The following AI Design Principles list the most important areas that need to be taken into consideration to ensure that all AI-powered products are ethical, transparent and user-centered.
The AI Design Principles do not replace any existing Maersk policies, but should rather be seen as a supplement to these. If you want to know more about AI at Maersk, then visit the Maersk AI Hub.
Transparent interactions
Clear communication about AI interactions helps users understand when AI is involved, what decisions it makes, and how confident it is in its outputs.
- Clearly label AI-generated or altered content. Always distinguish AI-driven content using explicit visual cues, such as badges or subtle labels e.g., “Generated by AI”.
- Design AI to be unobtrusive yet always accessible. Surface AI when it is genuinely helpful in the user workflow e.g. Suggested correction by AI.
- Use progressive disclosure for detailed explanations. Allow users easy access to detailed explanations behind AI decisions through tooltips, expandable panels, or side drawers.
- Help users assess reliability. Make sure it is transparent how reliable the AI outputs are e.g. by displaying confidence levels.
User control and confidence
Users must always feel and be in control. Prioritise human oversight and carefully manage AI interactions to safeguard outcomes and protect users’ data.
- Enable easy user overrides and confirmations. Provide clear and accessible UI controls allowing users easily to confirm, adjust, or override AI-suggested actions.
- Require confirmation for critical tasks. Require explicit user confirmation for significant AI-driven actions such as shipment rerouting, order cancellation, or document approvals.
- Show human escalation paths. Make options for escalation to human support clearly visible and readily accessible.
- Respect privacy policies. Provide transparency on how personal data is handled, complying with GDPR and Maersk’s Data Privacy standards.
Error handling and recovery
Graceful error handling when the AI solution provides a wrong output. It reassures users, maintains trust, and reduces frustration during AI-driven interactions.
- Use actionable error messages. Use understandable, non-technical language that informs users what went wrong and how to recover.
- Provide quick recovery options. Offer a straightforward way to retry or correct errors without starting over from scratch.
- Design transparent feedback loops. Embed transparent feedback loops, allowing users to report issues or suggest improvements. Regularly assess and improve the AI´s ability to meet evolving user needs and expectations.
Ethical data handling and privacy
Responsible data handling when interacting with AI maintains user trust, ensures regulatory compliance, and upholds Maersk’s reputation for reliability and integrity.
- Show how the data will be used. Communicate how user and company data will be utilised, stored, and protected when shared with AI solution. When it makes sense, ensure encryption and secure handling of sensitive data in compliance with Maersk’s encryption standards.
- Ask for user consent within conversations. In conversational flows, explicitly ask for and confirm user consent - before personalised features or data collection.
- Mitigate AI bias. Regularly assess AI systems for biases and communicate the criteria and data behind AI recommendations (e.g AI suggests suppliers only from a single region).
Please write to us on our Teams Channel. We encourage and welcome any type of contribution and feedback.