Data governance: one of the pillars of digital trust

Companies’ digital transformation offers enormous possibilities for automating and even industrializing processes. These changes are driven by what is called artificial intelligence, often referred to as the ability of machines to make decisions.

There are actually several ways of using digital technology in gradual degrees of autonomy and delegation of skills. In the early stages, machines assist in decision making; in the later stages, machines make the decision. Five stages of skill delegation can be described. First, the machine can be used to describe the data, i.e. to read it and to make a representation of it, graphically or by sound for example. In the second stage, it can be used for diagnostic purposes, i.e. interpreting a situation, such as displaying a warning if a critical value reaches a certain level. The next stage is prediction: the machine predicts the future based on the analysis of the past or the present. The stage that follows is prescription: the machine gives advice on what should be decided according to its analysis of the situation, like a GPS that changes the route according to traffic. The final step is the decision: the machine is autonomous, it reads the data, analyses it, interprets it and translates it into action. This is what happens with the autonomous car, the automatic cash register, the intelligent heating system, etc.

In short, the more we automate, the more we delegate the analysis and interpretation of data, and even decision-making. The degree of automation allows, or not, correction, moderation or human interpretation. In other words, the more we automate, the more fundamental the quality of the data is, since it has a direct impact on the decision-making process without any human interpretation. The tendency to reduce or even eliminate decision-making time by using technologies capable of reacting in real time further reinforces the delegation of skills to the machine.

The degree of delegation depends on several factors, both voluntary and involuntary. Sometimes, because of the costs of developing fully automated systems, companies decide to keep manual steps alongside automatic processes. This decision can also be driven by the desire to maintain the social and human dimension, such as in supermarkets where automatic checkouts and traditional checkouts co-exist, or even bla-bla checkouts that promote social interaction. In these two cases, the choices are opportunistic, based on cost/benefit approaches. In other cases, the low quality of the data does not allow the processes to be fully automated. One might think that data quality can be improved through financial investments. But this is not enough because data quality is a complex matter that is strongly linked to corporate culture. Data is an asset that is unlike any other and its particular characteristics require adapted working methods^[1][2]. In order to be effective, digital transformation must not be perceived as merely technical, it must also be accompanied by a cultural transformation.

Whether the machine is simply an executor of algorithms or whether we talk about machine learning with learning capabilities does not solve the problem of data. The computer calculates, searches, makes mistakes, tries again and progresses! But it can’t distinguish between good and bad data. It is up to us humans to make available data that we judge to be accurate and in line with our values. Unbiased, with integrity, accessible, diversified, representative, objective … these are the criteria that we can favor when we select the data that we provide to an artificial intelligence.

Also, sometimes, data is difficult to interpret. Perhaps you are familiar with those funny images of chihuahuas that look, to the human eye, like muffins^[4]? The human eye misunderstands them and the algorithm even more. As Yann LeCun explains, artificial intelligence has no model of the world – or at least a limited one. But the interpretation of data is a relative matter. Knowing whether you are facing a chihuahua or a muffin consists in recognizing an image. If in doubt, the human eye relies on the context (a plate laid in the kitchen suggests that it is a muffin). This is something that the machine struggles to do since it does not (yet) have a model of the world. Delegating the interpretation of data to machines can therefore be a source of error or bias that must be dealt with and anticipated.

At a corporate level, ensuring the quality of data must go through data governance, which is inseparable from corporate governance: corporate culture, training, management of data collection, use and storage, risk management, definition of roles and responsibilities and, more generally, a critical look towards the autonomization of processes. There is no doubt that in several years, data will be managed, accounted for, standardized and even regulated as is the case today with financial assets. In the meantime, companies must impose “soft rules” if they want to take advantage of digital technology without suffering the risks and constraints.

In conclusion, automating processes means delegating the interpretation of data to machines. The result is a growing need for corporate governance that takes into account the dimension of the data, a distinctive asset that requires an adapted corporate culture and rules to manage flows and stocks, roles and responsibilities. This is one of the pillars of digital trust.