Building an AI governance strategy that works

As the use of AI becomes more and more ubiquitous across industries, becoming as widespread as electricity and clean drinking water, the conversation around the new technology is beginning to move from how to implement AI to how to implement it responsibly. How does AI differ from other software technologies that we have been using to build products, and is there a need for new regulations and new compliance frameworks?

That conversation hasn’t yet spurred most organizations to action. In a recent survey by Wakefield Research and Juniper Networks, 63% of companies said that they are at least most of the way to their planned AI adoption goals — but only 9% have fully mature governance policies.

Any organization that either uses or develops AI technology should start to put more focus on its AI governance practices. Otherwise, they risk being surprised by AI legislation now being developed while also putting their business and customers at risk due to improperly developed AI.

The necessity of governance

Every new technology brings about new questions about proper use and governance, and AI is no exception. However, because AI solutions are designed to execute tasks on par with human domain experts, proper governance of tasks that typically require human cognitive reasoning can be particularly difficult.

For example, AI in self-driving cars could be trusted to make split-second decisions, without human intervention, that have massive implications on the lives of people both inside and outside the car. Determining how AI should and should not be used, and what safeguards can protect users, bystanders and the manufacturer are important questions that are fundamentally different from previous types of governance.

Business leaders, customers and regulators are also beginning to ask questions about organizations’ use of AI and data, from how it’s developed to how it’s governed. If AI is making determinations based on data, how do I know if the model was trained with data that the trainer had the right to use? How do I know what that data is being used for?

Government leaders are already starting to turn their eyes toward regulation. The AI Act proposed in the European Union would have wide-reaching implications on how businesses can and should use AI. Even the U.S. is beginning to issue guidelines around the use of AI. In early October, the White House Office of Science and Technology Policy released its long-awaited “Blueprint for an AI Bill of Rights.” These are guidelines for companies to use in the development and deployment of AI systems. While not yet true mandates, businesses need to be prepared for that to change.

Organizations of all kinds need to prioritize governance at the same rate as adoption to safeguard their business, their technology and their customers.

AI governance strategies

Businesses need to develop a governance structure designed to mitigate both current and future risks as local, state and federal governments start to draft AI governance legislation.

Steps organizations can take today include:

• Clarify how and when AI will be used within the organization, both for current and future planned uses. AI is still a vague term that many organizations use to describe wildly different technology. Fundamentally, any technology that completes a task that previously required human cognition or reasoning can be broadly understood as AI. Organizations need to have a clear understanding of where and how their organization is using AI, regardless of whether that AI was purchased or developed in-house.

• Develop consistent standards and ethics across the enterprise that include safeguards around ethical uses of AI. Every major organization in the world already has standards, practices and safeguards around the technology they use. AI is a new technology, and while similar to other software technologies, it has some differences. Since AI solutions tend to do tasks on par with human domain experts, so all the same rules, ethics and liabilities that apply to bad human behavior tend to apply to AI. The behavior of an AI solution tends to change over time as models are retrained or the environment around it changes. After organizations have audited their use of AI, they should also audit their standards and ethics statements, both public and those used to train employees, and ensure they are comprehensive enough to include all of the organization’s uses of AI as well.

• Ensure that all governance policies are cross-functional and cover the entire AI ecosystem, including external ones. After you have a clear understanding of how AI is used in your organization and what your standards are for ethical uses of AI, it’s important to be sure that any external AI in use in falls under those same standards. If you develop best practices, but purchase AI-enabled solutions from a vendor that doesn’t follow the same governance rules, you leave your business and your customers vulnerable.

• Innovate with governance in mind. The best AI solutions aren’t worth creating if they can’t be responsibly integrated into existing governance policies. Once AI governance policies are clearly enshrined and established, ensuring that teams are trained on them and that alignment with goals remains the top priority is critical to long-term success.

Conclusion

AI isn’t just the future anymore; it’s very much the present reality for many organizations. AI governance, however, has still lagged behind AI adoption, but it’s an essential part of avoiding potholes on the path to AI success. IT teams that are staying ahead of the curve will play a crucial role in setting the future strategy within their organizations and their industries by leading in setting standards for AI.

Bob Friday is chief AI officer at Juniper Networks.