RAIUC03-BP03 Identify the type of AI required by your AI use case
Selecting the appropriate type of AI solution is a critical decision that fundamentally shapes your project's success and risk profile. Your choice of traditional ML, generative AI, or agentic AI must align with your specific use case requirements, data availability, and desired outcomes. The decision impacts everything from development complexity and resource requirements to explainability and risk management needs. A misaligned choice can lead to project failure, increased costs, or unmanageable risks, while the right selection creates a foundation for successful AI implementation that meets business objectives while maintaining appropriate controls.
Level of risk exposed if this best practice is not established: High
Implementation considerations
-
Determine if your use case primarily involves recognizing patterns in complex but pre-defined input data. If so, you may need traditional ML. Examples include fraud detection, demand forecasting, or quality control systems where patterns exist but are too complex for explicit rules.
-
Determine if your use case requires understanding widely varying inputs, including natural language, and creating new content or providing human-like responses. If so, you may need Generative AI. Examples include media creation, code generation, and advanced customer chatbots.
-
Determine if your use case requires breaking down high-level user objectives into workflows, and potentially reconfiguring the workflows depending on the results of intermediate tasks, as opposed to just responding to queries or making predictions. The use of a natural language interface for users to communicate these complex, high-level intents and receive updates is one of the primary characteristics of this approach. If so, you may need agentic AI. Examples include research and travel assistants.
Resources
Related documents:
-
ISO/IEC 42001:2023
A.6.2.2 AI system requirements and specification -
NIST Artificial Intelligence Risk Management Framework (NIST AI 100-1)
: MAP2.1, MAP2.2, MAP2.3