Why does (all) AI suck at telling me what I should be crafting in Divinity?

Understanding AI’s Limitations in Game Guidance: Why Does It Struggle with Divinity?

In the realm of gaming, especially complex titles like Divinity: Original Sin (DOS), players often rely on a blend of research, community insights, and intuition to navigate their journey. However, many gamers—myself included—have noticed a recurring challenge: artificial intelligence (AI) tools frequently fall short when advising on in-game crafting decisions, vendor selections, or strategic choices. This raises an important question: why does AI, despite its advancements, struggle to emulate the nuanced decision-making humans excel at?

The Human Edge in Research and Decision-Making

Humans possess a remarkable ability to synthesize information from diverse sources—be it official wiki pages, Reddit discussions, or gut instincts. When figuring out what to craft or purchase in DOS, players often consult multiple community-driven resources to arrive at well-informed choices. This process, although sometimes time-consuming, taps into collective knowledge and contextual understanding that AI currently finds challenging to replicate.

The Limitations of Current AI Models

While AI technologies like ChatGPT and other state-of-the-art models have made impressive strides, they still depend heavily on pattern recognition and available training data. This makes them proficient at generating human-like text or summarizing existing information but less effective at dynamic, context-specific tasks such as recommending in-game items or strategies effectively. For instance, an AI might prioritize certain vendors or items based on superficial data, glossing over nuances vital for optimal gameplay.

Why Does AI Underperform in These Scenarios?

Several factors contribute to AI’s shortcomings:

• Lack of Contextual Understanding: AI models do not “experience” gameplay, leading to potential misinterpretations of what is most beneficial in a given situation.
• Insufficient Specificity: General models are trained on broad datasets, which may lack the granular, game-specific detail necessary for precise recommendations.
• Static Biases: AI may overweight less relevant information, such as the importance of routing or other mechanics, due to training biases, leading to suboptimal guidance.

The Challenge of Convincing AI to “Try Harder”

As players and developers, we often joke about prompting AI to give better answers—metaphorically asking it to “try harder.” But how can we improve AI’s performance in these niche yet critical areas?

Potential Strategies for Better AI Performance:

• Refined Prompt Engineering: Be explicit about what constitutes a helpful answer. For example, instruct AI to prioritize specific outcomes, such as maximizing damage output or resource efficiency.
• Incorporating Contextual Data: Feed AI models with detailed in-game data, including current party status, player goals, or recent game updates.
• Fine-Tuning Models: Develop specialized versions of AI trained specifically on Divinity gameplay data, community guides, and vendor inventories.
• Iterative Dialogue: Encourage a conversational approach where the AI clarifies uncertainties and receives iterative feedback to hone its recommendations.

Final Thoughts: Bridging the Gap

While AI holds enormous potential for assisting gamers, it remains a complement rather than a replacement for human insight—particularly in complex decision-making scenarios like those in Divinity. As prompt engineering techniques improve and specialized training methods advance, we can expect AI’s recommendations to become more aligned with human reasoning.

In the meantime, embracing the strengths of both—AI tools for rapid information retrieval and human intuition for nuanced judgment—will lead to a richer and more effective gaming experience. After all, understanding AI’s current limitations is the first step towards leveraging its full potential in our adventures.