The dark side of analytics
Only companies that have implemented or started to implement analytical data departments in their business models are called to stand out from their competitors. Large-scale data management technologies, machine learning, deep learning and artificial intelligence are, in contrast to traditional management, the iron weapons at the end of the bronze age. Those who have them and know how to use them will lead their respective sectors.
But of course, like any innovation, there are immense advantages, but there are always disadvantages and the new analytical techniques also have their dark side. This dark side is based on the, sometimes, little explanation of the models used, the so-called black boxes, the biases that can be experienced in the data lists used for the elaboration of these models, the legality when extracting the data, etc.
Governments, led in Europe by the European Commission in its document entitled “On Artificial Intelligence – A European approach to excellence and trust“, have already begun to advance in these matters and prevent the development of these technologies from taking us down undesirable paths.
From the beginning, Wenalyze has been concerned about this issue and has considered what we call trust elements that any analytical development on our part must comply with in order to follow our company’s quality standards.
– Ethics: the development, deployment and use of any artificial intelligence solution must adhere to fundamental ethical principles, in accordance with the guidelines of the European Commission’s High-Level Expert Group on Artificial Intelligence
– Explicability and interpretability: A model is explainable when its internal behaviour can be directly understood by humans (interpretability) or when explanations (justifications) can be given for the main factors that led to its result. The importance of explainability is greater when decisions have a direct impact on clients/humans and depends on the context and level of automation involved. Lack of explanability could represent a risk in the case of models developed by external third parties and then sold as “black box” (opaque) packages.
Explicability is only one element of transparency. Transparency is about making data, characteristics, algorithms and training methods available for external inspection and provides a basis for reliable model construction.
– Equity and avoidance of bias: equity requires that the model ensures the protection of groups against discrimination (direct or indirect). Discrimination can result from bias in the data, when the data are not representative of the population in question. To ensure equity, the model must be free of bias. However, it should be borne in mind that bias can be introduced in many ways. Techniques to prevent or detect bias exist and are still evolving.
– Traceability and auditability: the use of traceable solutions helps to follow all steps, criteria and choices throughout the process, allowing the repetition of the processes that lead to the decisions made by the model and helping to ensure the auditability of the system.
– Personal data protection: Although at Wenalyze the models that we currently analyse and exploit do not process such data, in the event that they could be processed in the future, we protect them from the design stage, with a reliable BD&AA system (Big Data and Advanced Analytics) that strictly complies with current data protection regulations.
– Data quality: the issue of data quality must be considered throughout the entire life cycle of the BD&AA, as consideration of its fundamental elements can help to gain confidence in the data processed.
– Security: new technological trends also bring with them new attack techniques that exploit security vulnerabilities. It is important to maintain technical vigilance on the latest security attacks and related defence techniques and to ensure that governance, oversight and technical infrastructure are in place for effective IT risk management.
Wenalyze will therefore always be vigilant of changes in technologies and in the techniques and models it applies to its business in order to avoid any use and/or diversion of the models it markets or makes available to its clients now or in the future.