Think of Artificial Intelligence (AI) as a cool friend who is good at tasks that require intelligence. This friend is not a human but a computer program or a machine. It can do things like understanding human language, recognizing patterns, learning from experience, and even making decisions. AI is like the brain behind your smartphone assistant, the recommendation system of your music app, or even the chatbot you are talking to right now! But remember, it is not human – it is a set of algorithms running on a computer. 

A responsible AI model is an AI system that is designed, developed, and deployed in accordance with the principles and practices of responsible. 

Transparency

The principle of Transparency in AI is like a clear window into how an AI system works. It means: 

• Telling people when they are interacting with an AI, not a human. 
• Helping people understand how the AI system works and makes decisions. 
• Explaining the AI’s decisions in a way that people can understand, especially if they are affected by these decisions. 

However, it doesn’t mean sharing all the technical details or codes that run the AI system. It is more about understanding the “how” and “why” behind AI decisions. 

Fairness 

Fairness is the ability of an AI model to treat all the individuals and groups in a fair and equal way, without discriminating or favoring them based on their characteristics, such as age, gender, race, ethnicity, religion, disability, or sexual orientation. Fairness also means that the AI model should respect the diversity, inclusion, and dignity of the people, and improve their social and economic well-being. 

Privacy and Security 

Privacy means that AI systems should respect the rights and preferences of users regarding their personal data and protect it from unauthorized access or misuse. Security means that AI systems should be resilient to attacks and threats and ensure the integrity and confidentiality of data and processes. Privacy and security also ensure that the AI model respects the rights and choices of the data subjects, such as users, customers, or employees, and follows the rules and laws that apply, such as the General Data Protection Regulation (GDPR). 

M365 Copilot 

Copilots follow Microsoft’s commercial and consumer data handling commitments, like other Microsoft 365 services. These commitments protect the privacy and confidentiality of the organization’s data.  

This means that no other tenants or user can access an organizations’ data without delegated permissions. permission. This applies to the Large Language Models (LLMs) as well which means that as a customer of Microsoft 365, your organizations’ data will not be used to train Copilot LLMs without approval. You always have control over how, when, and where your data is used.Copilot is meant to provide tailored responses, which may pose a question in mind if the data is being trained into an LLM or be used as a benefit for other tenants. Not only this when you prompt Copilot for a response, the information contained within the prompt, the data that is retrieved, and the generated responses stays within the M365 service boundaries. The data retrieved uses the context of the user, so only data that is retrieved is what the user has permissions to. 

It should not be surprising to hear that the copilot request does traverse multi-tenant shared hardware and in multi-tenant shared service instance, data protection is ensured through extensive software in depth against unauthorized use. Copilot encrypts all data in transit and rest, applies Role based access control and auditing policies and aligns to industry standards and regulations such as GDPR. 

Let us walk through an example of simple Copilot processing flow to understand what happens behind the screens:

Step 1: User submits the prompt to Copilot, say for example “IT policy Documents”. This goes to the Orchestrator and guess what it does the job of orchestration of the request. 

Step 2: Orchestrators send the search request using the context of user credentials to ensure that only the data that user has access to is retrieved, thereby keeping the search limited to M365 service boundaries. Orchestrator selects the best result from the search and uses a process called prompt engineering to add the information about “IT Policy Documents” to the prompt. 

Step 3: Copilot orchestration engine sends the prompt to the LLM to process the prompt to produce output. 

Step 4: Copilot subjects output generated to compliance and data protections policies using Graph API 

Step 5: Orchestrator produces the final response and shares it with the end user. 

Conclusion

Microsoft 365 Copilot represents a groundbreaking integration of responsible AI within the workplace, embodying principles of transparency, fairness, privacy, and security. By leveraging state-of-the-art Large Language Models (LLMs) while strictly adhering to Microsoft’s stringent data handling commitments, Copilot ensures that organizational data remains private, secure, and used solely for the intended purpose of enhancing productivity and decision-making.

The meticulous orchestration of user requests, coupled with robust compliance and data protection mechanisms, exemplifies Microsoft’s commitment to fostering a trust-based relationship with its users. In essence, Copilot is not just an AI assistant; it is a testament to the possibility of harnessing the power of AI in a manner that is both ethical and beneficial to all stakeholders involved.