Accelerate Trusted AI Solutions with IBM Cloud Pak for Data and CognitiveScale Cortex
Today’s era of the digital economy offers opportunities for every industry to leverage AI for the best business outcomes. Enterprises can fuel business growth by implementing AI solutions that make their products and services more compelling for customers. AI adoption will require an end-to-end platform to build and deploy trusted AI solutions that can deliver hyper-personalized customer experiences at scale. This blog post outlines how IBM Cloud Pak for Data and CognitiveScale Cortex enables enterprises to accelerate the implementation of trusted AI solutions.
Cloud-Native Architecture for AI Solutions
Cloud-native architectures are driving innovations across the spectrum of applications, be it in AI, automation, transactional, or edge systems. Cloud-native applications implement distributed, loosely coupled microservices and are deployed as containers. They deliver the scale and resilience critical for enterprise workloads. Red Hat OpenShift offers a container platform for cloud-native applications. With OpenShift, enterprises can leverage the cloud-native benefits coupled with the flexibility to build and operate portable applications for on-premises, public, private or hybrid clouds. This flexibility offers unique value for a business to drive AI closer to where the data resides.
Data and AI Applications with IBM Cloud Pak for Data
IBM Cloud Pak for Data offers an integrated Data and AI platform to collect, organize and analyze data for infusing AI into business applications and services. The platform provides modular capabilities, built as containerized services on OpenShift. The modules are packaged as assemblies to give users the choice of deploying them as independent components.
Foundational Services for Enterprise AI
The complexity of data landscapes, affinity to a varied ecosystem of tools and technologies, and compliance requirements imposed by regulations are key challenges in driving enterprise AI. IBM Cloud Pak for Data includes essential services to simplify this complexity and address the enterprise challenges.
· Data Virtualization (DV) service helps construct a logical view of data sourced from multiple enterprise systems. The virtual view hides the complexity of deriving datasets from different data sources and enables SQL-based queries for easy analysis.
· Watson Knowledge Catalog (WKC) helps realize an enterprise Data Catalog that makes it easy to search, access, and govern data with configurable policies.
· DataStage (DS) and Spark-based IBM Analytics Engine (IAE) offer scalable processing for driving data transformation workflows.
· Watson Studio (WS) provides comprehensive services to develop, deploy and monitor AI models. Data Scientists can use the choice of tools, ML/DL frameworks, and language environments to build AI models. The mix of tooling includes code-centric Jupyter notebooks, low-code guided ML interface with SPSS visual modeler, and no-code automated ML with AutoAI. Models can be deployed to the runtime using a standard approach irrespective of which tool or framework, or language is used to develop the model. The deployed models can be monitored for multiple dimensions of trust, be it fairness, drift, explainability, or model performance.
· Watson Assistant (WA) helps build a conversational NLP interface into any application, service, device, or channel.
The platform complements the UI-driven user experience with APIs and Python SDKs to drive all of these services programmatically. The APIs enable the integration of the services with an ecosystem of partner platforms and custom applications.
AI Solutions with CognitiveScale Cortex Platform
CognitiveScale Cortex offers a cloud-native platform to develop and deploy trusted AI solutions. Cortex Fabric and Cortex Certifai are the core components. Implemented as a collection of containerized services on Kubernetes systems like Red Hat OpenShift, it can be deployed in on-premises, private cloud, or public cloud environments.
Accelerate AI Solution Development with Cortex Fabric
Cortex Fabric provides a low-code, high-level programming model for quickly assembling AI solutions. The Fabric uses configurable, reusable building blocks referred to as Skills, Profiles, and Agents to accelerate AI development. The Cortex constructs conform to the YAML-based open-source CAMEL (camelai.org) specification.
The Cortex Skills are computation building blocks that accept inputs and can execute code to build outputs. A Skill can execute multiple Actions. The Cortex Profile, aka ‘Profile-of-One,’ combines different types of information about an entity for a unified semantic view. The declared attributes extracted from data sources, the inferred attributes predicted by AI models, and the observed attributes captured from different events are consolidated to build a single Profile view of the entity. The Cortex Agents are composed using multiple Skills, Inputs, and Outputs to orchestrate an end-to-end AI solution.
Fabric Studio offers a visual workbench to compose Cortex Agents. A small team of core developers build the Cortex Skills using Cortex Python libraries and import them into Cortex Registry for broader reuse across Cortex Agents. A larger pool of solution developers can use Fabric Studio to develop and test the Agents to deploy AI solutions on the Cortex platform. The deployed Agents and Skills execute as containerized workloads on the Cortex platform. The short cycle of solution development helps enterprises drive faster adoption of AI.
Enhance Trust in AI Solutions with Cortex Certifai
Cortex Certifai generates a composite trust score, aka the Trust Index (ATX) for AI models. Certifai Scanner probes AI models for different elements of trust based on configurable scan definitions and an evaluation dataset. The Trust Index is a weighted score derived from explainability, fairness, robustness, and performance metrics evaluated on the AI model. The Certifai AI Risk Assessment Questionnaire and the Policy tool allow business users to configure policies and thresholds to assess risk and ensure compliance by AI solutions.
CognitiveScale AI Solutions on IBM Cloud Pak for Data
IBM Cloud Pak for Data is an extensible platform designed to integrate with an ecosystem of partner solutions. All platform services are instrumented with APIs to ease the integration. AI solutions in Cortex can leverage the curated data assets and trusted AI models in Cloud Pak for Data to drive trusted business outcomes.
IBM Cloud Pak for Data and CognitiveScale Cortex are cloud-native platforms built on Red Hat OpenShift. Both platforms can be deployed and managed from a shared OpenShift cluster. Customers can develop and deploy end-to-end, trusted AI solutions with API-led integration between Cloud Pak for Data and Cortex.
API-led Integration Pattern for AI Solutions
Developers build Cortex Skills to access Data assets and drive AI services in Cloud Pak for Data using its APIs. The Cortex Skills will be reusable AI building blocks to compose Cortex Agents and orchestrate AI solutions.
Here is a reference architecture of a chatbot-based AI solution for a healthcare insurance firm.
The Demographics, Claims, and Prescriptions information on members is hosted in different enterprise systems. Data Virtualization (DV) in Cloud Pak for Data creates a single logical view of the member information by combining data from each system. The virtual table is published in a WKC catalog. The AI model is trained on existing members to predict the cost of care for new members. The chatbot engages in intelligent, contextual conversations using the unified profile of members and trusted predictions from AI models. The following set of Cortex Skills implement the chatbot-based AI solution.
1. ‘Member Profile’ retrieves data from WKC to build ‘Profile of One’ for members.
2. ‘Train AI Model’ trains model on Member Profiles using AutoAI in Watson Studio.
3. ‘Deploy & Score’ deploy the trained model to the deployment runtime, enables model monitoring, and scores new Member Profiles.
4. ‘Build Trust Index’ retrieves metrics from the model monitors to build a Trusted Index (ATX) score.
5. ‘Conversation Agent’ uses Watson Assistant to power AI Chatbot conversations.
The API-led integration pattern can be used to implement trusted AI solutions for any industry use case.
Summary
Enterprises have to accelerate AI adoption to fuel business growth. IBM Cloud Pak for Data and CognitiveScale Cortex are built as cloud-native AI platforms to deliver resilience, scale, and flexibility critical to enterprise workloads. Businesses can only rely on AI solutions they can trust. The API-led integration between the platforms enables low-code solution development in CognitiveScale Cortex to leverage the trusted data and AI services in IBM Cloud Pak for Data. The integrated services help accelerate and automate trusted AI solutions for the enterprise.
