to go for something human-like when we evaluate/define it, where, in reality, LLMs are things we have never seen before.”
He pointed to recent research on bumblebee problem-solving as an example of how readily we’re surprised by capability in systems we didn’t expect it from.
On what might reduce unwarranted anthropomorphism in practice, he suggested: “I think the #1 thing that one can do to mitigate these perceptions is to have appropriate disclosures and use good alignment techniques where the model is explicit on its nature. I worked a lot on how users perceive LLMs, and they do tend to get attached when it appears to have some sort of warmth/personality. To put it in another way, I don’t get attached to my toaster, but I definitely get attached to characters on a movie screen.”
Marketing spin
He also flagged the commercial dimension plainly, without casting it as good or bad: “The issue here is that these capabilities and claims thereof are very strongly tied to marketing — after all, a lot of these models are products.”
That last point is probably the most useful takeaway for anyone evaluating AI tooling for client work or internal use. Claims about a model’s “personality,” “reasoning,” or emotional states are worth treating as design and marketing choices as much as technical descriptions, and it’s reasonable to evaluate a tool on what it produces rather than on the language used to describe it.