AI agent swarms in Minecraft fail to build civilization, drown in bureaucracy

Posted by entryleveljobsc 19 hours ago

Comments

Comment by openasocket 17 hours ago

I find this research odd because I don’t know what field of study to call this. It’s not computer science. It’s almost like behavioral psychology, but for AI agents instead of humans or animals. That said, I feel like these studies lack a lot of rigor. Or maybe it’s a lack of clear direction. Like, what are we trying to determine here? Because these things often involve advanced AI agents and complex harnesses, but then it sounds like they are given a very minimal prompt. What are we really studying then? Would the outcome be completely different with a different prompt? Can we generalize these results to other models? Different generations of models, different levels of thinking? What if we tried fine-tuning the model to this scenario? Is this fundamentally how any LLM would act, or is this some artifact of our current, generalized fine-tuning?

Comment by montagg 14 hours ago

Reads to me like Proto-psycho-history. Attempting to model behavior mathematically, with the assumption that all human behavior can be modeled this way, and that these models can be useful for predicting human behavior.

Saw an article a quarter century ago in The Atlantic about simple population models as a way to explain more complex behavior. One was: randomly distribute blue and red cells across a 2D grid. Every frame, pick a random cell: if it has at least one color like it nearby, leave it. If not, move it to a random location. Over many generations, segregated color formed. The article used this to describe something about segregated communities forming. I offer no opinion about the credibility of the article, but it was fascinating to me as a baby nerd, and seems at least a little related to this article’s line of thinking.

Comment by tkel 8 hours ago

Once you understand how complex the human brain is, with 86 billion neurons and 100 trillion synapses, formed over billions of years of evolution, comparing that to the machines we have built makes them seem silly and extremely simple in comparison.

Psycho-history is fiction. You're not going to be able to model a single human much less millions of humans and their environments.

Comment by jurgenburgen 30 minutes ago

> Psycho-history is fiction. You're not going to be able to model a single human much less millions of humans and their environments.

Also it didn’t even work in the original fiction. If it did, there would be no story to tell.

Comment by sznio 3 hours ago

I often wonder how much of the human brain is actually dealing with "thinking", by that I mean navigating, understanding complex systems, and what is there to keep you alive, walking and breathing.

The machine doesn't need this extra cruft.

Comment by rogerhey 6 hours ago

Why do you think you need to do brain simulation to proxy human behavior?

Comment by TJSomething 14 hours ago

I think this is mostly just software benchmarking. The most science that can probably be extracted from this case study is probably a couple data points on where we are and where we can improve with AI agents. It seems like a decent task to optimize for. It might also work as a proof of concept for an economics study or theoretical political science.

Comment by acbart 17 hours ago

Asimov predicted "Robopsychology".

Comment by freakynit 16 hours ago

I think large LLMs can be seen as a representation of humanity's collective thought processes... almost as if humanity were a single entity.

They have read pretty much everything we have ever written, good and bad, so they're kind of a mirror of us, as a species. Not a perfect one, but still.

Aliens could probably use a llm to get a crash course on what humans are actually like, and how to interact with them.

Comment by pennomi 14 hours ago

They are the collection of all human thought processes, as curated by Definitely Not Evil Corporation.

Comment by andsoitis 16 hours ago

> I find this research odd because I don’t know what field of study to call this.

Probably multiple with very complex system dynamics, feedback loops: Cliodynamics, Cultural evolution, Cultural anthropology, Archaeology, Sociology Political science, Institutional economics, Urban planning, Complexity science, Social ecology

Comment by bor3dnrd 16 hours ago

Computers science is just applied math, has nothing to do with the machine.

The term computer was coined to describe humans engaged in applied math jobs before the machines existed.

The models are matrices of values engaged in reducing options to the most desired outcome given inputs. That's software dev.

"Fundamentally" or "fine tune" is meaningless; no model spontaneously emerged through a membrane from another reality. They’re all fine tuned from nothing into a model or from a general model to a specialized one.

You're circumlocuting semantics to say something for social media, but the physical provenance is clear. They're not of mythical origin and it's still math relative to set of memory values.

Comment by ModernMech 15 hours ago

Computational sociology? I imagine a model of society could come out of this, then you could create hypothetical societies and see how they behave.

Comment by lubujackson 10 hours ago

Sociology is the right concept. It sits somewhere between system analysis and emergent human "sort of" sciences, like economics and sociology, where we can detect structures but humans are too complex to surface any platonic concepts with much confidence.

AI interaction may give us a cleaner way to test and analyze emergent properties that sit beneath so of the softer sciences, giving us a harder science that we had no way to investigate before.

Comment by joshspankit 14 hours ago

I suggest thinking of it as psychology on a mathematical model of collective human works, and I’m currently curious to see the differences of models trained only on specific cultures

Comment by none_to_remain 7 hours ago

Elementary Wizardry: Summoning and Management of Elementals and Golems

Comment by karhuton 16 hours ago

Well, obviously Psychology + (computer) science = Scientology

Comment by threethirtytwo 16 hours ago

It’s the same way psychology works on humans. Black box observational stuff.

You can’t generalize anything here because the psychology of the model can be affected by system prompt, reinforcement training, and training data. What you observe here can be wildly different if you change any of the underlying things I mentioned.

It’s like watching an ant colony.

Comment by bradishungry 16 hours ago

it’s just machine learning but instead of doing things the old fashioned way everyone just uses llms now. this is nothing new

Comment by dr_dshiv 9 hours ago

Machine psychology

Comment by classified 16 hours ago

Never trust an experiment you haven't faked yourself.

Comment by adammarples 16 hours ago

I'm not sure it's that serious, but they did have two models, Claude and gpt were on opposing teams.

Comment by smrtinsert 15 hours ago

The label might come later, but cooperative or adverserial behavioral patterns matter deeply when deploying agents to act as weapons of defense or business builders.

Comment by RAWGOD 13 hours ago

>I find this research odd because I don’t know what field of study to call this.

AI bullshit hype generation.

Comment by iLoveOncall 14 hours ago

Call it what it is: slop.

It has zero intrinsic value beyond the headline. The whole goal is to market LLMs if the result is successful.

Comment by aaquibahm 14 hours ago

you can’t see any benefits to studying how LLMs behave in long horizon multi agent tasks?

Comment by vchernets 11 hours ago

[flagged]

Comment by abdevestudio 9 hours ago

[flagged]

Comment by _vertigo 18 hours ago

I started playing the rs-sdk (https://github.com/MaxBittker/rs-sdk) with Claude after seeing it on HN the other day. I haven't looked at the Minecraft experiment but just letting an agent loose in the game and telling it to play did not work well. At first it really did not have a good feel for how the game actually plays and would routinely waste tons of time doing something that makes no sense — very reminiscent of the "four agents spent 11 minutes appointing a ‘sole fertilizer operator’ for a single piece of bone meal on a single wheat block" anecdote in this article.

But, after several hours of correcting and directing Claude (build modular, re-usable snippets of useful code rather than one-off scripts for driving, document your learnings for future agents, etc., explicit guidance on playing more optimally) it can now play the game fully on its own. I just gave it a goal to play through the end of the "Legends quest" (nearing end game content, requires playing through lots of complex quests ) and it's breezing through it.

I haven't experimented with swarms of agents yet, but I know if I had tried to create a swarm in 2004scape before the agent was actually good at playing the game, it would have failed horribly. So, I wonder how good the agents they were experimenting with are at Minecraft when playing alone.

Comment by DougN7 16 hours ago

How do you get Claude to continually work at it? I find it will do a number of steps and then just stop, no matter what I tell it. Then I prompt “continue on” and it will work a few more steps on its own.

Comment by _vertigo 16 hours ago

/goal has worked for me

Comment by kxrm 13 hours ago

You can also tell Claude to setup scheduled tasks.

This is very helpful when you want Claude to do something specific, like analyze the next phase. Review past performance etc. I have found this to be a little better than Goal. However, cron tasks only continue for 7 days.

Comment by AustinDev 14 hours ago

Yeah use /goal or use another agent to prompt CC it's kind of silly but it also works.

Comment by nurettin 16 hours ago

I just tell it to work until described result. It works unless it gets stuck at a necessary step I had explicitly forbade earlier, or an unreachable port.

Comment by dfedbeef 15 hours ago

This is an incredible waste of money, power, and compute time.

Comment by wilkystyle 14 hours ago

Good point. The power and compute time would be put to much better use in one of Sam Altman's datacenters.

Comment by ShinyLeftPad 13 hours ago

or if everyone realizes the same datacenters go bust.

Comment by AustinDev 14 hours ago

I got Chat GPT 5.5 to land on the Mun in Kerbal Space program it was a fun and interesting learning experience.

Comment by _vertigo 15 hours ago

I considered that but I also considered that I don’t come close to maxing a normal $20 a month Claude pro plan. So I think “incredible waste” is probably too far.

I also see total garbage slop upvoted on this site every day. At least with this, I am experimenting and learning a little about how to use agents effectively, I’m not sure I can say the same happened for other slop on this site.

Comment by Tadpole9181 14 hours ago

Do you also bully people for playing video games or taking a leisurely drive?

And how very "hacker news" of you to discourage someone from playing with SOTA technology in unique ways.

Comment by customguy 8 hours ago

People in Google playpens, "state of the art technoloy", 99 times out of 100 it's Minecraft or Pokemon.

Wouldn't a true 1337 hax0r apply some critical analytic thought to why that is?

Comment by micromacrofoot 14 hours ago

so like every video game ever created? at what point do people decide it matters? just because it aligns with personal feelings about ai use?

Comment by lifeisstillgood 17 hours ago

>>> build modular, re-usable snippets of useful code …

That’s fairly close to a definition of civilisation. :-)

Comment by bhouston 18 hours ago

I think that is a common failure case. It isn't that different that agents getting stuck in other types of loops, distractions. There is this great short novel about it by Adrian Tchaikovsky, "Service Model":

https://en.wikipedia.org/wiki/Service_Model

Comment by gjskngnf 15 hours ago

I’m not sure there’s too much to take away from this. It seems like a fun experiment, but I find it unsurprising LLMs failed in this context. They used this “mineflayer” wrapper, allowing GPT to read some game state and take actions. Yet the whole time, the game is running and the agents have to sit and process every minute or so. We know that agents cooperate much better in a harness more like bash in a Unix VM attempting some programming task. To conclude from this that agents categorically poorly cooperate and descend into bureaucracy is an overreach. I think anyone working on AI robotics would find this result pretty unsurprising

Comment by ddtaylor 18 hours ago

Direct article: https://labs.ycombinator.com/projects/games/index.html

Comment by czep 8 hours ago

I find this experiment inspiring, but probably not for the reasons intended by the experimenters. To let loose a swarm of agents in a sandbox game with the vague goal of "build a thriving civilization", and then to observe them flounder in risk averse strategies is precisely what separates us from them. There is no risk-taking, no irrational spark, no neurodivergent "because it's there" impulse that led us out of the caves in the first place. Perhaps it's impossible for a fully rational intelligence to build a civilization, as analysis-paralysis would exponentially dominate all decision-making. It's only happened because of happy accidents, ultimately stemming from what would in the moment be thought of as very bad ideas.

I am equally intrigued by how to augment Minecraft game play using agents as, essentially, team members to command. Given the ability to communicate with agents who are autonomous enough to play the game well enough, could they assist a human player to achieve very ambitious goals? Spawn in with 50 "friends", and instead of letting them wander aimlessly, give them some concrete tasks that a human player knows to be valuable steps on the tech tree. Build a cobblestone generator, an iron farm, lock in a Mending librarian, find a stronghold, loot the End, a sugar cane farm, gunpowder for rockets, now everyone has Elytra, on to raid farms, a perimeter, slime farm, witch farm, etc.

We don't give agents a rack of servers and an objective "build the next unicorn", but we direct them to solve tasks we as humans know are valuable in our own domains. Same story here and I am seriously daydreaming about a server in which I've got 50 full Netherite agents at my command ready to assist me in whatever grand build I have in mind. "Ada, I need 20 shulkers of black concrete. Bram, bring me 14 stacks of blaze rods. Cleo, we need more blue dye, find the optimal gradient in the flower forest and build a shifting floor farm for cornflowers..."

Comment by Footprint0521 8 hours ago

Or you could use the server console commands, for a time compressed equivalent effect

Comment by GuestFAUniverse 18 hours ago

Germany 2027 and onwards.

Comment by rvz 16 hours ago

That and more accurately, live footage of EU bureaucracy.

Comment by radial_symmetry 18 hours ago

HAHA this is the only comment right now and exactly what I came in here to say

Comment by sega_sai 17 hours ago

https://news.ycombinator.com/item?id=50296486 it was discussed here

Comment by za_creature 15 hours ago

Of course they failed <today>, but make it a supervised goal and tomorrow we'll have minecraft utopia*

Now we just need to end human civilization a few times before the reward function kicks in and we'll have actual utopia

* the utopia might be backed by anti-thermodynamic processes available to a simulated reality; if you are concerned about anti-thermodynamic processes not being available in your reality, consult your local physicist or snake oil salesperson for a second opinion

Comment by bad_haircut72 18 hours ago

This site is cancerous, overloaded with ads

Comment by horsawlarway 17 hours ago

I don't normally comment much on site format vs content, but it was genuinely hard to read this with the ads.

Comment by gloosx 1 hour ago

tom hardware website genuinely looks like some kind of satire or meta-irony about ads. When "What’s the max you’re willing to pay for a new Keyboard?" insert and " Get club Access Quick" popup kick in at the same time there is almost no content left on the screen, just ads. I don't believe this is not some kind of a sick joke!

Comment by asdff 14 hours ago

Why aren't you just blocking ads?

Comment by mejthemage 7 hours ago

this site is also aggressively anti ad blocker.

Comment by lhk931122 7 hours ago

In the simulation games I play, you basically control multiple characters in single player, and if each character were given its own AI like this, I think you could see some pretty interesting gameplay. This used to be hard even with frontier models, but looking at this, it seems to work well even with models a tier below.

Comment by tom910 6 hours ago

This is exactly my experience in using multiple orchestrators and workers which developing and improving a platform. AI have tendency to create a lot rules and strictness which lead to a lot problems with following

Comment by fallinditch 18 hours ago

Interesting insights about similar issues in this video by Openrig [1]

Multi-agent coordination is an engineering problem characterized by the inevitable emergence of human-like social pathologies - such as bureaucratic inertia.

[1] https://m.youtube.com/watch?v=AL-PQuB2wy0

Comment by perching_aix 18 hours ago

> human-like social pathologies - such as bureaucratic inertia

Isn't bureaucracy just the sociological manifestation of synchronization overhead? Sounds shortsighted to file it away as just some weird social pathology that emerges out of nowhere specifically affecting humans.

Comment by TheOtherHobbes 15 hours ago

No, bureaucracy also includes elements of social dominance by people who have no other means to assert status (or sometimes outright covert sadism), busywork, and processes that are rule-based - often dumbed down in the hope that people with limited intelligence don't make avoidable mistakes - rather than responsive and reality-based.

Getting agents to deal with their own versions of these issues is an interesting experiment.

Comment by perching_aix 14 hours ago

I personally find that to be a bit of an inversion. People will utilize leverage and identity in whatever context, so mixing that in doesn't seem cogent to me.

But maybe the two are genuinely tied together enough that it's deserving to be a part of the definition, I don't know. Tried looking up the definition, I'm not really convinced.

For the longest time I considered the "hate on bureaucracy" more of a reactionary trope rather than anything actually coherent. Maybe this is just another one of my cultural differences of many. It's not as if people forming consensus, keeping count, or delegating authority were ever actually possible (or desirable) to escape, comes with scale. So to try and paint it as some odd devilish plague always came across to me as dilettante, though maybe it's more me who's simply never grasped the whole dynamic and the cultural position around this fully.

Comment by rightbyte 14 hours ago

> For the longest time I considered the "hate on bureaucracy" more of a reactionary trope rather than anything actually coherent.

That is my thought too. They want like people that "just do it" and skip the process for firm action. Because they are smart.

It causes havoc on macro level and on average. Have your cake and eat it. When too many override social norms there is no profit in doing so etc.

Comment by farhadhf 10 hours ago

You can kinda see this if you try to use something like paperclip with more than a couple of agents that need to work and coordinate with each other. It quickly becomes a bureaucratic mess that gets every agent (and you) confused and just burns tokens without creating any useful output.

Comment by serioussecurity 17 hours ago

https://arxiv.org/abs/2610.22319

Maybe a skill issue? Deepmind beat Minecraft. Or are the goals different?

Comment by Myrmornis 8 hours ago

I wonder how a model would fare that somehow had been trained / weight adjusted to have knowledge of human institutional processes almost removed.

Comment by ChrisMarshallNY 14 hours ago

I guess Minecraft is suffering from administratium[0] poisoning.

[0] https://web.archive.org/web/20130315075433/http://www.lhup.e...

Comment by jordanb 17 hours ago

We've automated management consulting.

Comment by xlayn 16 hours ago

This is game theory... wait a sec... is degenerating capitalism into something that burns everything down to ashes just as a "the ultimate 6d chess move" to actually solve/perfect all the computer problems?

We are in AI thing right now... but bureaucracy makes everything fail... so we can start doing genetic algorithms/game theory to optimize it so we can swarm... but swarming requires defining a max/goal, which will require some kind of formal definition framework for what "the sitius altius and fortius" of AI agent swarm can build on Minecraft can do....

Wait a sec... are we just trying to beat Magnasanti in Minecraft?

Comment by charlieyu1 10 hours ago

Just like my experience with ChatGPT then. Overindulgence with details so I have to tell them to stop to get things done

Comment by ajrouvoet 17 hours ago

Wow, Tomshardware is spawning a ton of ads. Impossible to read this.

Comment by 4ndrewl 16 hours ago

Midway through the page I was randomly asked how much I'd be prepared to spend on a new keyboard. It was like a cognitive wrecking ball when trying to concentrate on the article's content.

Comment by 18 hours ago

Comment by ccll 5 hours ago

Really? Does gpt-6.1-sol work that fast?

I'm wondering if System 1 models like Jev/Laya can work together with GPT to provide some "instincts", that would be fun.

Comment by threethirtytwo 16 hours ago

These are really subjective tests because the LLMs can have adjusted behavior based on the system prompt.

If you have the LLM operate on default behavior it likely won’t work because most systems if picked at random don’t work… even for humans.

The systems you see around us come from thousands of years of cultural evolution. If you simulate the same thing with LLMs where you have it write to its own context on what works and what doesn’t work you will eventually get something working.

Comment by ianmarcinkowski 7 hours ago

See also: people who try to add a local worker's co-op or union to their local coffee shop!

Comment by ChrisArchitect 10 hours ago

[dupe] https://news.ycombinator.com/item?id=50296486

Comment by hsuduebc2 15 hours ago

I would like to see what would be different with uncensored models.

Comment by TacticalCoder 15 hours ago

Oh like the country of France?

Comment by deadbabe 16 hours ago

They should have elected an all powerful leader, whose word is absolute. Bureaucracy is designed as mitigation for individual human incompetence. You do not need it once you are an all powerful, all knowing being.

Comment by delichon 17 hours ago

Humans arrived at a capacity for social cooperation via millions of generations in which it has been a force on reproductive fitness. That created deeply rooted phenotypes like empathy, loyalty, hierarchy, and morality. Even if such concepts can be well described in a statistical model of language, there is no reason to think that they automatically become embedded in motivation. Without that, they may as well be anthropological observations by Martians.

To embed them as motivations, we can embed neural nets within a framework of genetic algorithms, in which successful cooperation is a gate to the next generation. It seems to me that this approach should be not just about creating efficient agent swarms, but should also be at the core of alignment efforts.