One Size Doesn’t Fit All.

Tailoring your AI Journey to meet your needs and timeframe.

When I look at my life’s journey, there are some common steps along the way that many others (not all) experience; the early years, going to high school, graduating, going to university, starting a career, raising a family, and so on. But when I look more closely at my own path and decisions, I realize just how unique my life has been compared to others – even other family members.  Of course, I’m happy to share my life experiences with others if it can help in any way.  While there may be groups of people that share common steps along their journeys, as individuals, we each find our way at our own time and speed. I have come to accept there is no set formula to life. Similarly, when we shop for clothes, we try them on to see if they fit correctly – one size doesn’t fit all, whether they look good on us, if they are practical for what we want to use them for, their quality, value / price, etc.

When it comes to implementing AI projects, having interacted with many organizations over the years, I have learnt that each of their journeys is also unique.  Again, there may be common steps along the way – but one size does not fit all. 

Nonetheless, organizations seek advice on how best to succeed when it comes to implementing AI, and to help meet that need, IBM came up with the “AI Ladder”.

Figure #1: A graphical representation of The AI Ladder

Think of the AI ladder as a set of steps – four to be precise – that help put in place a set of capabilities designed to help prepare and propel an organization along their unique AI journey. 

The steps are as follows:

Collect data.

Quite often an organization may not even know all the information they have. Part of this step can involve the discovery of data, creating an inventory and cataloging it, identifying overlaps and transforming raw data into trusted / quality data.

Organize data.

There is no benefit if an organization collects all the data needed but cannot easily find or access it when needed. Furthermore, for the data that can be found, there’s not always proper consideration of the controls that are put in place to comply with security/privacy regulations, Organizations need capabilities to find information, policies that help them define who has access to what and under what conditions, and then methods to make that data available through governed data lakes, which make that information available for analytics, data science, and AI.  Data federation and virtualization technologies are key to realizing this step.

Analyze data.

Once you know what information you have, made it available, accessible in a controlled way then it’s time to discover insights and derive value from it.  IBM provides a broad range of analytics capabilities covering descriptive (what’s happening), diagnostic (why it is happening), predictive (what is likely to happen), prescriptive (what actions to take). This is where the AI intelligence is built.

Infuse AI.

In this step, AI capabilities are delivered and integrated within new and existing business processes and applications. Infusing them with AI, enables them to become progressively smarter over time with each interaction with data and people – without having to continually rewrite and maintain complex declarative rules and code.

These four steps may sound simple enough, but Artificial Intelligence requires a strong Information architecture. An information architecture is not a static entity.   In some ways it is analogous to a living breathing thing.  It expands, contracts, morphs over time, takes on new capabilities to reflect the dynamic needs of the business.  Some services may need to be prioritized in terms of how they respond – for example, reacting in real time to support telemetry devices out in the field, or defense systems, medical devices, navigation systems, manufacturing equipment, etc. Other services may require higher degrees of integrity and security, or confirmation of delivery of a transaction.

Spanning the 4 steps of the AI ladder is the concept of “Modernize” – to help make the client’s data ready for an AI and multi-cloud world.  Organizations should be able to simplify and automate how they turn data into insights by unifying the collection, organization, and analysis of data, regardless of where it lives, within a multi-cloud data platform. 

In summary, the AI ladder is designed to help enable organizations to: 

  • Deploy an open information architecture for AI
  • Modernize their data estate for a multi-cloud world.
  • Make data ready for AI
  • Put open source to work.
  • Infuse AI everywhere, with confidence to drive better decisioning.

A few words on Openness.

In recent discussions with many organizations, they underscored the importance of vendors’ participation in open source projects including Kubernetes, Egeria, Spark, Jupyter. Things are rapidly evolving, and it’s become clear to me that no single company can create AI solutions in a vacuum and hope to keep pace with the rate of change in this key segment of our industry. For example, the IBM Cloud Pak for Data offering is dependent on Kubernetes. In fact, it requires and relies upon Red Hat OpenShift for its underlying infrastructure – from accessing storage, managing security services and delivering a containerized microservices strategy. Hummingbird is dependent on Spark. Watson Studio is dependent on Jupyter. Watson APIs are dependent on open source packages. By appropriately engaging with the open source AI ecosystem, vendors can gain insight into the areas that developers are most interested in AND better understanding of where the marketplace is going.

Tools to Help deliver AI across the Lifecycle.

IBM’s Watson tools leverage open source.  They provide a set of integrated capabilities designed to help organizations – for AI model creation and lifecycle management, as follows:

  • Prepare: Prepare and innovate for data discovery and activation. Watson Knowledge Catalog allows users to access, curate, categorize, and share data and assets wherever they are.
  • Build: Enable organizations to build their AI models. Consider this like an artist’s studio, or a carpenter’s workbench. They can create and train models to help with predictions. At this phase in the AI lifecycle, it is critical to ensure companies use the right algorithms to build their models for making predictions. Watson Studio supports this ability to build models using AutoAI to automate much of the work.
  • Run: Once a model has been built, it needs to be put into production inside an application or a business process. When a model is deployed, it’s now running in the organization, so it can make claims decisions, pricing decisions, and so on, and can be re-trained as needed. This happens with Watson Machine Learning.
  • Manage: Once a model is built and running, the question becomes: how can it be scaled with trust and transparency? In order to address complex or diverse build and run environments, enterprises need a tool that can not only manage that environment, but explain how their models arrived at predictions, and scale it across their organization. By having the management in place, organizations can track who changed the model, when the model was deployed, and the lineage on the model. By tracking all these items, organizations can be confident their models are not biased, and that they’re explainable and transparent. IBM’s answer for this is Watson OpenScale.

Great!  How do I get Started?

How you start or progress your journey to AI really depends on the business problem(s) you are trying to address.  You may have a particular project in the works to help address a specific business issue.  You may be new to AI or you may be on your “n-th” AI project already.  IBM provides various on-ramps to data and AI. For example, if someone just wants to start experimenting with AI, by wrangling data to build and deploy a model, organizations can start with Watson Studio, a tool that is available for the desktop or cloud.  Watson Studio is designed for collaboration across multiple personas to build a pipeline from raw data through to model deployment and management.

At other end of the spectrum an organization may already have deployed a number of AI solutions and might be using product offerings like SPSS and Watson Studio but want to progress to a data and AI platform that can hook into their enterprise, leverage all the data, provide governance, security, data management functions and exploit all the capabilities within the AI ladder. This is where Cloud Pak for Data and Cloud Pak for Data System can help.  Cloud Pak for Data is a pre-integrated data and AI. platform that can use existing compute and storage facilities or can include its own hardware in the form of a hyperconverged system – pre-integrated, pre-tuned and optimized ready to go. Finally, Cloud for Data as-as-Service will provide these capabilities as a multitenant service managed on IBM Cloud with multi-cloud connection capabilities available as a “pay-as-you-go” service or as a subscription model.

For some organizations though, the best strategy might be to start at the top of this ladder. That sounds counter-intuitive, but the AI ladder isn’t linear. It’s possible to start at the top, pushing AI through the organization, and then go back to work on collecting, organizing, and analyzing their data. An organization doesn’t have to start from a blank sheet. They can start with pre-built AI applications that can be adapted to their business situation. For an AI transformation to succeed, organizations have to infuse intelligence across all of their workflows. And pre-built AI applications are less likely to be trapped in the limbo between the lab and production.

There is no doubt in my mind that Cloud Computing has accelerated the adoption of AI through the migration of data assets to the cloud. By “renting” rather than purchasing data storage and processing capabilities, organizations can transform the economics of data management. They are no longer responsible for managing data access and security, no longer required to make large capital investments in computer hardware purchases, only pay for and access their data as they need it, and as a result, accelerate adoption of Big Data and AI.

But am I ready for AI

This is a question I often hear from many people: “Is my organization ready for AI?  Where am I in terms of AI maturity?” These are valid questions.  It’s not just about technology – It also about skills and the culture within an organization.

The concept of the AI Ladder is premised on the notion that organizations require a prescriptive approach to understand where they are in their AI maturity. By diagnosing and understanding the stage of maturity, an organization can then employ the AI Ladder as a framework for outlining the steps and capabilities that will guide them toward the realization of the benefits that result from machine and human augmentation. The basic tenets of the AI Ladder can be summarized as:

  1. Start with the business problem that you are attempting to address
  2. Understand your data requirements – these are the foundation for AI success
  3. Develop the right skills to leverage AI capabilities
  4. Focus on algorithmic trust and data integrity to ensure credibility
  5. Recognize the need for cultural and business model change.

IBM provides organizations with help in the form of a Data and AI on-line readiness assessment. After answering several questions on Collect, Organize, Analyze, Infuse, Modernize, Strategy an organization is presented with a chart of where they are, recommended next steps, how they score across each stage and the ability to drill down into each section in more detail.  There is also an opportunity to book a consultation with IBM experts who can provide services to help guide organizations through a data and AI engagement.

Figure #2: Example results from a Data and AI Readiness Assessment

Summary and next steps…

Just thinking about AI can be a daunting task for some organizations, let alone getting started or implementing it.  From a solutions perspective, organizations can choose to start simple, by experimenting with data and AI services in the cloud either as part of a no charge-trial or on a pay-as-you-go basis. Similarly, offerings can be downloaded locally. For organizations ready to make a platform investment, Cloud Pak for Data / System / as-as-Service offers an optimized data and AI platform designed to accelerate the AI journey with pre-integrated and pre-configured components – and designed to scale to meet future needs.

When assessing one’s own maturity and readiness for AI, IBM provides a data and AI online readiness assessment.  As described earlier, it’s a tool detailing where an organization is in terms of their maturity and readiness, represented by scores for each step of the AI Ladder, with recommendations for potential next steps.

Organizations seeking assistance with their data and AI projects may engage IBM experts to provide skills and knowledge transfer through a range of services using tried and tested agile methodologies and sprints. Some services are available on a no-charge basis while others are chargeable. Engagements are typically tailored around a client’s specific problem and can be short or long term. They include (but are not limited to) analysis across the data life cycle, data pipelines, feature engineering, the design, build, testing, deployment and management of models, minimal viable product, design thinking and strategy workshops – as well as install and configuration of IBM data and AI offerings.To find out more about AI Ladder and how to get started click this link.

Steven Astorino