Sixty-five years ago, the world learned about ENIAC, the first smart calculator. Ever since, computing technology has progressed in leaps and bounds in response to the challenges of storing, managing and extracting value from an ever-increasing volume of data. Today, the exponential growth of real-time data generation is posing even greater challenges for individuals and businesses. How can companies tame this huge stream of data? How can they analyze zettabytes of information in emails, documents, news and comments in social networks and utilize them for better business outcomes?

These and other challenges have motivated companies like Amazon and Google to develop and introduce AI-powered solutions in their services. Their successes have inspired major players in banking, retail, medicine and other industries to extensively invest into and in many cases fully acquire startups in the field of AI. Today, companies and their customers all over the world are already reaping benefits from AI. In banks, it helps to check information for loans. In the energy sector – assesses the wear of equipment, in health care-determines the diagnosis of patients. According to Infosys, 86% of the world's major organizations are already using or starting to implement AI. Such pilots have started to bring the first set of results: they have shown to attract customers while also saving time and money for these companies.

Robots are increasingly able to perform both manual and routine cognitive tasks. They came into production in the 1960s, replacing people on assembly lines in factories, especially in countries where labor is expensive. Now, robotic process automation (RPA) is poised to take over offices. A combination of factors including growth of global competition, global rise in labor cost, and the need to reduce costs of business processes while increasing accuracy and repeatability of outcomes have made the RPA onslaught on modern offices inevitable. As the volume of information increases, so is the need to reduce the cost of its processing. With software robots potentially being able to automate up to 70% of office work, hiring new people is no longer the best option. This partly explains the dizzying growth of the technology: the RPA market is expected to surge to $2.9 billion from a mere $250 million in 2016, a tenfold increase, according to Forrester.

A colleague without a cubicle or walls

Contrary to common perception, software robots are not mechanisms on wheels, but are “virtual” employees who work side by side with humans albeit using a computer interface. In accounting, organizations use RPA to extract and transfer data between accounts and for transactional reporting and budgeting. In human resources, RPA helps automate HR tasks, including onboarding and off-boarding, updating employee information and timesheet submission processes. Companies in the financial services industry can use RPA for foreign exchange payments, automating account openings and closings, managing audit requests and processing insurance claims.

For a long time, the most popular way to automate these processes was through cross-integration of existing information systems. But as technology advances apace, legacy systems are becoming the most significant barrier to business. With the emergence of software robots came an array of benefits. Robots do not require salaries, do not rest and do not get tired and make fewer errors than humans. They are also easier to deploy and require no extra office space. In most cases, the payback period is just 6-9 months. So one can assume that in the next 3-5 years, the demand for these technologies will grow actively wherever it is necessary to quickly process a large amount of information on a given algorithm.

Of course, for now, the capabilities of software robots are rather limited: they do not know how to analyze complex data types, especially unstructured ones. Meanwhile, up to 80% of business-relevant information originates in unstructured form, primarily text: contracts, letters, news. As new opportunities appear that accelerate and simplify the introduction of intelligent technologies, office robots will be smarter with AI. These include open source libraries like Tensorflow, free courses in machine learning and new types of neural networks.

What to teach office robots

As office robots increasingly become part of value creation processes in the service sector, the former paradigms of automated manufacturing are being put into perspective again. The next wave of development should see businesses teaming up with technology developers to adopt a new approach to machine learning: competitive neural network, reinforcement learning and transfer learning. While these approaches are relatively new, they are bound to play a leading role in the development of AI in the near future.

How do competitive neural networks work? Let’s say two machines receive similar datasets at the point of origination. Then one of them starts to create new information based on them – for example, images of documents that look authentic. The task of the second system is to assess how convincing the outcomes are. Children learn this way – for instance by playing ball with each other after learning the rules of the game and the basic techniques. This too, is how unmanned vehicles learn about dangerous traffic situations and this is how ABBYY technology learns to extract meaningful information from various sources.

In reinforcement learning, the machine analyzes the environment using a virtual model that mimics the features of the external environment. That is how AlphaZero taught itself to beat the world’s best chess-playing computer program by calculating the combination of moves and choosing the winning one. These developments will soon move from games to business, where they will help investment analysts and risk managers to choose the most profitable option for the company. Finally, transfer learning offers an opportunity to use the same neural network for similar but not necessarily identical tasks. This explains how ABBYY technologies work. For example, if the solution is able to analyze labor contracts, the same system can be taught to work with sales contracts, which speeds up and reduces the cost of app development.

Within the next three years, these methods will be increasingly used to train systems in many large projects as they are simpler and cheaper to implement. The greatest success will be achieved by organizations that not only correctly combine the capabilities of the artificial and natural intellect, but also functionally combine RPA, AR, AI and other ultramodern solutions to achieve a synergistic effect from their use. Such an approach will lead to the creation of systems that can be truly called intelligent. In the meantime, businesses should start delegating tasks to software robots. They will not replace us in the workplace, but they will become our obedient hands, so that we can work more with our heads.

Henry Patishman is Director of Sales (Australasia) at ABBYY. Contact ABBYY at sales@abbyy.com.au or on  (02) 9004 7401 for any further information.