AI Safety

Creating a safe interaction

Date:

Jan 7, 2025

Ensuring Safe AI Interactions for Children

At InteractionLabs, we believe in pushing the boundaries of AI-driven storytelling while ensuring a safe and ethical experience for children. To achieve this, we are collaborating with the Max Planck Institute to work closely with leading psychiatrists and Pr. Iyad Rahwan, a pioneer in the field of Machine Behavior and an expert in AI ethics. Together, we are conducting rigorous A/B testing to evaluate how our AI-driven storybox affects children’s cognitive and emotional development, ensuring that our models align with the highest ethical and safety standards.

Technical Measures for AI Safety

To create a trustworthy and secure AI experience, we are testing multiple classic and state-of-the-art safety mechanisms to ensure child-appropriate interactions. Some of these include:

1. Filtering and Guardrails

Since our AI models run in the cloud, we use multiple safety layers to review and refine all outputs before they reach children. These measures:

  • Filter inappropriate content through a carefully designed content moderation system.

  • Ensure consistency by guiding responses along structured storytelling paths.

  • Avoid sensitive or unsuitable topics, keeping interactions appropriate for different age groups.

2. Reinforcement Learning with Human Feedback (RLHF) for Child-Safe Outputs

We fine-tune our AI using RLHF, where:

  • Psychologists and child development experts curate and rate AI-generated responses.

  • The model learns to prioritize safe, educational, and positive narratives over uncertain or sensitive topics.

  • Continuous feedback loops ensure adaptation to real-world interactions while preventing bias or harmful content.

3. Adversarial Testing & Red Teaming for Safety

Before deploying updates, we run:

  • Automated adversarial attacks to probe vulnerabilities in the model.

  • Human-in-the-loop evaluations where experts try to push the model to unsafe territory and refine its guardrails.

  • Continuous A/B testing in collaboration with child psychologists to assess the long-term impact of AI interactions.

Building a Safe and Enriching Storytelling Experience for Kids

Our mission is to make storytelling engaging, educational, and above all—safe for children. By working with top experts and applying strict safety measures, we ensure every interaction is positive and enriching.

We are dedicated to ongoing research, transparency, and collaboration with parents, educators, and child development specialists to create the highest standards for interactive toys.

Want to learn more or collaborate with us? Get in touch!