Infusing AI in a Cloud Journey

The AI “evolution” is moving us from the traditional, programmable IT, to an IT that begins to understand, and learn. Already many applications leverage natural human language interface. Eventually AI-infused IT will have pervasive domain knowledge and the ability to fully interact with humans using natural language. Rather than coding behavior into our IT systems, a logical rule at a time, we have the opportunity to build technology that learns the proper behavior largely on its own.

The Cloud Engagement Hub’s technical Points of View define application modernization and the supporting use cases which motivate such efforts. It outlines a technical approach for a modernization journey, including migration, containerizing and repackaging, to get applications and workloads running on different technologies and places. The purpose of making those technical investments is to deliver new capabilities faster, to deliver better business outcomes, by accelerating development and enriching the solutions. Among new enriching capabilities, arguably the most valuable is to infuse AI to existing solutions as they are modernized.

Our goal is to help our clients leverage AI to obtain better business outcomes, and our strategy is therefore to “Infuse AI” across systems and solutions. But what does this mean? To use an analogy, “What defines an Internet company?” Consider a traditional company, say a retailer. Add to it a Website, and voila, you have an Internet retailer, right? No. Just adding a Website presence does not transform a retailer into an Internet company. What characterizes Internet companies is how they organize their resources to leverage the Internet, including agile development with frequent upgrades, A/B testing, etc.

What defines an AI-infused company? We don’t know yet for sure what it all entails. But, we know that, as the above example of the Website, just deploying an AI model on an app is not enough to make an AI company. It may be a good starting point, but it is far from getting the benefits of AI. For a true AI infusion we need to organize our work and resources to leverage the power of AI. For example, this will include having a good strategy to acquire and organize data properly (that is, not fragmented in silos); to discover and leverage automation opportunities across the business; to empower specialized job roles (e.g., data scientists) to predict relevant trends, and more.

Why infuse AI on Journey to Cloud?

But why is this relevant to a Cloud Journey mission? In a MITSloan Global Executive Study , 85% of participants agree that they have an urgent need for an AI strategy, and most prioritize an “ Enterprises Journey to AI” as a strategic priority. 72% say that AI will deliver mainly revenue increases. AI infusion can improve decision making, increase automation, enhance the end-user experience, prevent failures, prevent fraud, keep compliance to regulations, etc. Business drivers could be, for example, enriching application solutions by infusing AI to:

1. Improve product quality by integrating better defect detection mechanism
2. Reduce operation costs by automating handling of simple customer complaints
3. Improve customer satisfaction by deploying chatbots to address user issues
4. Retain customers by integrating churn-prevention analytics.

To support the infusion of AI, as in the above examples, enterprises need to get the right data. We need an Information Architecture (IA) that supports Artificial Intelligence, as the mantra “There is no AI without IA”. Thus, we motivate the need to modernize solutions, as part of the Journey to Cloud, to obtain improved data access, which can then be ingested by AI models, to accelerate the ability of an enterprise to get business value. This approach is not just for infusing AI, but applies to other types of enrichment, for example, integrating Blockchain.

The Data Problems in AI

One of the common pain-points of organizations trying to extract AI insights is the effort involved in preparing and organizing their data so it can be consumed by the AI algorithms. Firms that have leveraged AI the most, particularly in the consumer space, have been very successful at properly organizing their data. But in most large enterprises, data is spread across a variety of environments, including public and private clouds, and traditional on-premises deployments, captive within traditional, siloed systems of record. Many organizations have moved data to a central accessible location, but often these efforts fail to deliver easy and controlled data access required to obtain AI insights. When that is the case, there are fractured views of the data, and it becomes difficult to obtain AI insights.

Data science is a team effort, so appropriate collaboration tools are required to coordinate access and actions. If an organization staff doesn’t know what data they have and how they are using it, the organization can be subject to regulatory non-compliance challenges. ​ If the data used to train AI models has unfair biases, and the resulting recommendations aren’t transparent and trusted, then AI won’t be embraced and won’t be used at scale. Companies embracing AI can have hundreds of experiments built using different tools and running in different environments. ​ They need to detect and proactively mitigate bias, to ensure that the performance of the models is fair given regulatory legal constraints.​ Data scientists should be able to explain how and why their models make recommendations, and they should be able to trace the lineage of the actual data that was used to build the models.

The AI Ladder

Enterprises need a data management strategy to provide flexible, organized access to all data, of every type, regardless of where it lives, and addresses the above concerns. A modernization effort would define and deploy an information architecture that provides an open, extensible foundation, with choice and flexibility, capable of communicating with other cloud platforms. IBM’s hybrid data management strategy to accelerate the Journey to AI is a prescriptive approach defined by a 4-step AI ladder: Collect, Organize, Analyze and Infuse.

1. Collect: Make the data simple and accessible at the right location, from any database or storage facility.
2. Organize: Ensure that data is trustworthy, complete and consistent at all stages of the information lifecycle: profile, cleanse, and catalog the data, provide protection and compliance, enable policy-driven visibility, detection, and reporting. ​
3. Analyze: Build, deploy, and manage AI models using integrated tools to explore and analyze both structured and unstructured data, and deploy them securely.
4. Infuse: achieve trust and transparency in model-recommended decisions, explain decisions, detect bias, etc., using provided solutions and services. ​

Applying the IBM AI Portfolio

Lets now review the IBM AI Portfolio of tools that support the above ladder to AI. Together, this set of tools helps AI practitioners to Organize, Build, Deploy, Catalog and Manage their AI data and models. Notice that these tools can now be deployed on multiple cloud platforms, including competitors’ public clouds. This is a good example of the value of modernization: by containerizing multiple traditional core offerings we are now able to deploy them in multiple Clouds. The IBM’s AI Watson portfolio, as shown in the figure below, allows to:

• Prepare the data. The Knowledge Catalog allows users to access, curate, categorize, and share assets, with controls that enforce access policies.
• Build models for making predictions with Studio.
• Deploy and run models in production with Machine Learning.
• Manage the models with trust and transparency, with OpenScale.

Watson Knowledge Catalog is the tool that a “Data Engineer” will use to discover, cleanse and prepare the data. It helps understand data quality, data lineage and distribution through data-profile visualizations, built-in charts and statistics. It helps discover and uncover data across multiple on-premises and cloud sources to unlock silo knowledge and catalog new data sources. The tool helps to govern access to the underlying data assets by using an active policy manager.

Watson Studio provides tools for data scientists to collaborate in building and training AI models, and to deploy applications in a hybrid environment to operationalize them. It provides visual and open-source tools to explore data, prepare it and develop models.

Watson Machine Learning is the runtime environment for training and deployment of AI models into production, using Apache Spark. Once a model is built and trained, it can be deployed, auto-retrained, and managed. Users can run experiments, provide model tuning and comparison tools to evaluate models across 100–1000’s of hyperparameter configurations.

Watson OpenScale provides visibility on how AI models are built, run, and managed. It monitors the models to track and measure outcomes and adapts and governs AI to changing business situations — for models built and running anywhere. Track performance of production AI and its impact on business goals, with actionable metrics in a single console. Maintain regulatory compliance by tracing and explaining AI decisions across workflows, and intelligently detect and correct bias to improve outcomes.

In closing, …

Finally, AI infusion should be given due consideration as part of any modernization journey. It should be considered as part of the business case for such a journey and should be implemented in parallel with other modernization activities. Business value comes from enriching the solutions, as they are modernized, with new capabilities, and arguably most valuable will come from infusing AI as part of the journey…