Responsible Generative AI Guidelines
Using Generative AI Responsibly at Thematic
In 2022, Thematic integrated Generative AI into our foundational AI models, with a focus on security, privacy, and accuracy to enhance the extraction of insights from feedback data. We maintain a rigorous approach to engineering our solutions, continuously evaluating and adopting the best Large Language Models to ensure responsible and efficient use of Generative AI. This guide provides an overview of best practices for using Generative AI at Thematic responsibly.
Understanding Generative AI
Generative AI, based on Large Language models (LLMs), encompasses algorithms that can create creative content. LLMs are proficient in processing complex relationships in natural language, generating text, and engaging in user conversations. These models rely on deep learning and neural network techniques and are trained on massive datasets, making them adept at natural language processing.
Best Practices for Responsible Use
1. Transparency: We prioritize transparency in our use of Generative AI. This involves clear communication with all stakeholders, including users and clients, regarding the nature and limitations of the system. Our summaries feature links to the original data, enabling users to easily verify results and gain valuable context. It is essential for users to check the original data for accuracy and relevance before using the summary to inform their decisions.
2. Human Oversight: Implement robust human oversight, such as your core user or Customer Success Manager, in your theme management and data upload process. When uncertain, involve your organization's champion or product specialist in Thematic to validate and interpret the AI-generated results, ensuring alignment with the context and nuances of the subject matter.
3. Feedback Loop: Help us audit and monitor the system for inaccuracies by providing feedback on AI-generated summaries. We have incorporated mechanisms within the product to facilitate easy feedback submission, allowing users to flag issues with our generative AI models and promote fair and accurate analyses through the thumbs-up or down button.
4. Privacy and Security: Safeguard sensitive data and adhere to stringent privacy and security protocols to protect the confidentiality of the information used on the platform. Users with permissions to manage access should ensure that users are only granted access to relevant data, in compliance with data protection regulations.