Elite Dangerous introduce AI, AI players exploit bug allowing them to maker super weapons and go Skynet, wiping out actual players.
A combination of an upgrade to the AI system and the addition of Engineering or Crafting, allowed the AI NPC’s to generate weapons that the designers did not expect, in effect super weapons.
Is this a warning for the future, a simple bug allows AI to go rogue hmm???
Googles DeepMind AI or IBM’s Watson + 3D Printing?!
You just better hope they don’t go after those warning us about the potential threat. You might be among the first victims.
Seriously though that’s just awesome and I find myself in agreement with one of the commenters who suggested that emergent behavior like that from AIs can sometimes make the game more enjoyable. At least for a short while.
Not sure if I will be able to find it but there was a game that used genetic algorithms (generational play) and neural networks, I think it was called Creatures. It had positive and negative reinforcement and was great fun (like an advanced tamagotchi).
But apparently the company that developed this technology managed to get a contract to work with the RAF, the RAF probably thought it would be a fun way to improve their combat flight simulators. Until the evolving and adapting pilot ‘critters’ started doing immelman turns and picking up advanced dog fighting techniques and beating the pilots!
Check out the link this was in the 1990’s. Think about it fly a plane with a computer, no G limits, massive weight reduction (life support/ejection/HUD/Controls) and much faster processing/response time.
Also have you heard of the memristor, the fourth electrical component that has never quite arrived even though things started hotting up about around 2008 (wikipedia).
In theory a memristor works in a very similar way to the neurons in our brains, remembering what it stores and responding to inputs. It could provide a massive breakthrough in computing, as it would massively drop the energy needs of mobile devices and boost startup times to near instant.
And it could also pave the way for adaptive computing, where the very chips can be adapted on the fly to do different tasks. As well as boosting AI research and development.
Maybe some years ago the AI pilots became more and more intelligent and eventually took over the world.
They then used these memristor things to build super computers and hooked mankind into a massive simulation.
Welcome to the matrix …
I think memristor (and memcomputing) is hold back by a lack of theoretical understanding for generalized application (ie how you programmed it), we just the theoretical framework that say it should work (ie prove that it is a universal computing paradigm like turing machine). it work theoretically like the brain but not practically (at least not yet).
the memristor seems to be held back (from what I understand) by lack of implementation details of the problem mention above (else it’s basically a Dram slighty modified).
There is also a problem of precision due to natural electric interference as it is an analogue system that operate on frequency and amplitude, noise kills it!
Yeah, of course, planes are better at flying themselves, but should we really build an impossibly good fighter pilot who you effectively can’t kill because he lives in code, who is programmed spesifically to kill people? That seems like some scary territory that could easily go wrong
As far as this Memristor thing, that article is way above my paygrade, I don’t really understand the spesifics, but your explaination sounds pretty interesting. I just feel like programming for something like that would require a pretty huge paradigm shift, as, from my limited understanding, it seems like memory management would have to change completely. Computer architechture in general would probably change. Perhaps this is why progress on it has slowed?
Also, the name Memristor sounds like something from Tron or something, ha ha
Memristor is also above my level, but I think part of it is because it’s because all the metaphor are academic’s. Wait till they vulgarize it and organize the theory into best practice, I’m sure turing machine as explain as soon it was invented would have left the layman scratching their head to oblivion.
But yes it’s an entirely different paradigm, as far all “code” I have seen weren’t “code” but physical cabling to encode a problem. The way I turn in a metaphor is that it seems to work like a complex version of multiple connected flasks, where the water try to minimize it’s energy level as given by the problem (ie find the optimal configuration) now how do you jump from that (if true) to encoding any problem, I don’t know
Good point even a simple but accurate AI player could be a nightmare in a computer game, imagine space invaders or storm troopers that could shoot straight.
Definitely. Considering its fully physics based its not toooo hard to avoid projectiles, as the aiming is based on speed and vector. Change either and it will likely miss. Aiming for the player is automatic as well though, so it makes it fair in that respect
There’s a lot of those, not necessarily games, but AI projects (sometimes in game shape) where the AI evolves through generations until it does what it was meant to do perfectly. And yeah that’s one of them.
.
