Today in the news we’re talking about things AI agents have done that are relatively harmless and/or at least don’t involve hacking the government-funded healthcare system in my home country: beat Minecraft! And look, I’ll be honest: as someone who didn’t allow Minecraft in my life, I can’t say I was entirely aware that “beating” him was something you could do. Isn’t it a voxelized sandbox where you can do cool things like build hard drives and giant temples and whatnot?
Apparently not! Or, at least, that’s not all, because according to a paper submitted to the arXiv preprint server last week, “four Gemini 4 Argon agents in a Minecraft sandbox survival world” managed to achieve their assigned objective: “defeat the Ender Dragon.” It took the agents a little over a day of cooperative play and cost about $520 in tokens.
Perhaps the most interesting detail that emerges from just a glance at the document’s summary is that the agents “wrote a total of 2,783 JavaScript programs.” If you are wondering why agents would write JavaScript programs, you will find this article to be an interesting read. The answer to the JavaScript question lies in the statement that “to our knowledge, this is the first end-to-end autonomous completion of Minecraft by AI agents, achieved using only the standard Mineflayer library, without pre-searched seeds or other game-specific tools written by humans.” »
So, what is the Mineflayer library? The answer can be found on the project’s GitHub page: it’s a “powerful, stable, high-level JavaScript API (for creating Minecraft bots).” In other words, the agents did not directly control the game’s characters; instead, they used Mineflayer to script the actions. As the paper explains, “Rather than seeing pixels or pressing keys, an agent acts by writing and executing a short JavaScript program, such as walking to these coordinates and mining every iron ore within a 16-block radius, and reading the text that comes back.”
To a human, this probably seems like an unnecessarily complicated and somewhat retrospective way of approaching the task at hand. But it says something interesting about what really happened here. As the document itself notes, these are general-purpose coding agents. What they are good at is writing code. As such, it’s ultimately less about (dramatic music) killing the Ender Dragon and more about the agents’ ability to take on a large and complicated coding task.
This required them to divide tasks in a way that made the most efficient use of their time, communicate their successes and failures, and so on. The paper notes that these kinds of capabilities are exactly what is required to tackle large-scale software engineering tasks: “Complex real-world software engineering and business tasks require the same long-term coding and multi-agent coordination. »
Regardless, reading this article is the kind of thing that makes the direction the AI industry is heading, with its brazen, unpunished intellectual property thefts and relentless attempts to integrate AI into any software imaginable, so endlessly frustrating. If we simply consider the article itself, it documents some fascinating phenomena. Take, for example, the section titled Emergent Behavior. “Nothing in the agents’ instructions mentioned roles… beyond the line that players decide how to share work,” the researchers write. “During the 229 messages exchanged before the victory, the team converged on practices familiar to effective human teams.” It’s really interesting!
The same goes for the fact that while the agents were ultimately successful, they often did royal stupid things with things that humans would consider simple: while playing on Easy mode, they still managed to die a total of 605 times, “mostly to monsters.” The paper also notes that “22% of the programs (written by the agents) ended in an error” and that it took them “nearly 14 hours” to figure out how to defeat the final boss, which…well, it’s not Malenia, to put it that way.
But, of course, this cannot be read without the more general context of this type of task being the prelude to ever greater automation of tasks formerly reserved for humans. Of course, it would be wonderful… if the benefits of this automation were shared equally, instead of falling exclusively to the same idiots who spend all their time telling us that this process is inevitable and that we should all suck it up and learn to code – oh. Wait. Maybe not that, heh.
Gn tech
