Ventana Research Analyst Perspectives

It is a mark of the rapid, current pace of development in artificial intelligence (AI) that machine learning (ML) models, until recently considered state of the art, are now routinely being referred to by developers and vendors as “traditional.” Generative AI, and large language models (LLMs) in particular, have taken the AI world by storm in the past year, automating and accelerating the development of content, including text, digital images, audio and video, as well as computer programs and models. We are at the very early stages of identifying enterprise use cases for generative AI but expect adoption to grow rapidly and assert that through 2025, one-quarter of organizations will deploy generative AI embedded in one or more software applications. Vendors already well-established in AI and ML, such as IBM, are introducing products and services designed to help customers adopt reusable foundation models for generative AI alongside existing task-specific ML models.  

IBM introduced its Watson brand to the world in 2010 when the natural language processing system appeared on the Jeopardy! game show, providing much of the public with an introduction to the real-life application of AI. In fact, we assessed the early efforts in our analysis in 2012, and it was an early leader in what is referred to as generative AI. Watson demonstrated that IBM was at the cutting edge of AI research and development. The company took advantage of the positive associations, subsequently increasing investment in AI/ML and utilizing the Watson brand for numerous initiatives — most notably in healthcare as well as its AI/ML development products. The use of the Watson name spread like wildfire during the rest of the decade, diluting its potency as it was applied to a range of applications and chatbots addressing increasingly diverse use cases including cooking, teaching, fashion and weather forecasting. IBM has taken a more judicious approach to the use of the Watson name in recent years. In early 2022, it divested Watson Health (now known as Merative). Similarly, IBM has rebranded several Watson-branded products and now uses the brand for a narrow set of automated digital assistant productivity tools: Watson Assistant for voice agents and chatbots, Watson Code Assistant for code development, Watson Discovery for document understanding and content analysis, and Watson Orchestrate for process automation. The company’s focus on AI development continues, however. At its Think customer event in May, IBM announced the launch of watsonx, the company’s new platform to address the development, testing, training, tuning and deployment of AI applications utilizing both ML and foundation models, the latter of which power generative AI capabilities.

The demand for AI is strong. Nearly 9 in 10 participants in Ventana Research’s Analytics and Data Benchmark Research use or plan to adopt AI technology. Thousands of users and organizations have been experimenting with LLMs since the introduction of applications such as ChatGPT and Bard. As I previously noted, these applications are not without trust and privacy concerns, and the ability to trust the output of generative AI models will be critical to adoption by enterprises. IBM’s perspective is that although these applications are important, the bigger breakthrough is the foundation models that underpin them. Unlike ML models that are developed and trained with labeled data to address a specific task in a specific domain, foundation models are trained on a broad set of unlabeled data and can be used to address multiple tasks across multiple domains with a relatively small amount of fine-tuning. At Think, IBM articulated its perspective that foundation models will lower the barriers to enterprise adoption of AI by reducing labeling requirements and facilitating reuse. The company also described how the combination of foundation models with proprietary data and domain knowledge will enable organizations to rapidly fine-tune models and develop applications that are tailored to their specific requirements. Watsonx is the platform that IBM hopes organizations will choose as their preferred environment on which to do that and consists of three products designed to address the AI development life cycle, data storage processing, and AI governance.

Watsonx.ai provides a development environment running on the Red Hat OpenShift Container Platform for data scientists to train, validate, tune, and deploy foundation and ML models. Watsonx.ai provides access to open source models from Hugging Face as well as IBM-curated foundation models, along with tuning and prompt engineering capabilities in addition to functionality to address ML development and the MLOps life cycle. Watsonx.data is a data lakehouse environment available on IBM Cloud and Amazon Web Services, as well as on premises. It enables the storage of data in the Parquet format and provides support for Apache Iceberg for transactional consistency. Watsonx.data offers a choice of query engines with Apache Spark and Presto (complemented by IBM’s recent acquisition of Presto specialist Ahana), along with IBM’s own Db2 and Netezza. Watsonx.governance provides an environment to collaboratively manage, catalog and monitor AI models in the context of ethical concerns and regulatory requirements. It provides a suite of automated governance, risk and compliance tools and functionality for model monitoring and mitigating bias and drift, as well as explainability.

Watsonx is designed to be a comprehensive platform to support an organization’s strategic adoption of ML and foundation models, either standalone or in combination with IBM’s AI consulting services. Some of the functionality delivered in watsonx is new, particularly the foundation model capabilities. However, some is also available via other IBM products, such as Cloud Pak for Data, and it is not entirely clear if or how current users of IBM’s data processing, governance and AI development capabilities will access watsonx. Nevertheless, I recommend that any organization exploring the potential for generative AI include IBM’s watsonx in their evaluations. It is clear that IBM is taking a considered approach to the enterprise applicability of generative AI and can be a trusted partner for enterprises developing a strategy to take advantage of foundation models.

Regards,

Matt Aslett