"AI body movements" for NPCs already seems limited to whatever non-AI (PC) body movements are. If anything, pre-scripted (cut-scene) animations already outdo pretty much all "player is running, use standard 'running' sequence for avatar" stuff for multiplayer. If your MMORPG/equivalent players obtain "a funny little victory dance" perk that they can activate at will then a non-AI NPC can do exactly as much to copy the player pressing B-Triangle-LeftTrigger (or whatever it is) that then just invokes the feature that has to have been involved.
If you definitely mean "not bashing into walls", I can probably say that if Lara Croft could bruise then my playing of the original Tomb Raider, back in the day, would have turned her black and blue (even in the bits without actual enemies, or fatally vertical geology/architecture, just trying to run round a maze of corridors trying to find various right blocks to push and pull to let me into the next maze of corridors). Game-sprite controllers were almost always much better at basic navigation (by simple pathing) than an inexpert player, they just lacked the intelligence to discover that one corner where the perfect shot was, or that exact sequence of jumps that might get you through a gap in the intended virtual-glass-tunnel the level designer thought they were constraining the player in.
There's better and worse non-AI NPC 'brains', of course. You don't get Wolfenstein guards covering each others advances as they try to flush you out, and also a perfectly calculated shot on an exploding barrel would also be a game-killing experience as well, if it was what happened when a game-engine knows that this is what would ultimately kill the player-avatar and end the run.
Some form of collaborative (but hopefully not too collaborative) flocking behaviour does happen (or anti-flocking, but similarly using avoidance and accounting for all the fellow NPCs it can see), and you could get realistic (low-density) crowds spawning on the streets of San Andreas, I think, that weren't dumb enough to clip each other or step out in front of traffic. (Normally, at least when currently unaffected by the player's "demolition derby"-driving or "Dallas book depository"-sniping.)
Pre-programmed behaviours can be lacking, but can also be comprehensive. AI just means that you start with virtually all options on the table (whether it be a twitch of a leg, or the freedom to run off a cliff) and then with broad stroke rules (don't fall over/don't try to headbutt a train or anything else moving/static) develop a more intangible ruleset of behaviours that work with what interactions a PC might expect (some enemy avatars might be expected to attack on sight, others be more tricky; a shopkeeper avatar probably shouldn't attack at all unless it's that kind of game and the player-character now has that kind of rep), whilst obeying the game-universes various physical rules as much as necessary ('elemental spirits' might be allowed to noclip the environment a bit!).
Throwing in "AI makes a game better" begs loads of questions, though. What's currently lacking? Is it suffering from insufficient variation? Or from too much pre-programmed non-sequitur? Are you trying to take rails off the NPCs or add new psychological rails to the player? Are you trying to fill a multi-user environment with more 'users' at quiet times, without anyone realising? [...etc...] And how will your AI accomplish this?
I play Urban Dead, a very simple web-game, and supposedly everyone you meet (or get attacked by) is a real person logging in and responding to how everyone else who logs in moves through the environment, fixes (or breaks) things, heals or 'hugs' those that are currently humans while shooting or needling those currently zombies. Certainly no official server-side NPCs. I have no doubt that some active 'players' are something like browser-scripted automata (perhaps just to wander round and avoid trouble, maybe some are doing zerg functions for others, padding up a one-man "mall tour" wrecking spree with 'outrider' characters who can at least spy ahead). Though why you'd want to do that is another matter. You don't have to talk with your fellow players (English or Zombish or whatever comes natural to you), but a "broadcast zerg" that tries to find powered radios, sprays graffiti or speaks direct information/insults to anyone it meets could be hand-guided, pre-scripted with a "message" or tap into a GPT engine and do pretty much the very same thing in 95% of any resulting interaction with an actual player. And this in all an environment where the training and deployment of an all-singing-and-dancing AI is simplified by the many (externally reproducable) restrictions. The limit of actions per day, interactions with the world, length of messaging (speech, radio, graffiti, 'SMS'), etc. And if it 'hears' someone report "2Z SE mall, doors open, damaged gennie" and doesn't comprehend its meaning... well, probably not all human players do, at first, and at the very least its human "controller" can decide whether to add semantic training to it (or give ChatGPT a chance to query its own knowledgebase on the issue).
Fortnite-like environments will have a lot more challenges (for human players, too), with so much more 3d *stuff* (as pure data or otherwise) and nuances, and a pre-programmed 'bot-character might already be indistinguishable from a given quality of human player. If you need them to live-chat (especially in audio) with actual players then that's another thing, but I imagine that's also not compulsary.
Obviously offline (especially large-map sandboxy) games lack anyone real, so the plan is to replace current NPCs (perhaps a little predictable/unhelpful) with AINPC variations? Still basically scripted, just far more loosely. More unpredictable, possibly far more unhelpful (or not as valid in thebofficial role of an adversary) at the same time as a consequence, but that depends on the pre-training and QC.
I'm sure some of these things are not answers in search of a question, but as a broad sweep I'm not sure I see the excitement in most of the contexts. Interesting ideas, but a bit like saying that something "now has Blockchain", perhaps. Specific examples might shine through, of course, and populating a simulation with (learnable?) AI agents and seeing how far it goes does intrigue me. We shall see.
Poll
4 (44.4%)
Author
Topic: What will save us from AI? Reality, the Universe or The World $ Place your bet. (Read 6030 times)