OpenAI has launched the company's Red Teaming Network, an initiative enlisting a contracted group of experts to enhance the company's capabilities in AI model risk assessment and mitigation strategies.
This photo illustration shows the ChatGPT logo at an office in Washington, DC, on March 15, 2023. - The company behind the ChatGPT app that churns out essays, poems or computing code on command released on March 14, 2023, a long-awaited update of its artificial intelligence (AI) technology that it said would be safer and more accurate than its predecessor.
OpenAI Red Teaming Network
OpenAI, the ChatGPT creator, said in a blog post: "We're announcing an open call for the OpenAI Red Teaming Network and invite domain experts interested in improving the safety of OpenAI's models to join our efforts. We are looking for experts from various fields to collaborate with us in rigorously evaluating and red teaming our AI model."
This move represents a transition from internal adversarial testing towards a collaborative approach with external experts. The term "expert" is defined broadly here, encompassing a diverse range of domain knowledge and lived experiences.
These experts play a crucial role in crafting domain-specific risk assessments and evaluating potentially harmful capabilities in new systems.
This initiative aims to deepen and broaden collaborations with external experts, a pivotal component of OpenAI's approach. It's viewed as a complementary measure to established governance practices, including third-party audits.
"Red teaming can catch (albeit not fix, necessarily) biases in models like OpenAI's DALL-E 2, which has been found to amplify stereotypes around race and sex, and prompts that can cause text-generating models, including models like ChatGPT and GPT-4, to ignore safety filters," OpenAI said.
The OpenAI Red Teaming Network comprises trusted and experienced experts who contribute to risk assessment and mitigation efforts on a broader scale. It differs from one-off engagements and selection processes before major model deployments.
Members will be called upon based on their expertise to contribute to red teaming at various model and product development stages. Time contributions will be tailored to each member, ranging from 5 to 10 hours a year.
Assessing AI Systems
Apart from OpenAI's specifically commissioned red teaming initiatives, members will have the chance to interact with one another on general practices and discoveries related to red teaming.
This method seeks to encourage a diverse and ongoing input flow, transforming red teaming into a more iterative undertaking.
The network complements additional cooperative AI safety avenues, including OpenAI's Researcher Access Program and evaluations conducted in open-source environments.
Assessing AI systems necessitates comprehending an extensive range of fields, varying viewpoints, and diverse life experiences. OpenAI extends an invitation to professionals globally, with an emphasis on diversity in geographic location and domain of expertise throughout the selection process.
Related Article: Is ChatGPT Becoming Dumber? New Study Claims AI Chatbot's Performance Is Deteriorating
