..what is machine learning, exactly? Stanford University computer science professor Andrew Ng defines it as “the science of getting computers to act without being explicitly programmed.” In fundamental terms, machine learning is a branch of artificial intelligence that is meant to replicate the way humans take in information from their environment to make better-informed choices for the future*. But realistically, will machine learning transform the way businesses are managed?
It was a regular work day—I had entered my office premises, showed my face to the small screen on the attendance system and it marked my attendance. Once inside my cabin, I was reflecting on what just happened and I realized that face recognition and other forms of biometric solutions, including fingerprint scanning and retina scanning, have been indeed a part of a larger wave called Artificial Intelligence (AI) which began a few decades ago but seem to have gradually entered our daily lives. Household robots, robotic arms in automobile factories and hazardous environments, recommendation engines used in websites and searches on net, and self-driving cars are but some examples of how AI is penetrating our lives.
what is AI and machine learning
AI, which traces its roots to research done by scientists for over a century, has been explored in many different ways—though not always successfully. Mimicking physical tasks done by humans (robotics), replicating [the functions of] sense organs (motion sensing, face recognition, etc.), cognition, rule-based decision-making, solving puzzles, playing games (IBM’s Deep Blue against Gary Kasparov), replicating practices of experts in various domains (expert systems) are examples of approaches taken towards understanding AI and developing new applications for it.
AI is primarily about creating machines capable of demonstrating some aspects of human intelligence. The next frontier in AI is to build into the machine the human capability to acquire new learning based on past experiences, emerging patterns, and predictions. This is what is known as machine learning. It is this ability which can make the machine adaptable at the least and autonomous in its extreme form, as exemplified in part by
a self-driving car. AI has so far used statistical analysis, neural networks, and the like to help in training a machine based on past data and patterns. Autonomy, however, means that the machine not only operates on every new situation based on the training data provided, but also self-corrects the learning data.
enterprises: automated or autonomous?
With increased usage of IT-based technologies, mobile and the cloud, along with reduction in costs of processing, transmission, and storage, all enterprises have accumulated a vast amount of transaction data as well as subject-based repositories of the key aspects of their business such as products, processes, employees, assets, and customers. This accumulation and centralized availability of transaction and subject-based data have been a result of implementing workflow, business process automation, enterprise applications, content management solutions, and data warehouses.
Social media and the IoT have created the need for a culture of real-time sensing and responding. This goes far beyond the traditional means of utilizing data in the form of MIS and business intelligence. Many of these organizations have thus reached an advanced stage of what can at best be described as automation.
Some of them have gone a step forward and brought in analytics in the form of end-user analytics (ad hoc query and data visualization), and used analytics in areas such as customer segmentation and targeting for the purpose of planning promotional campaigns. A few of them have built rule-based decision systems—these rules can be modified but the need for a new rule or a change in an existing one still needs to be done manually.
Perhaps the more established field of manufacturing has been using more sophisticated forms of automation such as robotic arms or SCADA and process control systems. However, these too tend towards automation and do not qualify to be called autonomous systems.
Some evidence of AI and machine learning is however evident in solutions such as chatbots such as the female voice in a GPS system which helps us navigate while driving the car, or a front-end device which after a bit of training starts recognizing text written by a user using a stylus, or a voice-based system which recognizes the voice of a person and acts on a command. Even advanced analytics solutions now being used in customer [relationship management] and risk [management] banking and net forensics do have an element of learning, if not autonomous behavior.
towards an autonomous enterprise
What differentiates an autonomous enterprise from an automated one is the ability to sense the context, predict future possibilities, and act on the basis of these predictions without any human intervention. For example, the risk and compliance process in a financial institution could acquire new knowledge based on previous risks, and predict future risk scenarios based on extraneous syndicated information on contextual factors. Equipped with this information, the risk management system could alter its threshold levels or trigger certain actions on its own. This would be considered an autonomous system.
Likewise, an online autonomous health advisor agent who has access to real-time monitoring data of people living in the local area could predict the outbreak of an epidemic, or at least be able to diagnose a patient more accurately based on his or her health data and available community information.
What organizations need to do is study their own business processes and decisions, and identify where AI or machine learning could be applied to substitute or supplement the human expert responsible for making the decisions or running the processes with an equivalent self-learning machine or algorithm. Approving loan applications; predicting defaults; advising on investments, insurance or health; proactive recommendation engines in various situations, and provisioning of a guest room based on knowledge of past preferences and behaviors of a guest (in the hospitality industry) are but a few possibilities.
It has already been discussed that much of the legal processing could be automated—perhaps someday even judgement agents/bots could help clear the backlog of routine legal cases which are currently overloading our judiciary.
The ‘smart’ behavior expected in smart cities could be another great opportunity for changing the way we live. A self-programming signaling system which alters itself depending on traffic conditions, or an autonomous road barrier which redirects traffic if the normal route gets choked, are all interesting possibilities in the real world.
impact of autonomous enterprises
While automation in enterprises helped in delayering an organization by taking away the burden of information capture, processing, and reporting thereby impacting the supervisory and white-collar clerical layers, autonomous processes could take away much of the operational and to some extent, even tactical decision-making. It will make business processes quicker and more reliable, accurate, relevant, and scalable. In its extreme, an autonomous organization could well continue to keep working without manual intervention until the purpose for which such an organization was created is no longer relevant in the real world. Redefining the purpose will remain a human task at least for a long time to come.
AI and machine learning have a lot to offer. The day is not too far when you may stop owning a car and just step into the first empty car near your home and expect it to drive you to your destination; when various services and offers are sent to you just when you need it; when some bot or agent proactively detects your health problem and activates the healthcare delivery system.