MDS Assistant is an experiment.
Hi, I'm the MDS Assistant and can help you with information from our guidelines and components.

I do not provide code and development guidance. For coding assistance, please check out our experimental MCP server.

AI Checklist

A summary and a great way to get started.

This checklist is a summary of the AI Guidelines and is a great way to get an overview of the main considerations that need to be taken into account when designing and developing AI-driven user interfaces and conversational interactions within Maersk applications.

The list below should be seen as a guiding checklist and not as a mandatory set of rules that needs to be followed strictly. Some points might not be relevant and meaningful in certain cases and contexts.

The checklist does 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.

Transparency

Done
Label AI-generated content

Use clear visual indicators such as badges, colour coding, or labels like 'AI-generated' to distinguish AI outputs from human content (comming soon).

Done
Show confidence levels

Display confidence indicators for AI predictions or recommendations (e.g., percentages or labels like low/medium/high) to help users assess reliability.

Done
Use progressive disclosure

Allow users to progressively access explanations behind AI decisions through tooltips or expandable panels.

Done
Introduce AI up front

Ensure conversational AI introduces itself at the beginning of interactions to set expectations and build trust.

User control

Done
Allow overrides

Provide users with simple and accessible options to override AI-suggested actions.

Done
Require confirmation for critical actions

Before executing any high-impact or irreversible AI-driven task, require explicit user confirmation to ensure intent and prevent errors.

Done
Offer human escalation paths

Clearly present options to escalate to a human agent or support channel, especially when the AI reaches its limitations or user confidence is low.

Done
Use quick-action prompts

Provide contextual quick-action prompts (e.g., 'approve,' 'modify,' 'contact support') in chat interfaces for easy task completion.

Ethical handling and privacy

Done
Communicate data practices transparently

Clearly explain how user data is collected, stored, used and shared using plain and non-technical language.

Done
Obtain user consent

Explicitly request user consent for personalised AI features and processing of sensitive data.

Done
Monitor and mitigate bias

Continuously evaluate AI solutions for potential biases and take corrective actions to ensure fairness and inclusivity.

Done
Reinforce privacy and security in real-time

Reassure users about data protection and security measures during interactions, especially when handling personal or sensitive information.

Secure and resilient design

Done
Use strong authentication

Implement secure authentication methods, especially for sensitive actions, e.g. multi-factor authentication (MFA).

Done
Communicate secure states visually

Use clear visual indicators, such as lock icons or secure session banners, to show when users are in a secure environment.

Done
Audit security regularly

Conduct routine security assessments of the UI and data-handling processes to identify and mitigate vulnerabilities.

Done
Verify identity in voice interfaces

Securely verify user identities in voice interactions, e.g. confirming registered contact details.

Done
Confirm sensitive actions securely

Provide visual or audible confirmation steps for critical transactions or sensitive data exchanges to prevent accidental or unauthorised actions.

Error management

Done
Communicate errors clearly

Inform users of errors using non-technical, user-friendly language that explains what went wrong.

Done
Enable easy recovery

Offer straightforward options for recovery and/or retrying actions without restarting the entire process.

Done
Disclose AI limitations

Proactively inform users when the AI has limitations and system constraints that may affect results or performance.

Done
Handle misunderstandings gracefully

Respond to unclear inputs with polite and helpful prompts, e.g., 'Please repeat the shipment number'.

Done
Guide users toward resolution

Provide clear, contextual prompts to help the user resolve issues, e.g. 'Unable to find the shipment, please verify the shipment number'.

Conversational UI

Done
Visually distinguish AI and user Messages

Use distinct visual styles to separate user inputs from AI responses, e.g., using chat bubbles or colours.

Done
Show processing indicators

Include visual cues like typing indicators or listening animations, to communicate system activities.

Done
Provide real-time transcripts

Offer live transcripts or summaries of voice interactions, especially in multimodal interfaces.

Done
Display voice activity status

Use clear microphone status indicators, such as animations to show when the system is actively listening.

Done
Enable quick-response actions

Offer quick-response buttons or actions for streamlined user interaction and reduce typing in chat interfaces.

Done
Avoid technical jargon

Use plain language and avoid ambiguous or overly technical terms to enhance understanding.

Done
Offer expandable explanations

Allow the user to access detailed insights on AI decisions via expandable panels and tooltips.

Final checks

Done
Design for intuitive interactions

Ensure that the user experience is logical, easy to follow, and feels natural across different contexts.

Done
Differentiate AI-generated content

Make it super easy for users to distinguish AI-generated content from human-created content, and introduce clear labels and visual cues.

Done
Enable explainability

Ensure that the user can easily access deeper explanations behind any AI-generated output, including reasoning, data sources, or influencing factors.

Done
Comply with Maersk's data ethics

All handling of sensitive data must fully comply with Maersk's Data Ethics and Cyber Security Policy, including storage, processing, and sharing.

Done
Provide clear consent mechanisms

Provide clear, accessible, and manageable consent mechanisms for personalised AI experiences, allowing users to opt in or out at any time.

Please write to us on our Teams Channel. We encourage and welcome any type of contribution and feedback.

With contributions from:

Mia Stigsnaes-Hansen
Martin Oliver Christensen
Fangyu Zhou