Improving AI System Behavior: A Commitment to Accessibility and Customization
OpenAI is dedicated to ensuring widespread access to and benefits from AI and AGI. To achieve these goals, there are three essential building blocks that need to be considered for AI system behavior.
1. Improving Default Behavior
OpenAI aims for its AI systems to be useful and respectful to as many users as possible without requiring extensive customization. To achieve this, they invest in research and engineering to reduce biases in how ChatGPT responds to different inputs. OpenAI acknowledges that there is room for improvement in addressing both obvious and subtle biases.
Feedback from users is vital in identifying areas for improvement, especially in addressing instances when ChatGPT either refuses outputs it shouldn’t or fails to refuse when it should. OpenAI is committed to enhancing default behavior by learning from user experiences.
2. Allowing User Customization
OpenAI believes that AI should be a customizable tool for individual users within certain limits defined by society. They are actively developing an upgrade to ChatGPT that allows users to easily personalize its behavior.
While customization is encouraged, OpenAI recognizes the challenge of striking the right balance. They acknowledge that enabling extreme customization can lead to the misuse of technology or the creation of AI systems that mindlessly amplify existing beliefs. Therefore, there will always be some bounds on system behavior to prevent malicious use and power concentration.
3. Incorporating Public Input
OpenAI emphasizes the importance of avoiding power concentration in decision-making regarding AI systems like ChatGPT. To achieve this, they believe in involving users and those affected by these systems in influencing the rules and defaults.
OpenAI understands that practical implementation is a challenge, but they strive to include as many perspectives as possible. To initiate this process, external input has been sought through red teaming for technology assessment. OpenAI has also started soliciting public input on AI in education and is exploring partnerships with external organizations to conduct third-party audits of their safety and policy efforts.
OpenAI is at the forefront of piloting efforts to obtain public input on system behavior, disclosure mechanisms, deployment policies, and other crucial aspects. By prioritizing public participation, OpenAI aims to ensure the responsible development and deployment of AI technology.