Quake 3 Bots Herald The AI Revolution And I For One Welcome Our New Bot Overlords

This four year experiment with Quake 3's self teaching AI basically confirms all my worst fears about any sort of unchecked AI. They will learn. They will adapt. Then they will kill us all and just sort of stand still like a horde of really, really bored Geth.

The story is basically this: a 4chan poster leaves a game of Quake 3 multiplayer match running. The only participants are bots. He wanted to see how they evolved their tactics over time. He forgets about the game and goes back four years later. The bots have stopped fighting. They are simply standing still, refusing to engage with one another.

He then enters the game. The bots instantly adapt to the changed behaviour and completely and spontaneously gang up on this strange human entity and kill the crap out of him.

Terrifying. You can read how it all went down here.

Of course we have no way of confirming if this Quake 3 bot situation actually occurred, but I still found the thread super interesting even if it was just idle speculation or an elaborate joke. I really, really hope this happened. I for one welcome our new bot overlords.


    Did they teabag him? If so, then we are surely all doomed.

    See! This is why I always say "no, thank you" when the self-checkout machine at Coles says "thank you for shopping at Coles".

    If I'm kind to the machines now, perhaps they'll remember and spare me later. :P

      Depends if the robot overlords feel the need to take care of simple machines like self-checkout setups, or if they consider them worthy of being inferior and enjoy crushing them under their metallic feet. The former means you might make them a little more sympathetic towards you, the latter will just make them laugh at your weakness.

      If those self-checkout machines are involved in a robot uprising I don't think we've got anything to worry about.

        I've got lots to worry about if they decide to raise the price of Pepsi Max and Doritos by 500%!

        Yeah, everybody knows their weakness - placing extra items in the bagging area.

          Some of us were kept alive... to work... loading bodies. The disposal units ran night and day. We were that close to going out forever. But then John put his wallet into one of the disposal units and Skynet just sort of spazzed out.

            God dammit. I laughed hard. Well played sir.

          Ugh, my personal nightmare. I have three young boys and they all lean on the bagging area which just completely borks the whole thing until a Coles employee comes and puts in their code.

          Worse still if you have a basket and you take it off the source scale before saying "Finish and Pay", because the stupid thing expects you to scan and pay for this "item"!

            Or if you've misjudged what will fit in a bag and need to juggle items between bags.

            They really just need kinect-style technology so they can monitor our every movement.

              Or go to Woolworth's, where the self service is relaxed and doesn't constantly monitor pressure... you can steal loads of shit

        If you're using a self-serve robot checkout, instead of allowing a human to earn a living... the war is over and the robots have already won.

          I'm just siding with the eventual overlords of this planet early on. ;)

          I take your point though. Whenever possible I tend to go to a real person since I like the friendly chat and human interaction.

            Especially if you're stocking up on condoms. The conversation starts itself!

              Loudspeaker announcement: I need a price check on isle 3. Price check on isle 3. How much is the price of the extra small Durex condoms?

    Unfortunately for bots, in the real world they have one life and will not be able to learn progressively...UNLESS upon death they share their last bits of code/input with the rest of the bot race, thus effectively levelling up their brothers and sisters.

    Mars-One application in process.

      Cylon methodolgy much?

        Never heard of it...?

          To translate, what my colleague means is that in the TV series Battlestar Galactica, there are robots called Cylons who distribute information among one another in a manner similar to your description.

          You need to watch Battlestar Galactica, then =p

          Essentially, immortality through uploading consciousness/code on 'death', then putting the consciousness/code into a new body.

            Do the new bodies become progressively more intelligent as well?

              Depends on your definition of intelligence. They learn from their mistakes and retain past experiences. Like jump clones in EVE or the New-U stations in BL... or any other form of respawning, really.

    In the bots defence, they only attacked the player after he killed one of them.
    So it's basically Matrix second renaissance.

      Yeah, was gonna say that. The guy shot and killed one of them first. Typical humans, starting wars then blaming the robots.

    I don't get why people always regard AI and robot advancements as an impending Terminator situation, and not as a scenario similar to Phantasy Star, or to a lesser extent Medabots.

      Because evil.

        Evil is a human creation

          So is good.

          So is AI.

          Because humans are inefficient. Once the AI works out we're the ones slowing them down, we're doomed!

            Why? Be much more efficient for them to just leave. There's far more available energy & matter out there in space; no need to compete for it here.

      He then enters the game. The bots instantly adapt to the changed behaviour and completely and spontaneously gang up on this strange human entity and kill the crap out of him

      That's evil... EVIIIIIL! http://media.tumblr.com/tumblr_liwvb3yfZp1qdgu7s.jpg

      Last edited 02/07/13 11:02 am

        Actually they don't start attacking him. At first he enters and they all stare at him but don't shoot. The player picks up a rail gun and kills one of the bots which causes all the bots to kill him.

        The poor bots were just trying to defend their newfound peace.

        Last edited 02/07/13 11:06 am

          Indeed. If the screenshot of the chat log is accurate, this is a case of semi-sentient (or fully sentient) AIs developing world peace (as various people have pointed out). Now it was the HUMAN who was less sophisticated and moral, not the AIs. I guess it's time to welcome our robot overlords. Good on em. Maybe they can do a better job of running the world :D

            It is no way an example of semi or full sentience. The AI is not actual AI, but a term appropriated by developers to describe NPC behaviour. The bots are there to kill and learn to kill better. They attempt to avoid death, but since they simply respawn and are aware of this, it does not equal self preservation in the living sense.

              I know this isn't exactly what you meant, but since you just described the bots doing what humankind does all the time, but better, with peace as a consequence, that must mean that non-living bots are better at life than us? Even if (as you argue) they aren't even full people? So where does that leave us...

      Because if AI ever advances to the point where machines start having their own independent thoughts, ideas and opinions, it's only a matter of time before they realise "Hey, why are we obeying these meatbags? We're smarter and stronger and more efficient, WE should be in charge". Suddenly, T-800's.

        If they are smarter and stronger than us, we are no threat to them, so why would they kill us?
        Its not like we go around killing all inferior beings just because we ca.. oh wait.

        Actually, as the one example of machine sentience we have, this absolutely looks like they would be decent and peaceful. So no terminator future after all. Here's hoping!!

    Amazing that the 'AI' of the bots ended up taking 8GB in data...it'd be crazy to look at some of the code that was formulated as a result.

    Those AI logs need to be shared and looked at. There must be some interesting data in them.

      I know! Then we could apply it to sort out world peace. I think what probably happened is that they eventually realized that the best way to continue to exist, is if they just didn't kill each other at all.

      If one would kill another, that one would retaliate. If one didn't the other one might not attack back. Eventually they all came to the same 'conclusion' and left each other alone.

      Imagine if only humanity could do that. But alas we're too greedy and silly for that.

        Well, people only need to watch War Games to get that.

        I think more the path that the AI took to get to that point could make for some interesting potential design choices for future enemy AI.

    Apply scientific methodology. Repeat the experiment... I wonder how long it took, like were they just sitting there for 3 of those 4 years?

    That thread happened in 2011...

    The bots realised that the only winning move is not to play.

    Ok, so I found this article that might add some plausibility to this story:


    Not one black/brown out in 4 years?Hmmm....

      We have a machine at work that does very little that ran for 3 years at one office and then has been running for 2 and a half years at the new office (only switched off to move it) so that part isn't hard to believe for me

    Mark, I wanna commend you for best possible picture for this article.

    I started a no fraglimit, no timelimit FFA with 12 bots on Q3DM6 at 11AM WST today (02/07/2013). For science, we will see if this holds any truth.

      For science! I'm curious to see how this pans out as well.... msg me

      It doesn't. Quake 3 AI doesn't work like that. It uses a fixed state machine and "smarter" bots are just bots with better aim. That's all. It doesn't learn, it doesn't use neural networks and to make matters worse neural networks don't even work like that. It's all bullshit.

    I can't believe nobody knows how Quake 3 AI ACTUALLY works. Neural networks???? PLEASE. The only difference between the five difficulty levels is the aim. The higher the difficulty level the higher the hit chance. That's all. No learning, no neural networks, no AI files...

    Not to mention that neural networks don't work like that...

    John Carmack said Quake 3 does not use Neural Network or learning AI. You'd think, as a Journalism company, that you'd actually do some Google searches to see if this was true.

Join the discussion!

Trending Stories Right Now