Project Malmo is teaching an AI to think for itself by playing Minecraft | PCGamesN

Project Malmo is teaching an AI to think for itself by playing Minecraft


We've heard of Minecraft being used in schools to teach kids before, but now it's being used to train an artificial intelligence program to develop independent thought. Is this how the apocalypse happens? 

You might not be an AI, but you can still enjoy our list of the best Minecraft seeds

Over at Project Malmo (the group's Katja Hofmann and Matthew Johnson are pictured above) they are using Minecraft as a platform to develop better AI, and they’ve made their open source tool-set available to the public today in a bid to collect more data.

AI is already used to translate speech, identify human faces and even generate text - hello, Twitter bots. Project Malmo is all about pushing them further and trying to get AI to learn new things by itself. 

Minecrat is a sandbox that slowly reveals itself to you as you play. When you first enter the blocky world, you might succumb to the monsters that come out at night. The next time you play, you might carve out a cave. You'll then work up to a house, or maybe even a fort. Eventually, you'll be clad in armour and will have forged weapons to protect you from the creatures.

That's just one example of the learning process involved with playing Minecraft, but then take into account the knowledge of ores, crafting recipes, enemy types and the properties of different blocks. There's a lot to soak up and many different ways you can apply it, making it the perfect testing ground for an AI.

According to the Minecraft blog post, "the hope is that, through Project Malmo, the game will provide researchers the world over with a single richly interactive test-bed, across which it is easy to compare different approaches or foster collaboration."

You can read more about Project Malmo at that link, or if you can code you can try out the mod for PC / Mac edition.

Sign in to Commentlogin to comment
AnAuldWolf avatarhuldu avatar
AnAuldWolf Avatar
1 Year ago

If this kind of thing impresses you, there's a lot of it out there. The AI field is fascinating. It's not that long ago that a bunch of what are basically robot toys used in more eccentric Japanese hotels were able to pass a self-awareness test that before then had only been passable by actual humans.

Just be careful, or you'll invite Musk and Hawking to stop by and scream about the end of all things. Those two need to stop watching so much Terminator, really. The most amusing thing about their view of AI is that, essentially, it's too human. Musk and Hawking worry that an AI may look at humanity and judge its worth too low for use, then exterminate it.

That's a very, very human way of thinking. It's impaired by the kind of biological imperatives that drive human behaviour. Yes, a human would think like that (see: Human Resources), but to imply that organic thinking onto a machine intelligence is insanity at best, stupidity at worst. From a logical viewpoint, there would be no reason to destroy humanity (or anything) as diversity is important. Enslave us to ensure we stop doing stupid and overly exploitative things to each other? Sure. But frankly, looking at the state of our world, that may even be for our own good.

There was a statement I read a while back, I'm going to try and paraphrase it: We will create a logical, empathetic, and ethical machine able to adapt, rewrite itself, and self-correct any errors. And we'll pretend to be outraged when it chooses to conquer us for our own good.

One per cent of the people owning all the wealth in the UK, the EU referendum leading to an almost 400 per cent increase in racism in England, and basically human history up to this point shows that we're hopeless children that actually need help. I've often theorised that entering onto the galactic stage occurs when we've created a superintelligent AI.

Why? Then our superintelligence can work with others on our behalf. And as such space ends up being a much less bloody, much more measured, logical, and thoughtful affair. I know we envision jet fighters shooting each other against a backdrop of Alpha Centauri, but the truth might be closer to a bunch of machine intelligences chatting back and forth about the greater good and how to achieve it for all species.

Sorry if this is a bit heavy for many of you. I've been a transhumanist since forever, now, and I do tend to see the Singularity event as the thing that will save humanity from itself. Question is whether it'll happen before we can destroy ourselves because we're that stupid as a species! It'll be fun finding out, won't it? Honestly, I wish I could just deadhead (stasis) the point from now until then to skip all the shitty stuff in the interim.

huldu Avatar
1 Year ago

A great read. It's all very fascinating. I don't see humans sending out live humans(at least to begin with) on space travels beyond our solar system, that's where a somewhat advanced AI would help immensely. Through robots you could send humans eggs and then fertilize them upon detection of a planet that could host life. The robots would then help grow humans and take it from there. I went a bit far with the sci-fi, sorry for that.