The Hybrid Intelligence of Knowledge-Based Bots: From Journalist Interactions to Therapy Assistants

Introduction

In 2010, I was getting contacted by lots and lots of journalists about Cyborg Anthropology and other concepts. Aaron Parecki and I discussed an idea for an automated assistant: a bot that could interact with journalists on my behalf. The system was designed to streamline the process of answering recurring questions while still allowing for human intervention when needed. Over time, it would grow into an increasingly comprehensive knowledge base, reducing the need for direct human involvement as its accuracy improved.

It's also how some knowledge base systems work, especially for domain-specific information like online web hosting platforms. These systems allow customers to ask a support question, and if the answer is in the database, it will show this set of potential answers, ranked by match percentage, in the search box that the customer uses. If there's nothing there, the customer can submit a ticket. A good knowledge base support system should add the finished result to the system after the ticket has been successfully finalized. Having a more automated system is possible, provided humans are placed in the loop in a robust way.

This concept aligns with the Centaur Model proposed by Garry Kasparov, in which a mediocre chess player assisted by a good system can outperform an expert player. Similarly, a well-structured knowledge system can enhance efficiency in various domains, from technical support to therapy, by integrating human expertise with automated assistance.

This article explores the original journalist bot concept, its application in broader knowledge-based systems, and a potential implementation for therapy bots that assist users between sessions while learning from real interactions.

The Journalist Bot: A Self-Improving Knowledge Base

Flowchart of the Journalist Bot Workflow. The system filters journalist questions based on confidence levels and updates its knowledge base over time.

Concept and Mechanism

The journalist bot functioned as a filtering and response system that categorized incoming questions based on confidence levels:

• 1. High Confidence (≥80%): If a journalist’s question closely matched an existing answer in the database, the bot would automatically send a slightly modified version of the stored response.

• 2. Low Confidence (<80%): If the system lacked sufficient certainty, it would notify me (by text, so I can quickly craft a response). That response would then go both to the journalist and stored in the database for future reference. The information would be used to generate slight modifications on the response so that every response was not perfect. In this way, the system would be able to grow over time in response to new information developed by me as well as requested by external querants.

This approach ensured both accuracy and efficiency, allowing repetitive questions to be handled automatically while unique or nuanced queries were addressed manually. Over time, the database would expand, covering a broader range of possible questions and reducing the burden on the human operator.

Expanding the Model to Other Knowledge Systems

The same principle applies to various knowledge-based domains, such as customer support in web hosting. Instead of immediately submitting a support ticket, users could interact with an AI-driven knowledge base that suggests potential answers based on prior issues. If no suitable match is found, a support ticket is generated, and the new issue is added to the knowledge system for future reference.

Such a looping system ensures that the database continually evolves, improving response accuracy and reducing the need for human intervention over time.

Flowchart of the Therapy Bot Workflow. The bot assists users by checking a therapy knowledge base and flagging uncertain queries for therapist review.

Applying This Model to Therapy Bots

The Role of AI in Therapy Assistance

Therapy is an area where knowledge-based systems could offer significant benefits. Many individuals seek structured ways to process emotions, reflect on therapy sessions, and reinforce therapeutic concepts between meetings with their therapist. A therapy bot, built using the principles of our journalist bot, could serve as a supplemental tool, providing continuity between sessions while ensuring sensitive or complex cases receive human intervention.

How a Therapy Bot Would Function

1. User Queries the Bot

• A person using the therapy bot asks a question or shares a thought.

2. Confidence Check

• If the system recognizes a closely related response from the therapist’s knowledge base (≥80% certainty), it provides a suggested reply.
• If the confidence is low, the system either flags the query for the therapist to review or saves it for discussion in the next session.

3. Therapist Involvement and Continuous Improvement

• Therapists review flagged queries and provide personalized responses.
• These responses are incorporated into the system, refining the bot’s ability to support future users.

Balancing Automation and Human Oversight

A therapy bot must strike a balance between automation and human judgment. Unlike technical support systems, therapy involves complex emotions and ethical considerations. The bot should not attempt to diagnose or offer critical mental health interventions but instead function as a Contextual Companion—reinforcing therapeutic strategies and capturing discussion points for future sessions.

Conclusion: The Future of Knowledge-Based AI

The journalist bot concept illustrates how a hybrid intelligence system can enhance efficiency while maintaining human oversight. When applied to therapy, such a system could serve as a valuable tool for reinforcing treatment strategies, helping clients process emotions, and ensuring that therapists have access to a structured history of client interactions.

By refining these systems, we can create a new class of AI assistants—ones that don't replace human expertise but enhance and extend it.