Why Age of Empires II is Giving LLMs a Reality Check

Ok wait because this is actually insane. We all know large language models (LLMs) are the talk of the town, with claims flying around about them having human-like traits like morality or language understanding. But what if those claims are a bit.. off?

The Unhinged Experiment

Some researchers decided to take a detour from the hype and ask: Do these LLMs really have these 'anthropomorphic' qualities? Instead of getting swept up in the buzz, they did something wild. They built a simple neural network on the classic video game Age of Empires II. Why? To prove a point that any system, whether built with LEGOs or running the Greater Boston Area, can seem to have these human-like traits. No cap.

LLMs aren't That Unique

Here's the tea, besties. The team argues that the supposed human-like attributes of LLMs aren't as unique as some folks claim. Sure, responses to prompts might be consistent, but how we interpret those behaviors could change depending on the 'substrate' or platform. In simpler terms, it's not just about the machine's brain but about where and how it's used. So, when everyone screams 'AI is human!' maybe take that with a grain of salt.

A Null Assumption

Let's talk about the 'null' assumption. Instead of jumping on the bandwagon and assuming LLMs are little humans in disguise, what if we start by assuming they aren't unique in having these traits? This approach could flip the script and make our AI experiments more grounded and informative. Not me explaining AI research at brunch again!

The researchers even tossed in a curveball, proving that Age of Empires II isn't only fun but functionally- and Turing-complete. Who knew playing a retro game could be so intellectually spicy?

So, Why Should You Care?

Why does this matter beyond the academia bubble? Well, for starters, it makes us question what we're really expecting from AI. Are we just projecting human traits onto these models because it's exciting and makes headlines? Or are we genuinely understanding what they can and can't do? The way this experiment just ate. Iconic.

No but seriously. Next time someone tries to convince you an AI has feelings or a moral compass, ask them if they've played Age of Empires II lately. Because if a 90s video game can mimic these so-called human traits, maybe we need to rethink what we're ascribing to AI.