IBM has announced a suite of new AI models for its watsonx enterprise platform including a new series of models called Granite that deploy generative AI to solve language and code tasks. Big Blue also unveiled a technical preview of watsonx.governance, a new tool for documenting and tracking the capabilities of foundation models.
This is the latest artificial intelligence move from the tech giant, building on the data and AI platform watsonx. It aims to produce models that can be deployed safely and securely by enterprise at scale and has recently added third-party open source models such as Meta’s Llama 2 to its platform. This will allow companies to build products on that model using their own data rather than rely on only using the IBM-built models.
The new Granite series of models use the same Decoder architecture that underpins many of the large language models used today and has been built to support enterprise-focused natural language processing tasks. This includes content generation, insight extraction and summarisation. It will begin rolling out later this month, the company said.
Unlike some other commercial models IBM says it will also provide a list of all the sources of data, a description of the data processing and filtering steps used in training the models. This will help with data governance and in being compliant with potential future legislation such as the EU AI Act.
IBM says training of its foundation models is focused on trust and transparency, including a rigorous data collection process, documenting and labelling sources. It includes responsible deployment of the models and applications of governance, risk assessment, bias mitigation and privacy concerns to make it easier for enterprise-grade compliance.
IBM watsonx documentation and traceability
A new governance toolkit is coming later in the year or early next year for IBM clients utilising watsonx AI capabilities. It automates the process of collection and documentation of those foundation models to provide compliance reporting on its capabilities and workflows. This ensures “humans are engaged at the right times” in the AI process.
“As demonstrated by the ongoing rollout of the watsonx platform within just a few months since launch, we are here to support clients through the entire AI lifecycle” said Dinesh Nirmal, senior vice president for products at IBM Software. “As a transformation partner, IBM is collaborating with clients to help them scale AI in a trustworthy way – from helping to institute foundational elements of their data strategies to tuning models for their specific business uses cases to helping them govern models beyond that.”
One of the biggest challenges with generative AI is appropriate data, particularly around ensuring a broad spectrum of input. Companies are turning to synthetic data to fill this gap and IBM says it is launching a new synthetic data generator into watsonx which will help clients create artificial tabular data sets from current data schemas. “This will allow users to extract insights for AI model training with reduced risk, thereby augmenting decision making and accelerating time to market,” an IBM spokesperson explained.
The company is also bringing generative AI to its hybrid cloud software and infrastructure products to improve automation. Specifically Instana and AIOps Insights will include Intelligent Remediation, which embeds watsonx.ai generative AI foundation models to assist IT Ops specialists to create summaries of incident details and provide a prescriptive workflow for engineers to make repairs.
Read more: IBM turns to AI for mainframe modernisation
