How We're Creating Character Intelligence in Project Etherea
Good morning, everyone! I hope you're doing well. Today I’m here to share a bit about the topic we discussed live yesterday on our YouTube channel.
Currently, our game already has over a thousand lines of code and has gone through a major process of changes and improvements. However, yesterday, my programmer and I decided to talk about how the inhabitants of Project Etherea will think, feel, and live their lives.
We explored how we’ll develop the intelligence system and which approaches we’ll take to make it truly unique! I translated a few topics below, but you can watch the full video with subtitles enabled to see everything in detail!
Life in the game takes place in different locations, and the characters' intelligence must adapt based on their proximity to the player. This way, we can deliver complex responses while keeping computational costs low.
The key difference in NPC intelligence lies in the level of detail, which can range from simple actions to complex emotional reactions.
The AI model we’ll use is specific to the game’s characters and should not be confused with other AI applications like chatbots. Our approach is designed to create a richer, more dynamic experience focused on interaction and NPC evolution.
We’re building a game that challenges players to explore new types of gameplay.
We gave a brief introduction to the concept of a "perceptron" in machine learning, which simulates the function of a human neuron.
We explained how perception works through inputs and weights, resulting in a weighted sum.
We discussed creating neural networks, where multiple connections allow the system to learn to recognize patterns.
We plan to build the neural network code from scratch.
We’re considering using language models like Gemini and ChatGPT to help create expected inputs and outputs for training, but it's likely we’ll do much of the groundwork manually to make it even more customizable.
We explained the concept of “inputs” and “outputs” using simple behavior examples, like the relationship between feeling hungry (input) and eating (output).
We shared more about the process of teaching patterns to the neural network, showing how NPCs can learn to act in situations like being tired or hungry.
Example:
Character is sleepy and hungry → Action: Eat and sleep.
Character is sleepy, hungry, and stressed → Action: Eat hastily → sleep and have a bad night’s rest.
NPC decisions impact the environment around them, creating a web of emotional reactions.
The game not only simulates characters' emotions but also reflects emotional states in objects and environments, like benches that “witness” these interactions.
The intensity of events, such as romance or arguments, determines how the environment prioritizes its future reactions, influencing new encounters between characters.
We discussed lowering bureaucratic costs by using a free development platform: Godot.
We defined what an ideal NPC is: one who not only takes care of their own needs but is also aware of their surroundings.
We also talked more about our vision of a synesthetic life simulator, where characters have a rich variety of sensory inputs.
And we shared some of the motivation behind our desire to create our own life simulator. The conversation explored the complexity of decision-making in simulation games, how player expectations often differ from reality, and how we’re working to ensure players perceive and feel what’s happening—not through a data-heavy interface, but through visual, sensory, auditory, and environmental feedback.
We want you to live the experience, not just rely on intrusive interfaces to understand your context.
I hope you enjoyed it, and I invite you to take a closer look at everything we discussed in our stream. ❤️
Thank you for your ongoing support and encouragement!
DESCUBRA COMO OS HABITANTES DO NOSSO JOGO VÃO PENSAR SENTIR E EVOLUIR (YouTube)