At IBM Think in February, IBM made several announcements around the expansion of Watson’s availability and capabilities, framing these announcements as the launch of “Watson Anywhere.” This piece is intended to provide guidance to data analysts, data scientists, and analytic professionals seeking to implement machine learning and artificial intelligence capabilities and evaluating the capabilities of IBM Watson’s AI and machine learning services for their data.
IBM declared that Watson is now available “anywhere” – both on-prem and in any cloud configuration, whether private, public, singular, multi-cloud, or a hybrid cloud environment. Data that needs to remain in place for privacy and security reasons can now have Watson microservices act on it where it resides. The obstacle of cloud vendor lock-in can be avoided by simply bringing the code to the data instead of vice versa. This ubiquity is made possible via a connector from IBM Cloud Private for Data that makes these services available via Kubernetes containers. New Watson services that will be available via this connector include Watson Assistant, IBM’s virtual assistant, and Watson OpenScale, an AI operation and automation platform.
Watson OpenScale is an environment for managing AI applications that puts IBM’s Trust and Transparency principles into practice around machine learning models. It builds trust in these models by providing explanations of how said models come to the conclusions that they do, permitting visibility into what’s seen as a “black box” by making their processes auditable and traceable. OpenScale also claims the ability to automatically identify and mitigate bias in models, suggesting new data for model retraining. Finally, OpenScale also provides monitoring capabilities of AI in production, validating ongoing model accuracy and health from a central management console.
Watson Assistant lets organizations build conversational bot interfaces into applications and devices. When interacting with end users, it can perform searches of relevant documentation, ask the user for further clarification, or redirect the user to a person for sufficiently complex queries. Its availability as part of Watson Anywhere permits organizations to implement and run virtual assistants in clouds outside of the IBM Cloud.
These new services join other Watson services currently available via the IBM Cloud Private for Data connector including Watson Studio and Watson Machine Learning, IBM’s programs for creating and deploying machine learning models. Additional Watson services being made available for Watson Anywhere later this year include Watson Knowledge Studio and Watson Natural Language Understanding.
In addition, IBM also announced IBM Business Automation with Watson, a future AI capability that will permit businesses to further automate existing work processes by analyzing patterns in workflows for commonly repeated tasks. Currently, this capability is available via limited early access; general availability is anticipated later in 2019.
Organizations seeking to analyze data “in place” have a new option with Watson services now accessible outside of the IBM Cloud. Data that must remain where it is for security and privacy reasons can now have Watson analytics processes brought to it via a secure container, whether that data resides on-prem or in any cloud, not just the IBM cloud. This opens the possibility of using Watson to enterprises in regulated industries like finance, government, and healthcare, as well as in departments where governance and auditability are core requirements, such as legal and HR.
With the IBM Cloud Private for Data connector enabling Watson Anywhere, companies now have a net-new reason to consider IBM products and services in their data workflow. While Amazon and Azure dominate the cloud market, Watson’s AI and machine learning tools are generally easier to use out of the box. For companies who have made significant commitments to other cloud providers, Watson Anywhere represents an opportunity to bring more user-friendly data services to their data residing in non-IBM clouds.
Companies concerned about the “explainability” of machine learning models, particularly in regulated industries or for governance purposes, should consider using Watson OpenScale to monitor models in production. Because OpenScale can provide visibility into how models behave and make decisions, concerns about “black box models” can be mitigated with the ability to automatically audit a model, trace a given iteration, and explain how the model determined its outcomes. This transparency boosts the ability for line of business and executive users to understand what the model is doing from a business perspective, and justify subsequent actions based on that model’s output. For a company to depend on data-driven models, those models need to prove themselves trustworthy partners to those driving the business, and explainability bridges the gap between the model math and the business initiatives.
Finally, companies planning for long-term model usage need to consider how they plan to support model monitoring and maintenance. Longevity is a concern for machine learning models in production. Model drift reflects changes that your company needs to be aware of. How do companies ensure that model performance and accuracy is maintained over the long haul? What parameters determine when a model requires retraining, or to be taken out of production? Consistent monitoring and maintenance of operationalized models is key to their ongoing dependability.