Last year, when Generative AI (GenAI) took the world by storm in print and social media, industry pundits and every news outlet talked about how human civilization was at an inflection point in the grand arc of perpetual growth and eternal prosperity!

Okay, maybe I overstated that a bit… But I’m sure you remember, like I do, when Gartner’s AI hype cycle clearly showed GenAI was peaking!  And rightfully so: The Q2 State of Generative AI report from Deloitte articulates business and data leaders who are first adopters had high expectations from GenAI to deliver value in the form of:

1. 56% improved efficiency and productivity
2. 32% reduced costs
3. 31% innovation and growth

These were leaders’ expectations. But the reality is quite different, because we live in a world of ‘fast adopters’ and ‘slow movers.’ In the survey, 65% of fast adopters were inclined to explore and experiment with GenAI to build new capabilities, whereas 32% of the surveyed slow movers had other operational priorities – such as keeping the lights on and reducing technical debt – before modernizing their organizations to be AI ready.

And that’s why the actual value derived from GenAI was only:

• 27% improved efficiency and productivity
• 18% reduced costs

Interestingly, one-third (35%) are realizing increased value from innovation and growth. This increase compared to expectations was likely due to investments in Talent Management and the emergence of skills-based hiring in cloud and AI/ML, with technology adoption from Google, Microsoft, and Amazon.

The survey also shows that 72% of respondents claimed an increase in trust with the use of GenAI. This amplifies the fact that AI-adept organizations are poised to scale their AI initiatives to capture business value through

• Innovation by investing in GenAI and other emerging technologies
• Improving operations by leveraging automation capabilities

Federal Agencies Benefit from an Emergent Approach

With the changing technology landscape, Federal agencies should adopt an emergent approach to Generative AI that consists of four dimensions:  Assess, Engage, Communicate, and Formulate. These steps, when worked iteratively and revisited periodically, helps federal data leaders adjust continuously to a rapidly evolving environment.

Using this approach with one of our federal customers, a reputed financial regulatory agency, our team at Citizant successfully developed AI intake request processes, an AI Use Case inventory, and Generative AI and Risk governance. We applied the four dimensions of the emergent approach to these three foundational initiatives as follows:

To benefit from the emergent approach of using Generative AI technology, the federal customers must establish a comprehensive understanding by diving deeper into AI risk challenges and drivers.

Understanding AI Risk Challenges & Drivers

AI risk challenges are generally classified as measurement, tolerance, prioritization, and categorization. The 36-page NIST AI Risk Management Framework (RMF) unravels the challenges and brings to the surface drivers that exacerbate AI risks with the adoption of emerging technologies. AI risks emerge because of uncertainties in AI systems – either due to their opaque nature or because of system performance in controlled environments. As advisors, we help CDOs understand the uncertainties in AI systems in controlled environments throughout the AI Lifecycle – design, data collection, build, validate, and deploy – by building measurement criteria with the help of various AI actors – model owner, model user, and model monitor.

The drivers of AI Risk Management are data inadequacies, accuracy/data drifts, and security/privacy. Organizations with siloed data ecosystems experience uphill battles to launch and complete AI projects and systems. Until an agency organizes and rationalizes its data ecosystem, federal CDOs report they experience “pilot purgatory” and find it hard to demonstrate tangible ROI from their AI investments.

It’s important to note that AI Risk Management is different and separate from traditional enterprise risk management. As a leader, it’s vital to manage GenAI based technology risks with a GenAI Risk Mitigation Roadmap and Operational Framework.

Strategies & Roadmap for GenAI Risk Mitigation

Federal data leaders can help establish effective AI Risk Management strategies such as:

• Engaging cross-functional experts early in any procurement cycle to assess procurement and contractual processes prior to award and adhering to NIST AI RMF best practices.
• Ensuring data assets are fit-for-purpose so downstream business and technical users can use the data to build and operationalize new AI capabilities. But these assets must be prioritized, managed, and safeguarded.
• Establishing prescriptive risk mitigation strategies – addressing AI risks in human bias, model extraction, and third-party AI technologies.

CDOs and CAIO’s can follow these steps as a general roadmap to GenAI Risk Mitigation:

1) Form a centralized GenAI Review Group
Initiate a coalition council on Generative AI with support from cross-functional stakeholders, each representing their business areas (risk management, privacy, cybersecurity, legal, procurement, IT, etc.) to define, develop, and drive Generative AI tech in the enterprise with an executive sponsorship from the OCDO.

2) Establish charters and roles/responsibilities of the GenAI Review Group
Clarify the purpose of the Generative AI Review group, including a charter that aligns with the mission and vision of the enterprise, but not limited to roles and responsibilities either.

3) Ensure diverse and cross-functional representation of SMEs
Invite a diverse group of stakeholders to stimulate diversity in perspectives, as the newly formed coalition group articulates the strategy and framework to address AI risks related to the future use of GenAI technology.

4) Define GenAI through AI Intake, AI Use Case Inventory, etc.
Ensure each stakeholder representative aligns with the definition of GenAI and evolved their understanding of AI challenges and drivers as described in the previous section. It is vital that each representative can contribute, using their expertise, to review and assess the guidelines for GenAI use cases. Establishing workstreams such as AI Intake, AI Use Case Inventory, and AI Risk Governance is a critical component of this step in the overall roadmap to GenAI Risk Mitigation.

5) Create an approval framework for GenAI use cases
Armed together with collective insights and set cadence from GenAI Review group, leaders must create an approval framework for examining GenAI use cases. These use cases, equipped with robust information on the inputs, transformation, and outputs, will position the GenAI Review Group to either approve or deny build and deployment.

6) Leverage existing agency Governance Councils for AI use case review and approval
Leaning upon existing governance bodies or councils will help accelerate the AI use case review and approval process.

7) Augment existing Enterprise Risk Management processes
Leveraging existing enterprise risk management processes can bolster risk posturing of new AI risks that need to have continuous monitoring throughout the AI Development Lifecycle – design, data collection, build, validate, and deploy. A heatmap that classifies GenAI use cases based on impact (critical, significant, moderate, and low) and their likelihood of occurrence (probable, likely, possible, and unlikely) will safeguard not only the enterprise deploying GenAI technology, but also cultivates trust among the end users of the technology.

8) Operationalize GenAI use cases
Operationalizing GenAI use cases with a firm grasp on AI Risk and Governance puts an organization and its leadership in the driver seat to take the enterprise into the next stage of the AI revolution.

In the “arms race” of GenAI, federal sector organizations can benefit immensely by exploring and adopting Generative AI in IT and Product Development, followed by Strategy & Operations, Finance, Marketing & Customer Experience, and Supply Chain.  In a future blog post, I will explore the prioritization of GenAI in these various use-case functions.