Knime adds new AI governance measures to analytics suite
The vendor's latest update adds connectivity to AI providers to enable GenAI development as well as governance features to address the growing use of GenAI and its inherent risks.
The latest release of Knime's open source analytics platform aims to enable customers to govern the use of generative AI so they can safely and securely use the new technology to analyze data.
Launched on July 23, the Knime Analytics Platform, which includes the vendor's Business Hub for storing and accessing data products and AI models, now provides access to a wide variety of large language models (LLMs) and models for adding vector embeddings so users can integrate their data with generative AI.
In addition, version 5.3 includes improved retrieval-augmented generation (RAG) tooling for users to discover relevant data for developing their own domain-specific generative AI models applications.
Beyond the new capabilities for building generative AI tools, Knime added a GenAI Gateway that lets administrators centrally govern who within their organization can access LLMs and other tools for developing generative AI models as well as models and applications once they are complete.
Given the concerns many organizations have related to the use of generative AI, which include making decisions based on faulty outputs and accidentally exposing sensitive information, Knime's update is significant, according to Mike Leone, an analyst at TechTarget's Enterprise Strategy Group.
"Users -- in this case, data scientists -- require appropriate guardrails and assurances that what they're accessing or how they're relying on a response doesn't jeopardize themselves or the business," he said. "They see the potential productivity and efficiency gains, and they require a solution that can deliver the levels of trust needed to broaden and scale adoption."
Based in Zurich, Switzerland, with two offices in Germany and a U.S. base in Austin, Texas, Knime is an open source analytics vendor whose platform includes data science capabilities.
The vendor's December 2023 platform update included an improved version of its generative AI chatbot, a new user interface that made it easier to navigate its set of tools and support for Azure OpenAI Service from Microsoft. Before that, in April 2023, Knime released an extension of its Business Hub add security and simplicity to data science model deployments.
New capabilities
Generative AI has been the dominant trend in analytics and data management ever since OpenAI launched ChatGPT in November 2022, which marked significant improvement in LLM capabilities.
LLMs such as ChatGPT and Google Gemini now enable true natural language processing. When combined with an enterprise's proprietary data, they foster true natural language interactions with that data that let non-technical users engage with data and trained experts eliminate some of the time-consuming coding tasks that occupy much of their time.
Due to the complexity of analytics and data management platforms, analytics use within organizations has been stuck for decades. Only about a quarter of employees have the requisite skills to work with data. To work with data, coding skills and data literacy training were needed.
Generative AI changes that by enabling the true natural language interactions that NLP tools developed by vendors in attempts to make their tools easier to use did not.
As a result, many vendors have made generative AI a focal point of their product development. They began providing generative AI chatbots to let users query and analyze data in natural language and integrating with LLMs and other tools to create environments where users can build their own models and applications.
However, concerns over the accuracy of generative AI outputs and data security risks associated with exposing data to third-party tools such as LLMs have led many enterprises to hold back on deploying generative AI tools.
Knime's platform update aims to reduce the occurrence of negative consequences with its new governance capabilities so that more enterprises will have the confidence to develop and deploy generative AI tools.
Knime's analytics platform is comprised of a series of nodes. GenAI Gateway enables an enterprise's data administrators to centrally configure which nodes and generative AI providers can be used by individual employees. The result is an access control system that sets permission limits on the use of generative AI tools based on a user's role and improves governance and security to engender more confidence in the use of generative AI.
That centralized control, meanwhile, is a means to successfully keeping generative AI outputs and the data that feeds them secure, according to Leone.
"I really like the GenAI Gateway," Leone said. "IT departments continue to get pressured from the business to support GenAI usage in different ways. By centralizing control and management, IT is being set up for success in enabling trusted model and infrastructure accessibility."
Users -- in this case, data scientists -- require appropriate guardrails and assurances that what they're accessing or how they're relying on a response doesn't jeopardize themselves or the business.
Eric AvidonEric Avidon
Andy Thurai, an analyst at Constellation Research, noted that Knime isn't the only vendor adding generative AI guardrails.
Tech giants such as AWS, Google and IBM are all building generative AI governance into the data management and analytics tools as are many smaller vendors such as Collibra. But whether Knime is the first vendor or not to put limits on how different employees can access models and applications, the vendor's governance measures are important.
"Any governance for generative AI is welcome in the current market as many companies focus on building and deploying the models," Thurai said.
Michael Berthold, Knime's co-founder and CEO, noted that AI governance was a priority for the vendor long before the launch of ChatGPT and the resulting surge of interest in generative AI.
Generative AI has perhaps made AI governance more important than ever given that it enables non-technical workers to take on some of the tasks previously executed only by trained experts. However, Knime's analytics tools have enabled development of data science models requiring the protection of sensitive information for the past few years.
As a result, the vendor has been working in concert with customers to develop AI governance measures with the new controls included in version 5.3 of Knime's analytics platform just the latest additions, according to Berthold.
"[AI governance] just wasn't on everyone's radar the way it is today," he said. "Now that AI allows many more people to compromise their data -- to send it to any wild AI provider out there -- governance has quickly become everybody's priority. The original model governance topic was a mix of our own ideas on the topic as well as conversations with more forward-looking customers and users."
In addition to the new AI governance capabilities, Knime's analytics platform now includes the following:
Integrations with Microsoft Presidio and the Giskard testing platform to help protect sensitive data such as personally identifiable information when sharing data to external LLM providers.
Access to AI providers OpenAI, Azure OpenAI Service and Databricks along with access to additional LLMs, such as Meta's Llama 3, to enable use of the newest models in connection with proprietary data.
Connections to Hugging Face's Embedding Inference to get access to open source embedding models and Inference Endpoints to power generative AI models for experimentation.
The ability to split text documents for RAG in a single step so only the text data relevant to a given model is moved into a model training pipeline.
While Thurai said that Knime is far from the only vendor offering generative AI governance capabilities, he noted that the combination of those governance measures with a data science platform to develop and deploy models that includes an array of connectors to LLMs and other sources is somewhat distinct from what other vendors provide.
"One thing that is different about Knime is that they are offering more than just governance," he said.
Next steps
With the latest version of Knime's analytics platform now generally available, AI governance will continue to be a focal point for the vendor, according to Berthold.
For example, Knime might add more access-control measures and additional integrations such as those with Presidio and Giskard.
"We'll continue to add model and AI governance features -- some developed ourselves and some with integrations," Berthold said. "We believe AI governance will require a more fine-grained governance strategy [such as] different model quality and security standards for different teams."
In addition, support for generative AI models and applications trained with unstructured data types in addition to text are part of Knime's roadmap, Berthold continued.
Leone, meanwhile, said that Knime's focus on AI governance is prudent. By adding more governance measures in concert with other developments that simplify using Knime's analytics tools, it will make the vendor's platform safer for a broad array of users within an enterprise.
"I think continuing to expand access of their platform to data generalists through effective governance and guardrails, intuitive interfaces, simplified processes and integrated automation will go a long way for Knime," Leone said.
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.
