Part of:The importance of enterprise AI governance
Use an AI governance framework to surmount challenges
As AI governance adapts to the rapidly expanding field of AI, businesses need a holistic framework to surmount challenges with clearly defined roles and responsibilities.
A comprehensive framework that combines business, technology and governance resources is needed to overcome common AI governance challenges.
As a practice, AI governance is relatively new, and like any other emerging field it faces certain challenges on its road to maturity. These challenges can be viewed through the popular rubric of people, process and technology (PPT).
Challenges of AI governance
The hype around AI makes you think otherwise, but AI is still in its early stages of enterprise adoption. So, governance challenges reflect that.
Lack of governance expertise. The first of these challenges is limited expertise. The need for AI governance and how it affects the success of AI projects is still not widely understood. This limited awareness further translates into limited availability of governance experts. Lack of governance expertise is not unique to end-user organizations. Different players in the ecosystem, such as product vendors and consulting companies, are in the same boat.
Ad hoc process. The second is that even among organizations that recognize the need for AI governance, it is done in an ad hoc manner that lacks a cross-functional approach. By default, technology teams have been leading AI projects. But the main focus of technical teams should be on AI models and model risk management, while other teams focus on other aspects of these projects.
Emerging tool sets. Thirdly, tools intended to streamline AI governance practices are emerging now, but they're difficult to directly incorporate into existing enterprise data science and machine learning workflows. Their emergence is partially good news, because tools usually bake in best practices, but challenges remain.
Overcoming challenges using a holistic framework
AI governance is a team sport where business, technology and governance experts all have their key responsibility areas (KRAs) (Figure 1).
This depicts an enterprise AI governance framework.
A holistic view of AI governance with clearly defined roles and responsibilities ushers in better accountability and transparency. Business goals should drive the enterprise AI strategy. Business goals also determine the investments and budgets for AI projects. The corporate social responsibility (CSR) and environmental, social and governance policies of an organization also provide inputs into the AI strategy.
The governance layer is often missing in organizations today, but it's a crucial interface between business and technical teams. Compared to traditional IT, AI projects were previously lacking in well-established standard methodologies. Governance can now establish the AI standards, methodologies and benchmarks, and the responsibility for performance metrics lies with the technical teams.
An AI governance framework also provides oversight and supervision while entrusting the technical teams with core technical aspects, such as data engineering, building models, choosing best-fit AI platforms and implementing MLOps tools. When the link between business goals and AI strategy is established, build versus buy decisions also become easier. For example, in the early days of AI, many organizations built their own tools for end-to-end machine learning pipelines, but with the maturity of MLOps products, buying is a viable alternative.
This framework helps ensure that a technical team focuses on technical risk, while a governance team looks at the implications and outcomes of AI for users.
As AI regulations loom, the governance team has a responsibility to keep track of and identify the requirements for compliance. The team helps liaise with any required regulatory authorities and provides the regulators with any required data and information. This team also draws up a risk management plan and may even serve as an internal AI audit function or facilitate external AI audits.
Risk must be managed in all stages of the AI lifecycle: training data, model development, post-deployment performance, impact and outcomes. This framework helps ensure that a technical team focuses on technical risk, while a governance team looks at the implications and outcomes of AI for users.
In many organizations, the technical team has been trying to perform all these functions by itself. Strong technical capabilities are necessary, but more elements are required for a successful AI governance framework. Technical expertise needs to be complemented with effective governance capabilities at the enterprise level.
Most organizations already have risk and governance experts who can be trained and engaged for AI governance. Having effective AI governance means a business has a strong foundation to scale, protect from risk and improve ROI on AI investments.
