Human oversight
Designers should set testing objectives and retain authority to reject AI-recommended changes. Automated test results should inform, rather than replace, creative judgement about difficulty, pacing and enjoyment.
Business & innovation IWIB4AI session Needs review
Power Coffee at King – Leveraging AI to Manage 18,000 Candy Crush Levels
AI helps test and tune 18,000 Candy Crush levels while balancing operational efficiency, human creativity and player wellbeing.
Draft – legal and case law references not yet verified
Summary
King uses AI to support testing and tuning across 18,000 Candy Crush levels for a global player base. The use case combines high-speed level testing and AI-driven A/B testing to support large-scale game operations. It illustrates how automation can deliver business value while raising questions about creative control and the objectives used to optimise player experiences.
The Problem
AI can test gameplay at a scale that human teams cannot easily match, but automated performance measures may not capture enjoyment, accessibility or creative quality. The central challenge is deciding what to optimise and retaining meaningful human judgement over changes that affect players.
Ethical Grounds
Fairness
Testing should examine whether level changes work for players with different abilities and play styles. Aggregate performance improvements can conceal a worse experience for less-represented player groups.
Autonomy
Optimisation should support enjoyable play rather than exploit frustration or vulnerability to encourage excessive play or spending. Engagement and commercial metrics should be balanced with measures of player wellbeing.
Privacy
Where A/B testing uses identifiable or pseudonymous player behaviour, data collection should be limited to what is necessary. Simulated testing and genuinely anonymous results can reduce exposure of personal information.
Legal Compliance
| Framework | Provision | Relevance |
|---|---|---|
| GDPR | Arts. 5, 6 and 13 | Where testing uses personal player data, King must establish a lawful basis, respect purpose limitation and data minimisation, and provide appropriate information about the processing. |
| GDPR | Art. 25 | Player-data testing pipelines must implement data protection by design and by default, including limiting identifiability, access and retention where appropriate. |
| EU AI Act | Art. 4 | As an AI deployer, King must take measures to ensure sufficient AI literacy among relevant staff, including understanding testing limitations and interpreting optimisation outputs. |
| EU consumer law | Directive 2005/29/EC, Arts. 5 and 8–9 | Where AI-tuned gameplay forms part of a commercial practice affecting purchasing decisions, it must not constitute an unfair or aggressive practice, including through undue influence. |
Sources
No sources linked to this use case yet.
Key Takeaways
- Keep human designers accountable for optimisation objectives and consequential gameplay changes.
- Validate automated test results against real player experiences across abilities and play styles.
- Balance engagement and business metrics with accessibility, enjoyment and player wellbeing.
- Use simulated or anonymous testing where feasible and apply GDPR safeguards when processing personal data.
Related Use Cases
From the Session
Power Coffee at King – Leveraging AI to Manage 18,000 Candy Crush Levels
Speakers
