Reply in collaboration with the SDA Bocconi DEVO Lab for Data Robotics

The Data Robotics Lab is an innovation laboratory developed by Reply in collaboration with the Bocconi DEVO Lab, a research centre within the SDA Bocconi School of Management focussed on the themes of digital transformation. The laboratory is a centre of expertise that combines the perspective and scientific methods of the Business School with Reply’s technological expertise.

The goal of the collaboration is to develop a benchmark for the exploration, experimentation and analysis of opportunities for the intelligent automation of business processes, enabled by the increasingly more evolved capabilities of modern Machine Learning techniques. These techniques, in fact, enable the expansion of the action perimeter of traditional software robotics solutions (Robotic Process Automation) towards new and less structured domains. This represents a first step towards the paradigm of Autonomics (business systems in which technology operates independently, carrying out a large number of activities essential to the functioning of the enterprise, without the need for specific human input.

The study thus focusses on the aspects of Data Robotics with respect to Artificial Intelligence and Machine Learning, to the stratification of the Artificial Intelligence software solutions market, to the available Artificial Intelligence technologies and their positioning with respect to company priorities (HIT Radar methodology, developed by the DEVO Lab in collaboration with the MIT Design Lab) and finally to the domain of high impact business applications enabled by Artificial Intelligence technologies.

When examining these themes in greater depth, it is important to emphasise the “human” nature of Artificial Intelligence. Machine Learning techniques represent a new paradigm in the computing world, such that the machine is able to identify the program best suited to provide a specific output (given a series of “constraints” or rules), rather than having to be exhaustively programmed to return a unique output. The flexibility of machine learning in the “autonomous” processing of input data is both admired and feared, especially when faced with scenarios that see Artificial Intelligence applications increasingly closer to autonomous decision-making abilities which can have a direct business impact. Nevertheless, according to the Data Robotics Lab, the true issue lies not in the software, but in the human resources and in the processes that feed it. As a learning engine trained on company data, Machine Learning will reflect the interpretation rules of the data supplied by those who feed the machines with new information and the characteristics of the data to which the machine is exposed.

This leads to two fundamental managerial considerations regarding:

• Decision-making bias: since it is a human who defines the machine’s “reasoning / learning” constraints in terms of nature and intensity, we must be aware of the risk of transferring a series of typically human prejudices and decision-making limitations to Artificial Intelligence;
• Data quality: often company data is dispersed across many locations, becoming disaggregated and inconsistent due to application updates, and is reprocessed at different stages by different business functions. An organisation must therefore plan to invest significantly in the structuring of ordered data flows, as well as in an analytical culture that values the quality of data as an essential element for the quality of decisions driven by the same data.

Thus, if technologically speaking Machine Learning continues to make progress, supported by increasingly more sophisticated algorithmic techniques and a growing and focussed computational power, on the managerial front it is essential to remember that:

• Artificial Intelligence is not a magic solution; it is software whose effectiveness increases with the quality of the data that feeds it
• There is a risk of an “algorithm bias”, namely of the transfer of human decision-making limits to the machines
• In order to ensure the full effectiveness of AI solutions, a significant investment is required in the quality of the methods used to generate and process data
• The main barriers to the dissemination and effectiveness of Machine Learning-powered solutions are data quality and the consistency of the analytical processes and culture
• In most cases, the paradigm likely to succeed will be that of the Human-in-the-Loop, in other words a condition in which man and machine work side by side, where one benefits from the data processing capacity, while the other learns in incremental logic from human-generated insights. The human will remain ultimately responsible for the decision-making process for many years to come.