Inworld AI made a notable appearance at GDC 2024, showcasing innovative demos of its AI Character Engine in partnership with major players in the gaming industry such as Microsoft, Ubisoft, and NVIDIA. Fast forward to GDC 2025, this year’s representation was more subdued, lacking the high-profile collaborations that characterized their previous outing. However, there were still significant developments brewing behind the scenes. We had the opportunity to speak with Inworld AI’s CEO, Kylan Gibbs, to gain insight into the company’s recent activities.
Reflections on Company Evolution
Inworld AI has been on a fascinating journey over the last four years. Initially, the focus was on developing a character engine, which functioned as a server-side application that connected to various game engines through an SDK. This approach aimed to simplify AI complexities for designers and narrative writers. However, feedback indicated a strong desire for more control; developers wanted the ability to run logic locally.
This led to a significant pivot in our product strategy. We have transitioned from a server-based model to assembling our technology into libraries that developers can utilize directly within their game engines. This shift means that our tools now offer a C++ runtime that can be adapted for multiple engines.
Another critical area of focus is observability and telemetry. Many game developers struggle to understand AI operational transparency, which often complicates troubleshooting. Our portal tool provides developers with essential telemetry insights, allowing them to diagnose issues effectively. Our goal is to promote local logic and ideally local models, aligning with every game developer’s desire for seamless integration.
We’ve innovated a tool that leverages our cloud resources to distill models for local use. However, we recognize the variability of consumer hardware capabilities. To address this, we’ve introduced a hybrid inference model that enables local storage of the AI model while seamlessly falling back on a cloud version when necessary. For instance, a powerful PC equipped with a GeForce RTX 5090 could run everything locally, while a Nintendo Switch would utilize the cloud to provide the necessary computing power.
Challenges in AI: Controlled Evolution
A pressing challenge in the AI landscape for both games and consumer applications is model obsolescence. The rapid pace of AI development means that a game launched today with its current model may become outdated within six months. Developers need the capability to continuously select the most effective models from both third-party sources and our own offerings, optimizing them based on current user engagement.
Our objective is to collaborate with developers, allowing them to avoid large-scale commitments to specific cloud providers. Instead, they can adopt the most advantageous models available and customize them for their unique requirements. This is particularly important as models are typically engineered for broad tasks, whereas in gaming, we aim for excellence in niche applications.
Trends in Gaming: Exciting Developments
As AAA game development cycles tend to be lengthy, we observe a surge in mobile browser-based applications. One standout example is Status by Wishroll—an innovative role-playing game where users engage as characters in a Twitter-like universe. Remarkably, they reached 500, 000 users within just 19 days, with players spending an average of an hour and a half each day on the platform.
Another noteworthy title is Little Umbrella. Their team, known for Playroom, has successfully released several engaging games, including The Last Show, which operates similarly to a Jackbox-style party game but incorporates real-time AI orchestrations for multiplayer scenarios.
We also collaborated with Streamlabs to create an advanced streaming assistant that operates within Fortnite. This system monitors gameplay dynamics in real-time, processing user comments and streamer interactions to deliver actionable insights with minimal latency. Such cutting-edge applications highlight the necessity of hybrid and local inference for optimal performance.
Gaming Assistants & Companions
When it comes to functionality, Streamlabs serves as a gaming coach, albeit typically for seasoned streamers who may not require extensive guidance. However, it can assist novice players by providing contextual tips and strategies. We see two main types of AI companions: disembodied assistants, like Streamlabs, which operate externally, and embodied companions that are integrated directly into the gaming experience.
Embodied companions can enhance onboarding and gameplay experiences by providing real-time suggestions based on player actions, ensuring a more engaging interaction. These companions can also help ease gameplay milestones by offering hints when players face difficulties.
Additionally, we’re witnessing emerging use cases in multiplayer settings, where AI can simulate player behavior or fill in when human players drop out, creating a continuous gaming experience.
The technology is adaptable, with capabilities to run on advanced hardware like the NVIDIA GeForce RTX 5090 while maintaining compatibility across a range of devices. Although we support local execution of models, cloud storage serves as a backup for when local hardware limitations arise.
Enhancing Game Monitoring
In response to queries regarding effective game monitoring, we recommend employing visual language models or OCR technology to oversee critical on-screen regions such as mini-maps in MOBAs. Developers typically establish pixel-specific settings within the game to facilitate monitoring, though it is essential to recognize that sometimes the game state can provide sufficient data without needing extensive visual analysis.
Quality Assurance Testing with AI
While Inworld AI’s primary focus is on enhancing player-facing experiences, its technology can support quality assurance (QA) testing through custom development. Our framework allows studios to craft their own testing agents using our tools, enhancing testing efficiency and creativity during the prototype phase.
For developers seeking to monitor game states and potential unintended interactions, our telemetry solutions are invaluable. By integrating telemetry within the game code, developers can evaluate character responses and NPC interactions to refine gameplay experiences significantly.
Tackling AI Hallucinations
In addressing AI hallucinations, our ability to create specialized models and use validation measures helps mitigate this issue. For instance, within games like Status, while some creative latitude is allowed for characters to express unique ideas, we maintain rigorous control over data structures to prevent nonsensical outcomes.
Advancing Dynamic Crowd Technology
Dynamic crowd technology aims to breathe new life into gaming environments that have stagnated over the last decade. By simulating more complex interactions—such as characters communicating or taking action based on their environment—we can craft a more immersive experience that makes game worlds feel alive. We are focusing not just on enhancing individual characters but also on enriching the overall game state, generating quests, and creating immersive events dynamically.
Empowering Developers with Toolsets
Inworld AI provides template-based frameworks that simplify the integration of our technology into games. These templates foster easy adaptability for different developers, ensuring they construct their experiences within their unique game worlds effectively.
Investing in Metahumans: Current Trends
With the gaming industry’s shift towards stylized characters over hyper-realistic metahumans, it’s clear that developers are increasingly favoring uniqueness and identity in character design. Games like Metaphor: ReFantazio showcase high-fidelity yet distinct character design, indicating a broader trend away from the limitations posed by metahumans.
Latest Collaborations and Future Releases
We are currently collaborating with Nanobit on their interactive novel game, Winked. Each episode draws players into the narrative, and characters become beloved figures throughout the experience. Integrating Inworld AI means players can interact with these characters between episodes, keeping their engagement alive.
Looking ahead, while I can’t disclose information about specific AAA titles due to confidentiality agreements, we anticipate announcing significant projects during our upcoming showcase in June. Make sure to stay tuned for exciting new developments in AI-driven gaming.
Closing Remarks
Thank you for joining us in this discussion about Inworld AI’s advancements and visions for the future of gaming.
Related Articles:
Michael Bay and Sydney Sweeney Collaborate with Universal on Film Adaptation of SEGA’s Out Run
13:59
Exclusive PlayStation 5 Pro Teardown Reveals Advanced Technology Inside the Console
11:37
The Challenge for The Witcher IV: Establishing Ciri’s Importance Alongside Geralt as the Central Character, with Narrative Insights from Cyberpunk 2077
11:08
Leave a Reply ▼