What do you do if your game AI is not realistic enough?

Evaluate whether the AI's decisions align with what a human or real-world counterpart would do in similar circumstances. Are the decisions too predictable, overly aggressive, or lacking strategic thinking? Does the AI respond appropriately to obstacles, cover, or changes in terrain? Is the interaction with objects fluid and believable? Does the dialogue or behavior reflect realistic social dynamics, emotions, or personality traits?

At Carry1st, when we integrated bots into our mobile board game to address player liquidity issues, we observed an unintended increase in player churn. This led us to hypothesize that the absence of genuine social interaction was detracting from the player experience. Our approach to addressing this was multifaceted. By enabling bots to send emojis and make snarky comments, we tried to inject a sense of personality and unpredictability into their interactions. This tweak aspired to create a more engaging and dynamic gameplay environment. It shows the importance of understanding player expectations and the nuanced ways they engage with game AI.

The initial step in enhancing your game AI involves pinpointing its shortcomings. Is it too predictable, challenging, or dull? You need to examine your game AI from various perspectives, including the player's experience, the objectives of the game design, genre expectations, and any technical constraints. Gathering feedback and data through methods like playtesting, surveys, analytics, debugging, and reviews is crucial. Once you've identified the issues, you can prioritize the most pressing and impactful solutions to address them effectively. This detailed analysis ensures that your efforts are focused on making meaningful improvements to your game's AI.

Choosing the right AI technique is pivotal when addressing the difficulty curve, which is often underappreciated in its complexity. We experimented with various difficulty progression models—linear, exponential, and flowing—to find the sweet spot that aligns with player expectations. Data showed that players favored a high win rate, around 90-95% (D1-D5), and deviations led to increased churn. This insight was invaluable. It guided us to tailor the difficulty curve using a data-driven approach, ensuring that the AI challenges players while maintaining engagement. Essentially, it's not just about selecting an AI technique but also about adapting it based on behavior data to strike that delicate balance between challenge and enjoyment.

Once you've identified AI issues, the next step is choosing the right technique to fix them. Options include finite state machines, behavior trees, rule-based systems, neural networks, genetic algorithms, or reinforcement learning. Consider factors like complexity, performance, and entertainment value to make the best choice for your game.

The third step involves leveraging existing tools to create or improve your game AI. Numerous options exist for game developers, including general-purpose frameworks, specific plugins, and assets. Examples include Unity ML-Agents, Unreal Engine Behavior Trees, GameMaker AI GML, Godot AI Framework, and AI Toolkit. While these tools can streamline development by providing ready-made solutions or templates for common AI tasks, it's essential not to rely on them blindly. They may not perfectly suit your game or offer sufficient customization and control.

We learned early on that our initial implementations rarely hit the mark straight out of the gate. Hence, establishing a robust framework for AB testing and actively analysing the player's response through various feedback loops is a game changer. This also aligns it with player expectations, which are often unpredictable and varied. By setting clear expectations for each iteration and being adaptable in our approach, we've managed to enhance player engagement significantly (think 33% to 40% D1 retention). Remember, the essence of a successful AI isn't just in its creation but in its continuous evolution in response to real-world player interactions.

The final step involves thoroughly testing and iterating your game AI until it meets your goals. Testing ensures that your AI functions properly, without bugs or glitches, and aligns with player expectations and game design. Experiment with different scenarios and tools like unit testing, integration testing, stress testing, profiling, and debugging. Gather feedback from players through playtests, surveys, analytics, or reviews. Based on this data, iterate on your AI, making necessary adjustments or enhancements to achieve the desired outcome.