Skip to main content

Discoverability of AI Products

A true AI Product must be discoverable.
This means consumers — human, system, or agent — can find, evaluate, and access it without bespoke arrangements.

Discoverability is not optional.
It is one of the core characteristics inherited from BPS and is strengthened in AIPS to address the unique needs of AI.

Why Discoverability Matters​

  • Transparency → consumers know what exists, with what risks and purposes.
  • Efficiency → prevents duplication of effort and hidden “shadow AI.”
  • Governance → allows oversight and enforcement of policies across the product landscape.
  • Ecosystem growth → enables composability of AI Products into larger systems.

Requirements for AI Product Discoverability​

  1. Catalog Registration

    • Must be registered in a product catalog, marketplace, or registry.
    • Metadata includes: identity, owner, purpose, capability type, risk class, prohibited uses.

  2. Searchability

    • Consumers must be able to query by:
      • Purpose / intent.
      • Capability type (model, agent, hybrid).
      • Risk level and compliance status.
      • Performance metrics.

  3. Metadata Standards

    • Must use structured metadata based on BPS and extended by AIPS.
    • Includes links to model cards, system cards, explainability reports, and governance declarations.

  4. Interoperability

    • AI Products must support discovery via open standards (e.g., JSON-LD, RDF, OpenAPI, Schema.org extensions).
    • Ensures cross-platform portability across enterprises, ecosystems, and marketplaces.

  5. Access Pathways

    • Discoverability does not guarantee open access.
    • AI Products must clearly declare access policies (open, restricted, licensed).
    • Consumers should know how to request or provision access.

Beyond Discovery: Evaluation​

Discovery is only the first step.
An AI Product must also expose evaluation information at the discovery stage:

  • Performance benchmarks.
  • Fairness and bias metrics.
  • Resource footprint.
  • Supported deployment environments.

This allows consumers to assess fitness for purpose before integration.

Summary​

  • Discoverability is a non-negotiable trait of a true AI Product.
  • It requires catalog registration, searchable metadata, interoperability standards, and transparent evaluation data.
  • Without discoverability (see Glossary), an AI Product risks being treated as an AI Asset, not a product.

Principle: If it cannot be discovered via a trustworthy discovery mechanism, it cannot be trusted, governed, or reused.