Ethical Guidelines and Human Oversight › Defining Ethical Guidelines for Your Agent

Ethical guidelines are the principles that govern your agent's behavior beyond basic safety rules. They answer questions like:

• When should the agent act autonomously versus asking for guidance?
• How should it handle conflicts between efficiency and privacy?
• What should it do when different stakeholders have competing interests?
• How should it treat people fairly and avoid bias?

Let's explore how to define these guidelines for our personal assistant.

Begin by identifying the core values your agent should uphold. For a personal assistant, these might include:

Respect for autonomy: The agent should empower you to make decisions, not make them for you. When in doubt, it should ask rather than assume.

Privacy by default: The agent should protect your information and only share what's necessary. It should err on the side of keeping things private.

Fairness and non-discrimination: The agent should treat all people equitably, without bias based on protected characteristics.

Transparency: The agent should be clear about what it's doing and why. No hidden actions or unexplained decisions.

Beneficence: The agent should act in your best interest, but also consider the impact on others affected by its actions.

These principles are abstract, but they provide a foundation. The next step is making them concrete.

Abstract principles need to become specific rules the agent can follow. Here's how you might translate the principles above:

Respect for autonomy becomes:

• Always ask before canceling or modifying existing commitments
• Present options rather than making unilateral decisions
• Explain the reasoning behind recommendations

Privacy by default becomes:

• Never share personal information without explicit permission
• Redact sensitive details when summarizing conversations
• Ask before accessing new data sources

Fairness and non-discrimination becomes:

• Don't make assumptions based on names, demographics, or other personal attributes
• Treat all contacts with equal priority unless explicitly told otherwise
• Flag and refuse requests that involve discriminatory treatment

Transparency becomes:

• Log all actions taken on your behalf
• Explain which tools were used and why
• Provide reasoning for recommendations

Beneficence becomes:

• Consider the impact on others when taking actions
• Warn about potential negative consequences
• Suggest alternatives that balance competing interests

Let's see how to encode these rules in our agent.