I’ve been thinking about the future of video games and I wonder when we will have truly intelligent non-player characters (NPCs) in large online worlds. Imagine games like Elder Scrolls Online or Final Fantasy XIV, but each merchant, guard, or random villager could actually engage in real conversations as if you’re chatting with an advanced AI. These characters would need to grasp context, remember past interactions, and move around the game world naturally instead of just staying in one place repeating the same lines. Right now, most characters seem quite robotic, but I’m interested in when the technology will evolve enough to make them feel somewhat human. Has anyone heard about developers working on this type of AI? What are the main technical hurdles that need to be addressed first?
honestly, i think we’r closer than we realize. just look at chatgpt, it improved so much in a few years. game devs are testing this behind the scenes. biggest issue is probably the cost of servers since running ai for loads of npc’s would be super pricey.
I’ve been watching indie devs mess around with this on smaller projects. The big breakthrough will come when someone cracks narrative coherence - AI can chat decently now but can’t keep storylines straight across multiple conversations. We need systems that generate quests and story branches from player choices without breaking the logic. From what I’ve seen, the winners will focus on a few smart NPCs instead of making every random villager Einstein. Processing stays reasonable when you only give deep conversation skills to story-critical characters.
I’ve been working with AI integration for a couple years now and it’s way harder than people realize. Yeah, LLMs can chat, but games need responses under 100ms or players feel the lag. Try doing that with dozens of NPCs running at once - it’s brutal.
Memory costs are insane. I built a prototype where NPCs remembered what players did between sessions. Each character racked up gigabytes of data over months and storage costs went through the roof. Most studios will probably go hybrid - use procedural stuff for small talk and save AI for important story beats.
I’m thinking we’ll see decent implementations in 2-3 years, but only in smaller games. Open world MMOs with hundreds of smart NPCs? More like 7-10 years unless someone cracks model compression.
Tech demos look incredible but production reality always hits different. Still fun to watch though.
Several studios are working on this right now, but they’re staying quiet about timelines. I spent a few years in game dev - the real bottleneck isn’t processing power. It’s building persistent character memories and behavioral systems that work for thousands of NPCs at once. Each character needs their own knowledge base, relationship tracking, and decision-making that doesn’t break when players interact with them multiple times. I’m betting we’ll see limited versions in 3-5 years, starting with premium single-player RPGs before hitting MMOs. Jess is right about server costs being a real issue, but cloud computing should help solve that faster than we think.