Dan Mitchell in Information Week: Track, trace and govern: Don't overlook AI outputs

Read the article in Information Week here.

Unlike traditional, legacy data sets, AI-generated content and insights tend to live inside a vacuum, created, used and taken for granted without proper governance. Unfortunately, for companies that don't provide proper oversight -- and proactively govern AI data -- they're susceptible to unseen risks.

In other words, ungoverned AI data can poison the well. It makes companies vulnerable to legal or compliance issues, intellectual property concerns, holes in data sourcing and accountability, and inconsistent data results. 

At the same time, leaders of data management who understand the importance of governing AI-generated insights and data face the challenge to do so proactively, rather than continually working backward to fix or react to data issues.

Ungoverned AI: What can go wrong 

Rather than simply taking AI-synthesized data at face value and pulling it out of a system, companies need to ensure that all synthetic data and generative AI (GenAI)-powered insights are tagged, tracked, traced, stored and properly governed.

Enterprises can move too quickly, ingesting AI-powered data out of a system, saving it to a file share and then rolling it into their systems without tracing the history. Companies lacking proper AI data governance can face unexpected results, legal issues and decisions based on suspect sources. Items to keep an eye on can include:

• Regulatory frameworks such as the EU Artificial Intelligence Act.

• GenAI-developed infographics that cleverly use images owned by other organizations, resulting in the company paying royalties.

• Marketing copy generated for a campaign that might experience a hallucination, which causes the AI to directly borrow from a text or quote that can't be used, raising legal issues.

• Large synthetic data sets built with AI used to train models, which are then pushed into production. Companies that don't track who created that data, when and where might lose that foundational knowledge going forward, causing teams to re-create the data set repeatedly. 

Continually re-creating data sets through AI causes inconsistencies in the data, because each pull might be different. Constantly remaking large synthetic data sets -- only to have them disappear -- is like building and melting icebergs. AI-driven insights are incredibly helpful and convenient for business teams to leverage, but the process doesn't need to be reckless and wasteful.

AI output governance

Before insights are generated, enterprises need best practices in place to govern how AI data is used. This includes foundational steps such as tagging, tracing, storing and establishing accountability around AI data. Other key tactics include:

• Pull all data sources to the center. Companies need to centralize data sources -- AI-generated, internal data, external sources, etc. -- into the cloud, where it can be tagged, tracked and not get filtered to different locations outside the center.

• Eliminate silos. Different business teams may vary in how they use and create data, naturally causing them to work in silos. All teams need to work together from a single source of truth.

• Don't take AI for granted. Culturally, companies should impress upon business teams not to take AI for granted. Just because insights, content and images are easy to generate with AI, it doesn't mean governance steps should be overlooked.

• Be vigilant in how AI is tagged. Ensure users note AI outputs by which specific AI model was used and what version. Include the timestamp for when the AI was generated; which user is initiating a request; and what content is getting pulled (analysis, recommendations, summaries, content). Apply the results with confidence scores.

Cross-functional collaboration

Companies that deliver a tight data management system rely on total collaboration across the organization. IT and legal teams, compliance officers and every business unit must work together to develop guidelines that work for them and are easy to follow and protect the organization.

AI works fast, and users tend to leverage models for immediate satisfaction, but a lack of governance creates risk and compliance concerns. Tracing, tracking, storing and properly building AI data can improve the overall AI literacy of their systems and accelerate AI ROI by delivering dependable results and reducing redundant workflows.

Going forward, regulations around AI are expected to intensify. Companies tagging, monitoring and governing AI outputs now will build infrastructure that's able to navigate regulatory changes and be a scalable, profitable asset.