CognitiveScale: AI for the Enterprise, 5 Years in the Making
Five years ago, I began what has been the most exciting professional journey of my life. On July 1, 2013, I left my job at IBM after 7 years and started CognitiveScale in my dining room. The timing was right—it was the early days of the third AI revolution, and we were at the beginning of a market that would take many years (perhaps decades) to fully mature. With the hype and excitement surrounding IBM Watson, I had many opportunities to build early cognitive systems in my role as the leader of IBM Watson Labs. During this time, I made two key observations about the challenges and potential of AI in the enterprise. First, for the modern enterprise, data is plentiful, but only a small fraction of it is usable to create differentiated business value; and organizations were spending millions of dollars getting the right data ready for AI before seeing any value from AI. Second, as human beings, we have more tools available today than ever before to store data, query data, analyze data, and visualize data; however all of these tools (including the Watson Q&A system) suffer from the same limitation_ as the human user, you have to know what question to ask in order to get meaningful insights from data. Too often, the unknown unknowns are what contain the key insights required to make an impact.
These two observations, and the challenges they create, are what excited me and our earliest investors about the potential of building AI-powered business systems that instead of requiring the human user to query, ask, or click, would instead learn how to provide the user with the right insight at the right time using a variety of data sources even beyond those that humans themselves could fully analyze or understand. What if the human user need not ask a question? Instead, she was assisted by an AI that learned how to fill in gaps in her knowledge and augment her capabilities—giving her superhuman ability. It would be like each user having their very own J.A.R.V.I.S. supercomputer to help them navigate the healthcare system as a patient, or make better investment decisions as financial client. The implications for the enterprise workforce were also very exciting; each employee could expand their skills using augmented intelligence to work smarter and faster by using AI to amplify processes intelligence.
It was with this vision and excitement about the potential of AI that CognitiveScale was born. We began focusing on healthcare, banking and financial services, and commerce industries to find high impact problems we could solve. Still, one big obstacle needed to be addressed_ most organizations struggle to find the right data and thus the right business problems for AI. In the early days of IBM Watson commercialization, there were a handful of clients that basically decided_ “Watson is the answer, now let’s go find a question.” While these early adopters had an appetite for multi-year, multi-million dollar science projects, most organizations did not. When we started CognitiveScale, we quickly learned that in order for AI to be adopted, it needed to show value quickly—in other words, AI needed to be Practical. We developed a methodology—called 10-10-10—that allowed us to identify a target business problem in 10 hours, stand up an AI system to begin solving that problem in 10 days, and measure the value the system creates in 10 weeks. Our goal is for our clients to start seeing value in 90 days or less. The 10-10-10 methodology includes picking problems that focus on one of two areas_ AI-powered customer engagement or AI-powered process intelligence. This allows our customers to harness our core technology, assets, and experience to accelerate the build out of a new AI system. We also provide the tools to measure the performance of the system against business measures and KPIs. Instead of only focusing on statistical measures like precision and recall that are hard to translate into measurable business impact, we translate AI system performance into business terminology such as conversion lift % or cost savings generated per day.
Over time, we realized that beyond the first few AI systems, our clients were beginning to think about the future and how to scale up to many new systems across the enterprise. Because of the specialized nature of AI systems which include a complex anatomy of data, machine learning models, knowledge representation, business logic, and user experience, we found that our clients needed a set of tools and methods to reliably and repeatedly build and deploy AI. AI needs to become Scalable as an enterprise-wide capability in order for it to succeed—not just a bespoke and specialized dark art performed in a lab in Silicon Valley. We also recognized that the IT landscape was evolving very quickly—a super-convergence of cloud, mobile, AI, blockchain, and big data technologies was happening all around us. In order to achieve scalable AI in this ever changing landscape, we created Cortex—a set of tools, methods, accelerators, and an AI virtualization layer that allows clients to run their AI systems on multiple cloud/AI platforms like AWS, Microsoft Azure, Google Cloud, and IBM Cloud platforms, as well as inside their private cloud or private data center environments. By helping to abstract and orchestrate the underlying cloud services provided by these platforms, our customers now have a single programming model to define, build, and deploy AI systems consistently—even in hybrid cloud environments or multi-cloud environments. The development of Cognitive Skills—an open specification based component model for building composite AI applications—and the CAMEL family of specifications was a key breakthrough to enable an ecosystem of reusable and composable AI assets within and between enterprises. Like a virtual machine for AI, Cognitive Skills allow data scientists and developers to package ML models, business rules, and business logic into a common component model that is then deployable on their cloud/AI platform of choice. To further this concept, we also created the world’s first visual AI composition environment called Cortex Studio, which makes it easy for developers to assemble, configure, and deploy AI systems. These innovations combined with the world’s first AI asset store, the AI Global Marketplace, are all aimed at scaling and accelerating AI in the enterprise.
As AI becomes more prevalent in our everyday lives, questions of ethics, responsibility, and legal liability start to move to the forefront. When AI becomes essential to critical decision making functions in enterprises, and even governments, we must be prepared to answer questions about fairness, interpretability/explainability, and robustness of the machine learning models and underlying data ecosystems that power these AI systems. Not to mention the role of AI in our daily lives—healthcare, transportation, commerce, and banking to name a few—where AI will be directing, and in some cases deciding, what actions to take. The movement towards Responsible AI is critical to make sure that our civilization at both global and national levels is protected from rogue AI systems that may not follow the expected ethical norms society expects. To advance this effort, this year CognitiveScale has pledged to positively impact 1 billion lives through the application of responsible AI in the industries and geographies we work in. This will come through deployment of responsible AI systems, development of open standards and specifications, sponsorship, and participation of non-profit organizations (such as AI Global, the IEEE, and the World Economic Forum) that focus on the ethical implications of AI in society, and through the development of the world’s first AI Ethics Switch in partnership with the D7 group of digital nations and the World Economic Forum who just named CognitiveScale a 2018 Technology Pioneer in part for our leadership in responsible AI.
As we look towards the next five years of the company, I am encouraged to see the increasing level of collaboration between academia and industry in the areas of machine learning and ethics, blockchain and trust protocols, and development of open standards and specifications for AI systems—all of which will ultimately make the third AI revolution more than just today’s buzzword bingo. CognitiveScale will continue to focus on helping our clients pair human and machine to transform customer experience and process intelligence, all while making AI Practical, Scalable, and Responsible for all.