IBM has announced, in his conference Think 2023the platform watsonxfocused on generative AI and data, and which will allow companies to have the necessary tools to train, tune and deploy AI models, including foundational ones. In addition, it will equip companies with machine learning capabilities. All this with trusted data, and with the right speed and governance, from a single point and ready to run in any cloud environment.
With watsonx, IBM offers a complete AI development studio, with access to foundational and open source models chosen and trained by IBM. It also gives access to a data warehouse to allow training and tuning data to be collected and cleaned; and a toolkit for AI governance.
In addition, it enables customers to meet the needs of their organization in five key areas of their business: interacting and conversing with customers and employees, automating business workflows and internal processes, automating IT processes, protecting against threats, and addressing sustainability goals.
Clients will have access to the necessary set of tools, technology, infrastructure and consulting expertise to develop their AI models, as well as fine-tune and adapt those already available. To do this they will use their own data, which they can deploy at scale in an open and trusted environment.
IBM watsonx.ai, watsonx.data and watsonx.governance
To do this, the IBM watsonx platform is made up of three sets of products: IBM watsonx.ai, IBM watsonx.data and IBM watsonx.governance. The first will be available starting in July, and includes what AI creators need to train, test, tune, and deploy traditional machine learning and new generative AI capabilities driven by foundational models. It has an open and intuitive user interface, which gives access to all its elements.
Among them are a range of foundational models, training and tuning tools, and cost-effective infrastructure, facilitating the entire data and AI lifecycle. From its preparation to the development, deployment and supervision of models. It also includes a library of foundational models chosen and trained by IBM.
These models use a broad set of curated business data, and are backed by a thorough filtering and cleansing process. They also have a data lineage that can be audited. In addition, they are trained in both language and other modalities, including code, time series data, and tabular, geospatial, and IT Incident data. In principle, it will have limited access to certain selected customers, and they will only be able to access a first set of foundational models in trial version.
The models that IBM will offer watsonx.ai are classified in several categories. Among them fm.code, fm.NLP and fm.geospatial. The first ones are created to generate code for developers automatically, through a natural language interface, with the aim of improving their productivity and allowing them to automate various tasks.
In fm.NLP there will be a collection of Large Language Models (LLMs) for specific industry domains using selected data, where bias can be more easily mitigated, that allow for rapid customization using customer data. Finally, fm.geospatial is a model based on climate and remote sensing data. Its mission is to help organizations understand and plan for changes in patterns of natural disasters, biodiversity, land use, and other geophysical processes that may affect their lines of business.
On the other hand, as part of a collaboration between IBM and HuggingFace, watsonx.ai will be based on the latter’s open source libraries, offering thousands of its models and open data sets.
The second of these categories, IBM watsonx.data, will also arrive in July, and will offer a data warehouse adapted and developed on the open architecture of the data lakehouse type. In addition, it will be optimized for AI governed data workloads, with support for queries, governance, and open data formats for data access and sharing.
This solution allows the management of workloads on premises and in multicloud environments. With IBM watsonx.data, a business can dramatically reduce the costs of storing data through workload optimization. In addition, it will provide a single entry point for user data access, while applying multiple search engines tailored to information discovery. It will have integrated governance, automation and integration tools with company databases and tools. This will reduce the complexity of the configuration, and improve the user experience.
IBM watsonx.governance, finally, is a set of governance tools for AI, focused on enabling trustworthy AI workflows. It will arrive before the end of 2023, and its mission will be to make governance operational and help mitigate the risk, time and cost associated with manual processes. In addition, it will offer the necessary documentation to achieve transparent and explainable results.
On the other hand, it will offer the necessary mechanisms to protect the privacy of customers, proactively detect bias and deviation from models, and help companies and organizations comply with their ethical standards.
Watsonx.ai will be in all IBM software products
In the future, IBM plans to integrate the foundational models of watsonx.ai into all of its major software products. Thus, it already plans to launch at the end of 2023 Watson Code Assistantwhich leverages generative AI to allow developers to generate code from a single command in English.
They are also developing AIOps Insights, that is, to make the capabilities of AIOps better with foundational models ready for the code and the natural language process. In addition, Watson Assistant and Watson Orchestrate will combine the IBM digital work products with a foundational natural language processing model to improve both productivity and the customer service experience.
Another novelty that will allow this aforementioned integration is the development of Environmental Intelligence Suite (EIS). If all goes according to plan, IBM EIS will feature the geospatial foundational model, and available for testing in late 2023.
