Collibra adds AI governance to data management platform

Governing AI

Just as data needs to be governed, so too does AI.

When only teams of data scientists and other experts worked with their organization's data, there wasn't a pressing need for data governance frameworks. However, as regulations arose on the use of data and self-service BI platforms such as Tableau and Qlik emerged to enable business users to work with data, organizations put in place guidelines that allowed those business users to confidently work with data while simultaneously ensuring data privacy and the organization's regulatory compliance.

Until recently, AI was almost exclusively the domain of data scientists and other data experts.

Generative AI is changing that.

Large language models (LLMs) have vocabularies as extensive as any dictionary. Therefore, when integrated with data management and analytics platforms, they enable users to engage with data using true natural language rather than code. In addition, LLMs understand intent, can be trained to generate code and can automate processes.

As a result, just as self-service analytics tools enabled business users to analyze data, generative AI technology enables business users to take on tasks that previously could only be done by trained experts. That includes using AI models and applications as part of their normal workflow.

Meanwhile, more regulations related to AI are coming, with the advent of the European Union's AI Act the most immediate example.

Therefore, just as self-service analytics necessitated the advent of data governance frameworks, generative AI now necessitates development of AI governance measures.

Collibra developed its AI Governance platform to address that need, according to Felix Van de Maele, the vendor's co-founder and CEO.

He noted that after a year of hype about generative AI, many enterprises are now developing models for specific applications. However, if not done carefully, risks can arise including exposing sensitive data and running afoul of not only existing regulations but also those that might be coming.

"There's a lot a stake and there are a lot of risks if you don't do [AI] well," Van de Maele said. "With the hype, we're also seeing a lot of recognition of the need to do [AI] well, and governance is critical there."

In fact, the need for AI governance might be greater than the need for data governance, he continued. With analytics, there's always a human to validate results before people make decisions and act. But with AI, models make scores of predictions in milliseconds without explanation and are sometimes trained to take action on their own.

"With AI, the problems are the same [as analytics] but the consequences are potentially way worse," Van de Maele said.

Using Collibra's new AI Governance tool, customers will be able to do the following:

• Define and document AI projects with an intuitive interface that enables both business users as well as data experts to work with AI models and applications.
• Use the Collibra Data Catalog to link AI applications to the data that feeds them so that data quality can be checked on a consistent basis with Collibra Data Quality & Observability.
• Protect sensitive data such as personally identifiable information with data access policies established through integrations with Collibra Data Privacy and Collibra Protect.
• Ensure regulatory and legal compliance, including applying risk ratings to different project categories so reviews of controversial projects can be prioritized.
• Provide context to the data that feeds AI by giving the data details such as goals, business value and team member roles so that there's greater transparency about the data and collaboration becomes easier.
• Give visibility into the lifecycle of an AI model -- its lineage -- so that users can know whether the data used to train the model is of high quality and can be trusted.

Henschen noted that AI Governance rightly addresses data quality, data security and privacy protections. However, the risk ratings and project prioritization features are among those that stand out.

AI is absolutely in need of governance, and metadata management and governance platforms such as Collibra have been addressing this need for a long time. The flurry of interest in AI sparked by [the] GenAI craze has magnified the need.

"I'm seeing plenty of functionality I would expect," Henschen said. "The risk ratings and triage features look interesting. And I'm particularly impressed that they're providing a business lens on AI governance in the form of documenting use cases, goals, business value [and] team member roles."

While helping customers more safely and effectively develop and deploy AI models and applications is the aim of AI Governance, the impetus for developing the tool came from conversations with customers, according to Van de Maele.

He said many of his conversations with users over the past year have centered on AI. Organizations are recognizing that they have to build AI models and applications to remain competitive but do so without exposing sensitive data or violating regulations.

"Everybody recognizes that they have do AI -- that it's a foundational thing for their business," Van de Maele said. "But they also realize that they have to do it in the right way. There's a lot at stake."

Next steps

Once AI Governance is generally available, Collibra will continue to add capabilities as regulations change and as AI infrastructures evolve, according to Van de Maele.

Toward that end, the vendor plans to add more integrations with vendors that provide tools that make up the AI infrastructure.

"The AI infrastructure landscape is changing quickly so we have to keep up with changes and trends," Van de Maele said.

In addition, Collibra intends to add and improve features that enable different personas within organizations to work with AI. The vendor also plans to add depth to AI Governance's monitoring and observability capabilities.

Henschen, meanwhile, said that while there's a heavy emphasis within AI Governance on governing the data used to train models, there is not yet a lot that specifically addresses governance of the actual models.

AI Governance provides some such capabilities, he noted. But there's room for growth.

"As models multiply, growing complexity will introduce the need for additional governance capabilities that go beyond data [to the models themselves]," Henschen said. "I'm hoping [there's more of that] in their plans."

Eric Avidon is a senior news writer for TechTarget Editorial and a journalist with more than 25 years of experience. He covers analytics and data management.