EA tests the use of virtual players who in Battlefield 1 teach themselves the basic skills virtually from scratch by using artificial intelligence. ... EA tests self-learning AI agents in Battlefield 1
Next step they will teach them how to go to work and earn money. And then Micro transaction the HELL out of them.
I wonder if they are using this to train a "tactically" smart AI to implement into a future Battlefield game, as you say, previous games did have bots, but they were kind of stupid...
This is an interesting project, but a bit hypocritical and wasteful: * One of the reasons EA does MTs is because, according to them, making games is too expensive. And yet here they are, spending all this time and money constructing an AI that learns how to play a game. How exactly is them doing this ever going to yield a profit? * "Back in my day", games came with bots you could play against, and for the most part were pretty skilled. Whether you were playing a FPS or an RTS, a lot of them could form strategies and understand the game mode. This was back when crappy 32-bit single-core CPUs were still popular. So, it really doesn't make sense to me why EA is doing this. This method is expensive and inefficient in comparison.
Some issues from past experiences (ways bots were designed), that seemed to be also in the video of this early thing, that it's not really visual based, but they're given location based information - target through objects, or more often through pass-through objects, like bushes - aren't optically fooled. Still, definitely a fun system to design and play around with.
It feels like no one is playing the game so they want Bots to fill the gaps? That's how it seems to me.
I imagined a potential method would be to deploy them with new guns/gear/camos only found in loot boxes and let them play on public servers...and win. Making people want to open loot boxes to get that stuff?
So they learn by watching real human players? Then I expect the AI to run around in circles banging into things, never spot or shoot at enemy, never drop ammo or health, and then complain on reddit that the enemy are overpowered until DICE nerf things for them.
That's consequence from them being controlled by one AI here. Secondly, you very well approximate position of other players based on sound they make. AI does it better. Sound in CS 1.6 was more than enough for you to shoot people accurately through walls. (If you knew which can be shot through.)
I think having Bots in a multiplayer game allows for cheaters to create aimbots and other cheats because they are already included in the game if bots are built into that said game. That's how cheaters used to make cheats years ago. Not sure if that's still the case now but highly likely. Imagine if you could use this AI, hard to detect because it's based off of human behavior.
Hopefully this will ultimately lead to smarter AI bots in single player games. I'm currently playing Vermintide 2, and the bots are not very intuitive; they are helpful, but not intuitive.
Your all missing the actual point they want to make these bots so they can mimic real players that way when there games are dead because everyone is boycotting EA they will deploy the bots with real names and hope people can't tell the difference / prove there bots not real players
I do not mind bots in Vermintide 2. They are more helpful than many players on higher difficulties, Especially because their skill is compensated with higher health. What I Do not like is that there is no sneaking as enemies will notice you no matter direction they are looking or how slow, crouched you move.
Am I the only one who finds this pretty cool? Ok yes. They had programmers who had to meticulously program each and every strategy by hand using a fixed-function AI model that was neither adaptable nor creative. Bots have always been cake to learn, and the only way they make them bettor or worse is simply adjusting response time and artificially nerfing accuracy. Evolving neural networks is a very expensive process, yes. Actually running a neural network is at worst as hard as traditional fixed-function AI, but more often much faster. This would mean EA would evolve the AI on their servers, and then would release an evolved neural network in a small, fast "cooked" package. This "cooked" version would be what would run on the client with significant ease. So we are talking about smarter, more creative AI with less of a performance impact and less man-hours of AI programming to ultimately achieve. Oh, and let's not forget that the vector/tensor/neuron mathematics is exactly what GPUs are built for, so we'd be able to see more GPU-powered AI as well. So like, this sounds pretty good. We wouldn't have to have AIs that are stupid but can only beat you via over-the-table knowledge and program-level cheating. Bots that feel real. We should hold a large Turing test and see if humans can tell who are bots and who are players after a system becomes more evolved.
No, I also think it's cool. It's just super hypocritical of EA. That's not true at all. I'm sure that applies in some cases, but I've played games that offered custom maps and had very intelligent AI. They could adapt to different game modes and objectives, too. Regardless, BF1 doesn't have that many maps, and most of them aren't difficult to navigate. A fixed-function AI would not be difficult to make. That's besides the point - it's the initial investment that matters most, and this particular investment just doesn't make sense. Keep in mind, EA's primary demographic is people who lack enough skill that they're willing to pay to be better. I'm pretty sure a dynamic AI is a bit overkill. For the record - I totally agree that a smarter and more creative AI is awesome, but again, you have to consider who is working on this. GPUs are struggling enough with 4K as-is. Though tensors don't really have much to do graphics, it's still going to add load the VRAM and PCIe bus. Now, if you had a dedicated server that ran GPU-operated bots, that's something I'd totally be cool with. But the thing is, when I'm playing against bots, it's because I can't or don't want to play against other people online (whether that be a lack of connection or nobody to play against). That means I'm going to want to run this locally. If I want multiple bots, that could really slow the game down. Meanwhile, you can run dozens upon dozens of "traditional" bots in FPSs with minimal load on the CPU. That sounds like a pretty cool idea. For the record, I'm not trying to attack your points, but you have to keep in mind that creating good bots has been accomplished several times. But in the past decade, developers have got lazy, because good AI is difficult to make. It's much faster and easier to not even try, and just make the game online multiplayer.