Article reader Listen + reading controls
Article reader
Preparing the reader…
Reading settings
Responsible AI needs distributed ownership¶
Microsoft's chief responsible artificial intelligence officer has published a reflection on the company's program, emphasizing leadership commitment, inclusive governance, and actionable standards. The timing is useful. Generative systems are moving into products rapidly, and many organizations are discovering that an ethics statement does not tell a product team what to do on Tuesday afternoon.
Responsible artificial intelligence (AI) needs a central function. It cannot remain the central function's job alone.
Central governance sees across the enterprise¶
A dedicated responsible-AI group can define principles, set minimum requirements, convene specialists, and identify patterns that no single product team can see. It can maintain standards, tools, training, and escalation paths. It can also provide a degree of independence when schedule or revenue pressure narrows a local team's attention.
That enterprise view matters. The same failure may appear in different products under different names. Without a central learning mechanism, every team treats it as novel.
But centralization has limits. A governance office does not possess all the operational knowledge needed to assess a use. It may not understand the decision tempo, the informal workarounds, the quality of source data, or the consequences of a false negative in the field.
Local ownership connects governance to work¶
The people building and operating a system control many of the choices that determine risk: task definition, data selection, interface, model configuration, user training, and response to failure. They need both authority and accountability.
This is why a federated model is stronger than either extreme. The center establishes non-negotiable expectations and supplies expertise. Product and mission teams own the evidence for their systems. Riskier uses receive deeper independent review. Lessons flow back into the shared standard.
Research on high-reliability organizations helps explain the design. Reliable organizations combine clear priorities with deference to expertise: decisions migrate toward the people with the most relevant knowledge when conditions become complex. Responsible-AI governance should do the same without making accountability migrate so far that nobody owns the outcome.
Name more than a committee¶
Governance charts often show a council at the top and review boards below it. The harder question is what happens between meetings. Each consequential AI system should have named roles for:
- the mission or business outcome;
- product behavior and user experience;
- technical performance and change control;
- data quality and provenance;
- security, privacy, and safety risk;
- operational monitoring and incident response; and
- acceptance of residual risk.
One person may fill several roles in a small organization. What matters is that the responsibilities are visible and that conflicts have an escalation path.
The National Institute of Standards and Technology (NIST) Artificial Intelligence Risk Management Framework places Govern across the other risk functions for good reason. Governance is not a stage completed before mapping and measurement. It is the arrangement that determines who performs those activities, how evidence is challenged, and what happens when it is inadequate.
Let exceptions improve the standard¶
No enterprise standard will fit every context. Teams will need exceptions. The weak response is to treat an exception as a private waiver. The stronger response records the rationale, conditions, owner, review date, and what the exception teaches about the standard.
If similar exceptions recur, the policy may be unrealistic, the tool may be missing, or the organization's use cases may have changed. Exceptions are data about the operating model.
Distributed ownership does not mean distributing liability until it disappears. It means placing decisions close enough to expertise to be informed, while keeping them connected to enterprise standards, independent challenge, and institutional learning.
That is what turns responsible AI from a specialized program into a capability the organization can exercise at scale.
Sources and research trail¶
- Microsoft, “Reflecting on Our Responsible AI Program: Three Critical Elements for Progress” (May 1, 2023).
- Microsoft, Responsible AI Standard, v2 (2022).
- National Institute of Standards and Technology, Artificial Intelligence Risk Management Framework (AI RMF 1.0) (2023).
- Weick and Sutcliffe, Managing the Unexpected (2015).
- Rakova et al., “Where Responsible AI Meets Reality” (2021).