THE EMERGENCE OF AI IN ENTERPRISE IT

The Emergence of AI in Enterprise IT

K R Sanjiv K.R.Sanjiv, Senior Vice President and CTO, Wipro

Ramaprasad K R Chief Technologist and Distinguished Member of Technical Staff, AI & Cognitive Computing, Wipro

For centuries, one of the more ambitious goals of mankind has been the creation of machines that rigidly “obey” commands. Around 322 BC Greek philosopher Aristotle imagined robots when he wrote, “If every tool, when ordered…could do the work that befits it…then there would be no need either of apprentices for the master workers or of slaves for the lords.” Since then, inventors, scientists and innovators have refined the idea of robots – from Leonardo Da Vinci’s clockwork knight to the Stanford Research Center which developed Shakey, the first mobile robot, Sony’s AIBO, Honda’s ASIMO and Google with its driverless cars. The quest has slowly turned from pre-programmed machines that did repetitive tasks to those that can sense the environment learn and respond to it. But the future belongs to advanced information processing or cognitive systems. These will bring about an epic shift in society, business and governance. AI (Artificial Intelligence) falls into two broad categories. The first is Natural General Intelligence (strong AI). Here, the focus is on building machines that think like human beings. The second is Applied AI. Here, the focus is on the use of advanced

machine learning and knowledge engineering techniques to build smart machines. In other words, Applied AI works at developing machines that act like people. The technologies that enable AI applications can be classified as Cognitive Computing technologies. Cognitive computing is a branch of computing that involves imparting cognitive capabilities to computers, so as to enable them to solve fluid problems. These problems are full of ambiguity; require contextual processing of a differing number of disparate sources which may not be known beforehand. Just like humans get better through practice and their goals change with their level of expertise in processing such problems, a cognitive computing system also improves itself through learning techniques. A traditional rules logic based computing approach will not be able to solve these problems. The central hypothesis of cognitive science is that thinking can best be understood in terms of representational structures in the mind and computational procedures that operate on those structures. Designing, building and experimenting with computational–representational models is the

Like humans get better through practice and their goals change with their level of expertise in processing such problems, a cognitive computing system also improves itself through learning techniques 21

WINS IG HTS Volume XXI

central method of developing modern AI applications (see Figure 1: ‘Cognitive Versus Traditional Systems’).

Vectors that shape AI applications

22

significant corpus of historical information in a specific domain will enable an AI application to extract key concepts, entitiesrecognition, associations, and hierarchies and generate what we call smart data by merging domain knowledge with ontologies.

Among many aspects that differentiate cognitive systems from traditional systems, the major ones are the ability to continuously reprogram themselves thus remaining flexible and the ability to interact in ways that are natural to humans. Apple’s Siri is one such example – Siri lets you do everyday things by talking. We are witnessing the arrival of television sets and mobile phones that respond to gesture, a glance or even the way we hold the device (example: face down for a mobile = mute).

The central hypothesis of cognitive science is that thinking can best be understood in terms of representational structures in the mind and computational procedures that operate on those structures.

The biggest thrust to cognitive computing has come from the availability of data. A

The emphasis has to be on access to a significant volume of associated data. AI

THE EMERGENCE OF AI IN ENTERPRISE IT

systems can be developed only when we have a significant corpus of data. For example in the Application Management and Infrastructure Management corpus of problems, diagnosis and resolutions, ITSM ontology will be essential to hyper automate processes and auto remediate problems without human intervention. When we explore some of the key characteristics of an enterprise AI application we find that there are six characteristics that are actually computable and relevant. These characteristics will shape the AI applications: • Naturally Interactivity: Improved humancomputer interaction, with mechanical middle layers such as a mouse being eliminated; these systems are conversational and have dialogue oriented natural language interfaces • Knowledge Representation and Meaning: Ability to ingest and represent knowledge; use automated knowledge models;

dynamically extend links to internal and external knowledge sources • Algorithmic Intelligence and Hypothesis: Perform computations and pattern recognition leveraging historical data – statistics, machine learning, NLP, optimization, ranking & scoring among others; generate evidence based hypothesis based on confidence scores. • Continuous Learning: Learn and evolve with common sense logic, new information/ inputs, new analysis, new users, new interactions, scenario modeling and simulation • Reasoning: Leverage language structure, probability, fuzzy logic, semantics and relationships to draw inferences • Hybrid Data Handling: Capable of integrating multiple heterogeneous data sources (structured and unstructured, static and streaming) and facilitate synthesizing ideas or answers from them We have already witnessed the power of predictive systems in reducing down time in manufacturing and transport, improv-

Cognitive Systems function differently, coming closer to ways that humans think and work

Figure 1: Cognitive Versus Traditional Systems

23

WINS IG HTS Volume XXI

ing healthcare and boosting efficiencies in industries as diverse as utilities, mining and retail. These predictive systems are ensuring that inventory is trimmed, maintenance is just-in-time and the right skill sets are available at the right place to

24

minimize loss. AI applications will raise the bar by creating intelligent virtual assistants, rapid software releases, straight through processing, diagnostics and resolutions, process dissipation, digital experience sense-and-respond, etc.

THE EMERGENCE OF AI IN ENTERPRISE IT

The Coming Wave of Autonomous Computational Systems

Not all of these systems will look the way we imagine them to be – part human, part machine in design, closer to a vision straight from a sci-fi movie. Rather, when applied to enterprise, these would largely be systems that use vast amounts of data, but apply algorithms that are shaped by a changing environment. Simplistically put, cognitive systems can sense using a variety of inputs ranging from sensor data to machine scanning of emails; they can learn through algorithms, statistical models, logic and probability; they can infer using analytics/computational intelligence/ artificial intelligence to mimic thinking and they can interact using natural language or gestures. “If companies take full advantage of intelligent automation,” says one Deloitte report, “the overall impact on business could rival that of the enterprise resource planning wave of the 1990.”

The future is now The development and deployment of such systems requires enterprises to become conversant with new disciplines and methodologies. They will need to create a deep understanding and competencies centered on the following 6 application categories for AI systems: • Anticipatory and Predictive Systems: These would allow organizations to be proactive rather than reactive systems

• Intelligent Virtual Agents: Graphical bots that can interact with humans and respond to words and gestures • Phantom Robotic Process Automation: This allows process automation without human intervention • Visual Computing & Human Computer Interface: These would include advanced models for data ) (images, video and text) representation (3D could be one example) and processing it using with new methods of interacting with machines such as language, gestures and glance to make the interaction more natural • Knowledge Processing Systems: These would include logic and decision trees that enable agents to work more accurately by acquiring, retrieving and processing knowledge on their behalf • Autonomous Robots & Drones: These would be intelligent machines that operate independently in environments that humans may find hazardous or impossible to access within limited budgets.

Major productivity gains will be unlocked by the wave of autonomous computational systems that can sense, learn, infer and interact. Enterprises are now plugging in cognitive computing technologies to develop AI applications. The vast data that they have in their data warehouse and AI engine which learn from the data and helps them do predictive analytics, automated hypothesis, verification and generation are enabling them to deploy such systems. Enterprises can create bots for process and task automation, virtualize knowledge, build mobile

25

WINS IG HTS Volume XXI

virtual agents for the digital customer that can enable enhanced interactions and user experience with the customers. The big shift will be at the intersection of business process that will be mapped to an AI engine to enable business efficiencies and productivity. AI applications developed using Cognitive computing technologies are among the most interesting recent developments in computational science. Several popular open source stacks are available using which IT service providers are creating loosely coupled services for a wide application across IT and business processes. Open source stacks, corpus of data and related domain specific ontologies can create killer applications. They impact practically every business discipline and replace human beings in several tasks and enhance abilities in key process

26

areas. The benefits of productivity, speed, quality and scalability are immense. They take practically every function and business process to a higher level of performance. Integrating these systems at every level within organizations will also call for changes in people and process practices. For the moment, organizations must ask themselves how deep they want to plunge into leveraging cognitive systems. Enterprises should start experimenting through pilots using innovation offerings from IT services and only after proven pilots decide on vendor specific propositions. Do they have relationships in place with the ecosystem of AI labs and service providers who are already in the data, analytics, machine learning and natural language processing space? If not, it is time to do so.