Watsonx is IBM’s next-generation AI platform for building and tuning foundation models, generative AI and machine learning systems. The platform contains a studio, data store and governance toolkit. Additionally, AI-enhanced features are embedded into watsonx products, such as code development, AIOps, digital labor, security and sustainability initiatives.
Watsonx also includes data governance tools to make information more accessible for regulators and third parties. This is important because of new European, Latin American, and US government legislation currently being formulated. Data governance helps make processes easier for stakeholders and allows clients to manage regulatory changes.
Generative AI is a relatively new technology, and many companies still lack the expertise needed to create, train and run the models. According to IBM, enterprise clients without a history of AI experience typically have many questions about generative AI, like how do they execute a model? How do they make the chatbot customer facing or employee facing. How do they scale it up. And, importantly, how do they have confidence that the outputs will be accurate?
To support those questions, IBM Consulting is creating a Center for Excellence to support generative AI efforts. IBM plans to staff the center with more than 1,000 AI specialists.
AI, data and governance
Let’s look at the platform’s three main building blocks: watsonx.ai, watsonx.data and watson.governance.
IBM watsonx.ai is an AI studio where enterprise scientists and developers can build, run and deploy AI based on machine learning and generative AI using a library of high-quality, domain-specific IBM foundation models. The studio also includes a prompt lab for experimenting with a large spectrum of prompts on various foundation models.
Forbes Daily: Get our best stories, exclusive reporting and essential analysis of the day’s news in your inbox every weekday.
IBM has also collaborated with Hugging Face to provide access to thousands of open models and datasets. Clients can choose the best models and architecture depending on their business needs.
When using watsonx.ai, developers can use the IBM ModelOps environment to build and manage machine learning models through all stages, from development to deployment. APIs, SDKs and libraries are also available for all levels of expertise. Advanced practitioners can use the IBM tuning studio to customize models with labeled data or to create new trusted models from a client’s pre-trained model.
Watsonx.ai studio – Foundation models are an essential part of AI, so it is important to highlight their availability in watsonx.ai. All watsonx.ai models are trained on IBM’s curated, enterprise-focused data lake using its custom-designed cloud-native AI supercomputer, Vela.
Foundation models are pre-trained, large-scale neural network models built using a massive amount of data. Foundation models are the building block for various downstream natural language processing (NLP) tasks. They can be fine-tuned or adapted for specific applications such as generative AI models. I expect that IBM will be offering both encoder and decoder models.
IBM watsonx.data is a fit-for-purpose data store built on open lakehouse architecture. Developers can use it on-premises or in the cloud, and it can optimize workloads to reduce data warehouse costs by up to 50%. Watsonx.data also provides a single point of entry for users to access their data and multiple query engines for different types of analysis. The solution includes built-in governance tools, automation and integrations with existing databases and tools to simplify setup and improve the user experience. The expected availability is July 2023.
IBM watsonx. governance is a solution that can help operationalize governance, reduce risk and costs associated with manual processes while documenting transparent and explainable outcomes. An AI governance toolkit is being developed that will enable trusted AI workflows. This component will also include mechanisms to protect customer privacy, proactively detect model bias and drift and help organizations meet ethical standards. The expected availability is sometime this year.
Why data is critical
Improper use of data to create models can result in significant problems.
To avoid downstream problems, IBM has focused on providing governance throughout the entire AI lifecycle of watsonx. Governance also helps manage and understand continuous model changes, including training data changes.
It is crucial to have trust and governance in data that is used to scale generative AI across an organization. You certainly would not want to scale a model with the potential to produce unpredictable outcomes. A model that generated false information or hallucinations would be detrimental to a business. It can cause a loss of trust by its users, ethical issues, or errors in operational decisions to name just a few of the problems.
An excellent example of how hallucinated information can cause serious issues is the recent case of an attorney who cited numerous legal cases as reasons for the judge to dismiss the case against his client. However, the full text of cases supplied by the lawyer was all fictitious. The judge discovered that the attorney had used ChatGPT to build his case, which had “hallucinated” and produced entirely fictional cases and arguments.
If a model distorted or hallucinated business information or pricing that badly in an enterprise environment, it could be disastrous.
To prevent these problems, IBM has leveraged its long experience with traditional AI to create and implement governance throughout the entire AI lifecycle of watsonx.
In addition to minimizing problems, governance also has operational benefits. It can monitor performance and track model and training data changes. Ongoing model accuracy is important because changes to training data can affect model performance or even create bias. It is essential to catch and fix these problems as it occurs to avoid negative financial impact or operational issues,
Watsonx provides businesses with an end-to-end AI platform that includes a complete AI workflow. It offers an AI development studio with access to both IBM-curated and open-source foundation models, a data store for managing training and tuning data and a governance toolkit to help businesses scale AI solutions responsibly. Watsonx.AI also provides tools for traditional AI, ML, and generative AI.
The extensive market disruption caused by OpenAI’s ChatGPT has led to many new AI companies being formed, possibly with immature and unvetted products.
By contrast, IBM is hardly a newcomer to AI. It has one of the world’s most respected AI research teams with a long history of successful AI research and AI product implementations.
IBM will be rolling out watson.governance sometime this year. No one should discount or minimize the importance of governance. Clean data will keep the model “honest” and free from creating false and biased outcomes.
If an enterprise plans to use generative AI it must have confidence in the model’s safety and its suitability to eventually be moved into a production environment. Companies can only gain the necessary confidence in models by having visibility into its data. That is the only way to ensure it is clean and compliant with company standards and any mandatory regulations.
IBM has differentiated itself in this space with its data governance capabilities.
From a support standpoint, IBM has deep experience working with clients in highly regulated spaces and the safe application of technology. Support is essential and especially valuable for a company with little experience creating generative AI models.
If you are interested in more information about IBM’s AI research, you can check out my previous articles, including “IBM CodeNet: Artificial Intelligence That Can Program Computers And Solve A $100 Billion Legacy Code Problem” and “IBM’s AutoAI Has The Smarts To Make Data Scientists A Lot More Productive.”
In mid-February 2023 I wrote an article about IBM’s extensive pharma research using foundation models and generative AI. IBM has also developed and deployed AI models focused on increasing the efficiency of its internal business units. Those efforts have accumulated thousands of hours of experience that were directly applicable to the creation of the watsonx suite.
Finally, to better understand the computing power behind watsonx.ai, check out my deep dive into the cloud-based AI supercomputer that IBM originally built to train and run its internal foundation and generative AI models. That platform, which has since been named Vela, is now running watsonx.ai.