Exploring the AI and data capabilities of watsonx

Successful implementation of artificial intelligence (AI) is contingent on an AI strategy that takes into account the following considerations:

1. Open: It’s based on the best open technologies available
2. Trusted: It’s responsible and governed
3. Targeted: It’s designed for the enterprise and targeted for business domains
4. Empowering: It’s designed for value creators, not just users

Designed with these elements in mind, watsonx is a new AI and data platform that empowers enterprises to scale and accelerate the impact of AI across the business by leveraging data wherever it resides. IBM software products are embedding watsonx capabilities across digital labor, IT automation, security, sustainability, and application modernization to help unlock new levels of business value for clients.

The watsonx platform has three components: watsonx.ai (now available), watsonx.data (now available) and watsonx.governance (expected availability in November). In this blog, I will cover:

1. What is watsonx.ai?
2. What capabilities are included in watsonx.ai?
3. What is watsonx.data?
4. What capabilities are included in watsonx.data?
5. How can you get started today?

IBM watsonx.ai is our enterprise-ready next-generation studio for AI builders, bringing together traditional machine learning (ML) and new generative AI capabilities powered by foundation models. With watsonx.ai, businesses can effectively train, validate, tune and deploy AI models with confidence and at scale across their enterprise. 

By supporting open-source frameworks and tools for code-based, automated and visual data science capabilities — all in a secure, trusted studio environment — we’re already seeing excitement from companies ready to use both foundation models and machine learning to accomplish key tasks.

“IBM’s launch of watsonx was an awakening, and it has inspired us to deliver unprecedented innovations for our clients.” Sean Im, CEO, Samsung SDS America

“In the field of generative AI and foundation models, watsonx is a platform that will enable us to meet our customers’ requirements in terms of optimization and security, while allowing them to benefit from the dynamism and innovations of the open-source community.” Romain Gaborit, CTO, Eviden, an ATOS business

“We’re looking at the potential usage of Large Language Models. There are huge possibilities including connecting your controls to your internal policies.” Marc Sabino Head of Innovation, MD Citi Internal Audit

To help our clients take advantage of AI, we built a family of foundation models of different sizes and architectures, and carefully selected open-source generative AI models. Each IBM-trained foundation model brings together cutting-edge innovations from IBM Research and the open research community. These models have been trained on IBM curated datasets that have been mined to remove hateful, abusing and profane text (HAP). 

With multiple families in plan, the first release is the Slate family of models, which represent an encoder-only architecture. These encoder-only architecture models are fast and effective for many enterprise NLP tasks, such as classifying customer feedback and extracting information from large documents. While they require task-specific labeled data for fine tuning, they also offer clients the best cost performance trade-off for non-generative use cases. These Slate models are fine-tuned via Jupyter notebooks and APIs.

To bridge the tuning gap, watsonx.ai offers a Prompt Lab, where users can interact with different prompts using prompt engineering on generative AI models for both zero-shot prompting and few-shot prompting. This allows users to accomplish different Natural Language Processing (NLP) functional tasks and take advantage of IBM vetted pre-trained open-source foundation models. Encoder-decoder and decoder-only large language models are available in the Prompt Lab today.

Capabilities within the Prompt Lab include:

• Summarize: Transform text with domain-specific content into personalized overviews and capture key points (e.g., sales conversation summaries, insurance coverage, meeting transcripts, contract information)
• Generate: Generate text content for a specific purpose, such as marketing campaigns, job descriptions, blogs or articles, and email drafting support.
• Extract: Analyze existing unstructured text content to surface insights in specialized domain areas, such as audit acceleration, SEC 10K fact extraction and user research findings.
• Classify: Read and classify written input with as few as zero examples, such as sorting of customer complaints, threat and vulnerability classification, sentiment analysis, and customer segmentation.
• Question & Answering: Based on a set of documents or dynamic content, create a question-answering feature grounded on product specific content, such as building a Q&A resource from a broad knowledge base to provide customer service assistance.

Our viewpoint is that a single foundation model will not be the best fit for the wide range of enterprise use cases. That’s why we’re initially releasing five open-source models as part of the Prompt Lab sourced from Hugging Face, which can also be authored by third parties.

The models being released in the Prompt Lab include:

1. mpt-instruct2 (7b – decoder only) — Supports Q&A and Generate tasks
2. flan-t5-xxl (11b – encoder/decoder) — Supports Q&A, Generate, Summarize, Classify tasks
3. mt0-xxl (13b – encoder/decoder) — Supports Q&A, Generate, Extract, Summarize, Classify tasks
4. flan-ul2 (20b – encoder/decoder) — supports Q&A, Generate, Extract, Summarize, Classify tasks
5. gpt-neox (20b – decoder only) — Supports Q&A and Generate tasks

Subsequent watsonx.ai releases will include capabilities for prompt tuning and fine-tuning models as part of our Tuning Studio, as well as access to a greater variety of IBM-trained proprietary foundation models for efficient domain and task specialization. 

Within watsonx.ai, users can take advantage of open-source frameworks like PyTorch, TensorFlow and scikit-learn alongside IBM’s entire machine learning and data science toolkit and its ecosystem tools for code-based and visual data science capabilities. Data scientists, data engineers, and developers can work with Jupyter notebooks and CLIs in programming languages they are familiar with, such as Python and R, to deploy the pre-trained machine learning model for various Natural Language Processing (NLP) use cases, including complaint analysis using tone or emotion classification, entity extraction on financial complaints, and sentiment model analysis.

Additional capabilities of our ML and data science toolkit include:

• MLOps pipelines: Offers a collaborative studio for data scientists to build, train and deploy machine learning models with advanced features like automated machine learning and model monitoring. Allows users to manage their models throughout the development and deployment lifecycle.
• Decision optimization: Provides the industry-leading solution engines for mathematical programming and constraint programming to solve your optimization use cases with a choice of notebook or visual programming interfaces.
• Visual modeling: Delivers easy-to-use workflows for data scientists to build data preparation and predictive machine learning pipelines that include text analytics, visualizations and a variety of modeling methods.
• Automated development: Automates data preparation, model development, feature engineering and hyperparameter optimization using AutoAI.

IBM watsonx.data is a fit-for-purpose data store built on an open lakehouse architecture. It is supported by querying, governance, and open data formats to access and share data across the hybrid cloud. Through workload optimization across multiple query engines and storage tiers, organizations can reduce data warehouse costs by up to 50 percent.¹ Watsonx.data offers built-in governance and automation to get to trusted insights within minutes, and integrations with existing databases and tools to simplify setup and user experience. Later this year, it will leverage watsonx.ai foundation models to help users discover, augment, and enrich data with natural language.

Whether optimizing data warehouse workloads with multi-engine support or modernizing data lakes with high performance, governance and security, we are already seeing excitement from customers using watsonx.data as a new data foundation to accelerate their AI and analytics initiatives.

AMC Networks is excited by the opportunity to capitalize on the value of all of their data to improve viewer experiences.

“Watsonx.data could allow us to easily access and analyze our expansive, distributed data to help extract actionable insights.” Vitaly Tsivin, EVP Business Intelligence at AMC Networks.

STL Digital (STLD), the strategic IT partner of the Vedanta group, a global natural resources company, sees the potential of watsonx in driving the organization’s digital transformation:

“The power of watsonx.ai models, combined with the ability to leverage governed data in watsonx.data, enables our teams to build, train, tune, and deploy custom models at scale.” Raman Venkatraman, CEO of STL Digital

Watsonx.data is truly open and interoperable. It uses not just open-source technologies, but those with open governance and broad and diverse communities of users and contributors, like Apache Iceberg and Presto which is hosted by the Linux Foundation. Watsonx.data is also engineered to use Intel’s built-in accelerators on Intel’s new 4th Gen Xeon Scalable Processors, and uses multiple open-source query engines such as Presto and Spark. This provides for a breadth of workload coverage ranging from data exploration and transformation to analytics, BI and AI model training and tuning.

“We look forward to partnering with IBM to optimize the watsonx.data stack and contributing to the open-source community.” Das Kamhout, VP and Senior Principal Engineer of the Cloud and Enterprise Solutions Group at Intel

Watsonx.data supports our customers’ increasing needs around hybrid cloud deployments and is available on premises and across multiple cloud providers, including IBM Cloud and Amazon Web Services (AWS). Integrations between watsonx.data and AWS solutions include Amazon S3, EMR Spark, and later this year AWS Glue, as well as many more to come.

“Making watsonx.data available as a service in AWS Marketplace supports our customers’ increasing needs around hybrid cloud.” Soo Lee, Worldwide Strategic Alliances Director at AWS

Integration with watsonx.data also enables existing IBM Db2 Warehouse and Netezza customers to achieve a unified view of their analytics and AI estate. The next generation of Db2 Warehouse SaaS and Netezza SaaS on AWS fully support open formats such as Parquet and Iceberg table format, enabling the seamless combination and sharing of data in watsonx.data without the need for duplication or additional ETL. Watsonx.data allows customers to augment data warehouses such as Db2 Warehouse and Netezza and optimize workloads for performance and cost. Moreover, watsonx.data simplifies the process of combining new data from various sources with existing mission-critical data residing in on-premises and cloud repositories to power new insights.

“Building on our already existing Netezza workloads… we’re excited to see how watsonx can help us drive predictive analytics, identify fraud and optimize our marketing.” Bahaa’ Awartany, Chief Data Officer, Capital Bank of Jordan

We are primarily seeing customer adoption of watsonx.data across 4 key use cases:

• AI/ML at scale: Build, train, tune, deploy, and monitor trusted AI/ML models for mission critical workloads with governed data in watsonx.data and ensure compliance with lineage and reproducibility of data used for AI. 
• Real-time analytics and BI: Combine data from existing sources with new data to unlock new, faster insights without the cost and complexity of duplicating and moving data across different environments. 
• Streamline data engineering: Reduce data pipelines, simplify data transformation, and enrich data for consumption using SQL, Python, or an AI infused conversational interface. 
• Responsible data sharing: Enable self-service access for more users to more data while ensuring security and compliance through centralized governance and local automated policy enforcement.

Our approach to an open data lakehouse architecture combines the best of IBM with the best of open source. Capabilities within watsonx.data include: 

• Multi-cloud, hybrid cloud availability: Supporting both SaaS and self-managed software deployment models, or a combination of both, providing another dimension of cost optimization.
• Presto engine: Incorporates the latest performance enhancements to the Presto (link resides outside ibm.com) query engine. Presto is an open-source, fast, reliable, and highly scalable SQL query engine and is contributed to by some of the biggest companies in the world including Meta, Uber, Intel, and more.
• Multi-engine integration: Eliminate the need to keep multiple copies of data for various workloads or across database and data lake repositories for analytics and AI use cases. Presto, Apache Spark, Db2, and Netezza engines are fully integrated with shared metadata and data storage and work off Iceberg table format to access and query a single copy of data across the multiple engines.
• Open data and table format support: Store vast amounts of data in vendor-agnostic open formats, such as Parquet, Avro, and Apache ORC, while leveraging Apache Iceberg table format to share large volumes of data through an open table format built for high performance analytics.
• Enterprise compliance and security: Protect data, manage compliance, and maintain trust with built in-governance, automation, and enterprise security capabilities, and fit seamlessly into a data fabric architecture with the Cloud Pak for Data and IBM Knowledge Catalog integration.

• Easy to use, integrated data console: Bring your own data and stay in control of your data. In a few clicks, users can connect to existing analytics environments and start deploying fit-for-purpose query engines with integrated metadata and storage through a single point of entry. Seamlessly connect watsonx.data with various object storage such as AWS S3 or IBM Cloud object storage and registered databases such as MongoDB, MySQL, PostgreSQL, and more.
• IBM Ecosystem integrations: Providing robust integration with IBM’s ecosystem to allow users to seamlessly realize the benefits of existing IBM investments and streamline the flow of data and information between products with seamless integration for IBM Db2 Warehouse, Netezza Performance Server, IBM zSystems, and Cognos Analytics, with DataStage, IBM Knowledge Catalog, Databand.ai, and Watson Studio integrations coming later this year.
• Insights powered by generative AI: Later this year, users will be able to use natural language to explore, augment, and enrich data from a conversational user interface.

Test out watsonx.ai and watsonx.data for yourself with our watsonx trial experience.

For watsonx.ai, our new AI studio to support both machine learning and generative AI use cases, anyone can take advantage of watsonx.ai for free. Within the watsonx.ai trial, you get access to features such as 50K inference tokens, per user, per month, to play around with different sample prompts in the Prompt Lab.

With our free watsonx.data trial, you’ll receive USD 1,500 in free IBM Cloud credits to test drive a watsonx.data instance. You will be able to experience core capabilities such our multiple engines, support for open formats, built-in governance, and querying.