About Human-Agent interaction
There are many scenarios where AI agents and human users can collaborate. However, building a robust automation solution that limits these interactions – while preserving human control over critical decisions – makes the AI agent more efficient. The goal is to achieve a balance where AI agents operate autonomously to maximize efficiency, but humans remain in the loop to provide expertise and make final judgments when needed. These interaction scenarios can be roughly divided into three main categories, and in this piece, we take a look at what they are.
In business process automation, AI agents should always be created for a specific purpose, such as a business process or even an individual task. To prevent potential errors or unintended consequences, specialist AI agents operate within predefined guidelines and business rules, outlined in their runbook. The runbook specifies the agent’s allowed actions, as well as areas where it must consult with human users before proceeding.
This is great way to ensure AI agents remain within their intended scope and that humans maintain ultimate control over critical decisions.
1. User-Initiated Interactions: Triggering and instructing the AI Agent
In this scenario, the user takes the initiative and instructs the AI agent to perform a specific task. For instance, when creating a financial report, the user – perhaps a financial analyst – requests the AI agent to create the report. This spares the analyst from manually compiling the data and creating the report from scratch.
This interaction model is straightforward: the user issues a command, and the AI agent executes it. However, it might not always be crystal clear what the capabilities of an agent are. In such cases, users can ask the agent about its purpose and functionalities. This allows the user to understand how the agent can be useful and utilize it effectively for different tasks.
2. Agent-Initiated Interactions: Human validation/guidance needed
Whilst working on their tasks, AI agents might encounter situations where they require human input or validation. This typically occurs when a task demands human judgment or when the agent lacks access to certain information, such as unmapped tacit knowledge held only by human experts – often the case with niche exception scenarios.
For example, in loan application processing, an AI agent can encounter a loan application requiring further verification or that has unclear information. Instead of using its best guess and proceeding, the agent will instead reach out to a human loan officer, providing the relevant details and highlighting the specific issue. The loan officer, with their expertise, can then make a decision or provide the necessary information for the AI agent to continue processing the application.
This way, the AI agent can handle the routine, repetitive aspects of the process, while humans step in to address complex or ambiguous scenarios that demand judgment and specialized knowledge that has not yet been given to the agent. This will increase the accuracy and reliability of the overall process.
3. Agent-Initiated Interactions: Keeping Humans Informed
Transparency and accountability are vital for building trust in AI systems. It’s very important that AI agents are proactively informing users about task outcomes and providing comprehensive reports.
For instance, in inventory management, an AI agent responsible for the inventory would notify the user prior to, and upon placing, a restocking order. The notification would include details like the items ordered, quantities, expected delivery dates, and any encountered issues.
Beyond simply reporting on their tasks, AI agents can also analyze how the overall execution of a business process or a task has been going, and provide insights on how to improve further.
For example, an AI agent managing customer support tickets could generate a weekly report summarizing the number of resolved tickets, average response times, and commonly encountered issues. With the AI Agents’ data-driven insights, humans can make informed decisions and identify areas for improvement.
Share
More resources