The gap map follows the three layers you need to build an AI system, as defined at the 2024 Columbia Convening on Openness in AI. Each layer is split into its own product categories, and every category gets its own score based on the identified gaps within.
Product / UX
Product design choices and UX components determine what the AI system can be used for, what data it can access to execute its functions, and are responsible for providing a secure, appealing, and reliable experience to users; as well as collecting data to support developers.
Model Components
Training data, model weights, and adaptation pipelines determine how underlying AI models respond to specific inputs, shaping the underlying statistical behaviors that can be translated into AI outputs and activities.
Infrastructure
Compute, deployment, and runtime systems determine where models run, how reliably they respond under real use, and what constraints govern their operation; as well as providing the foundations developers need to build, host, and monitor production systems.