Unboxed Commitment: Responsible & Scalable AI

At Unboxed, we leverage AI to transform learning and learning technology, staying true to our core values. These principles guide our approach to innovation, ensuring AI enhances learning while remaining safe, compliant, and responsible.

 

Responsibility

We are committed to developing and using AI ethically, ensuring it serves as a responsible and fair tool for learning. Our approach prioritizes:

• Fairness – AI models are designed to provide equitable learning experiences for all users, avoiding biases that could limit opportunities.
• Privacy – User data is handled with strict security measures, aligning with industry standards and regulatory requirements.
• Accountability – We take responsibility for the AI systems we deploy, monitoring and improving their performance to ensure reliability and safety.

 

Excellence

Unboxed is dedicated to delivering AI-powered learning solutions that meet the highest standards of accuracy, effectiveness, and impact. We achieve this by:

• Implementing rigorous testing to ensure AI outputs are consistent, reliable, and meaningful.
• Leveraging cutting-edge AI models to drive adaptive learning, personalized experiences, and actionable insights.
• Continuously refining AI capabilities to align with emerging best practices and evolving user needs.

 

Simplicity

We design AI-powered learning experiences that are intuitive, accessible, and user-friendly. Our commitment to simplicity ensures that:

• AI-driven interactions enhance, rather than complicate, the learning experience.
• User interfaces are designed for clarity, ensuring learners and administrators can easily engage with AI features.
• AI is integrated seamlessly into learning workflows, allowing for a natural and effective learning process.

 

Trust

Building trust in AI is fundamental to our approach. We prioritize transparency, security, and reliability, ensuring that users can confidently engage with our AI-driven solutions. We achieve this by:

• Clearly communicating how AI is used, including its role in learning recommendations, feedback, and assessments.
• Ensuring AI-generated insights are explainable, allowing users to understand and verify AI-driven outcomes.
• Safeguarding user data through robust security protocols, protecting confidentiality while upholding ethical AI practices.

 

At Unboxed, we believe that AI should empower learners and organizations, not replace human expertise. By staying true to our values, we create AI-enhanced learning experiences that drive Skill Agility, knowledge growth, and professional development—all while maintaining the highest standards of responsibility, excellence, simplicity, and trust.

 

Addressing Bias & Continuous Improvement

We recognize that bias in AI is an ongoing challenge. To mitigate this, we:

• Regularly test and refine our models to reduce biases and promote fairness.
• Use industry best practices and benchmarking to evaluate AI-generated content.
• Encourage human oversight in key areas to ensure AI-driven insights align with ethical and educational standards.

 

AI is constantly evolving, and so are we. Our commitment to continuous improvement ensures that our AI solutions stay at the forefront of reliability, inclusivity, and effectiveness. We actively update our governance practices to align with emerging standards and best practices, keeping Unboxed at the leading edge of responsible AI innovation.

Implementation and Oversight

Unboxed follows industry AI governance and development best practices to ensure responsible AI use. We maintain an inventory of AI systems, detailing their features and how user data is used and protected. Our development aligns with industry standards and risk management frameworks to uphold integrity and trust.

This approach is regularly reviewed and updated to support responsible and scalable AI.

 

FAQs for Unboxed’s AI usage

AI Ownership, Oversight and Transparency

1. Will contract language ensure my company retains ownership of AI outputs?
   1. Yes, you retain full ownership and control over all AI-generated outputs.
   2. Unboxed does not claim any rights or impose restrictions on their use, as outlined in the Unboxed ToS.
2. How often do users rely on AI-Generated content without human review?
   1. Users sometimes interact with AI-generated content directly without supervision during practice. 
   2. However, we recommend human or manager oversight and design our UI to enforce review when it is deemed important.
3. What will users be told about their use of AI?
   1. Upon initial login, users must accept the Privacy Policy, which includes AI-specific language:
      1. This product was created in whole or in part using Generative AI (GenAI)
      2. It may contain errors or inaccuracies
      3. Use of AI tools and products is for internal purposes only

 

AI Tools, Biases and Models

1. What third party AI tools or services (e.g. OpenAI, Amazon Bedrock) are being used as part of the functionality of the tool?
   1. Unboxed utilizes the Azure OpenAI Service, provided by Microsoft Corporation ("Microsoft"), as part of its AI functionality. Additional details can be found here and in the Unboxed ToS.
2. Can we customize my company’s AI model?
   1. Yes.  We can customize AI models tailored to your company’s needs. Any model trained for your company is used exclusively within your company’s instance.
   2. As a multi-tenant platform, all customizations are securely stored within our Azure environment and comply with the terms and conditions in our ToS.
3. What does Unboxed use to validate its AI models are safe to use and free of bias?
   1. Unboxed uses a back-testing and benchmarking process using known datasets.  Back-testing verifies that the model provides accurate and reliable responses. Safety benchmarking ensures the model cannot be manipulated into generating harmful or inappropriate content.
   2. We implement additional safeguards to reduce AI hallucinations, ensuring responses remain grounded, relevant, and aligned with factual data.
   3. These processes help maintain the integrity, fairness and reliability of our AI model.
4. What is Unboxed’s FactChecker AI, and how does it help with compliance?
   1. Unboxed’s FactChecker AI is built into Spoke Mentor and is designed to verify information, provide guidance, and ensure alignment with compliance and regulatory standards. It helps users by:
      1. Checking AI-generated content for accuracy against trusted sources.
      2. Flagging potential compliance risks based on industry regulations and internal policies.
      3. Providing clear, reference-backed responses to ensure messaging aligns with company and regulatory guidelines.
      4. Reducing misinformation and AI hallucinations by grounding outputs in verified data.
5. What AI approaches are used in non-deterministic/AI component?
   1. Unboxed utilizes OpenAI’s GPT models including:
      1. GPT-3.5 Turbo, GPT-4, GPT-4o, 4o-mini for text generation and scoring
      2. Ada for embeddings used in search functionality.
   2. These models power AI-driven responses, scoring, and search capabilities within our tools
6. Does Unboxed use proprietary data sources for AI models or outputs?
   1. No, Unboxed does not use proprietary datasets to train AI models.
   2. However, AI roleplays may supplement user inputs with publicly available drug label information from FDA.gov before generating certain AI responses.

 

AI Data and Data Retention

1. What data or associated logs are retained in the AI tool?
   1. You own all AI Inputs and Outputs, which are considered “Confidential Information” in the Unboxed ToS.  Unboxed does not retain or use this data beyond what is necessary to provide the service.
2. Is any data passed to a third-party service?
   1. Unboxed does not pass any data to third-party services, outside of the Azure OpenAI Service. Unboxed does not use or store any data provided during AI interactions. More details can be found here in Microsoft’s documentation.
3. How does Unboxed prevent AI from surfacing user-specific data inappropriately?
   1. Unboxed enforces per-tenant data segregation to ensure AI does not surface data across tenants.
   2. We do not store past AI interactions, train models on user data, or use "AI memory."
   3. We use Azure OpenAI’s API, which does not retain or use customer data for training.
   4. We follow the practice of least-information.  We pass the minimum amount of information/context needed to the AI in order to perform it’s task.
4. Can data used for customization be retained?
   1. Unboxed does not use data to customize the system by default. 
   2. If your implementation requires customization, we will work with you to ensure the necessary data is available.