What is the current stance of Swiss companies on machine learning (ML)?
Ireneu Pla: Swiss companies are expressing a growing interest in machine learning, but the landscape is not uniform. Some players are simply curious and are trying to discover how ML can help them, while others are betting on artificial intelligence (AI) to innovate and improve their efficiency. More skeptical organizations which are waiting for a stronger proof of value can also be found. A common observation among organizations which have more data and could make the most out of it, is that they are caught in the tyranny of the urgent. Without proper prioritization or support, ML efforts are easily set aside as general digitalization efforts/needs take the front stage. Exploring this field requires some investment and having the right expectations. A successful experiment is, after all, the first step to actually benefit from ML.
How does ML actually make it from proof of concept to production?
Ireneu Pla: To obtain productive systems from a working proof of concept, a holistic perspective is needed. This means leaving the realm of the data scientist and dealing with an engineering challenge. It’s crucial to consider that ML can be computationally demanding and that there are good approaches to deal with mistaken predictions. When users are supposed to interact with the results, careful UX design is also key. The perceived quality of an ML system greatly depends on how the information is used and presented. These efforts are often underestimated, and with the increasing complexity of software systems may lead to a promising model ultimately being disregarded.
What challenges does ML pose for cybersecurity?
Ireneu Pla: Having ML at the heart of interactive software and critical processes opens new avenues for cyberattacks. While one of the values of ML is its ability to learn from data and improve with new examples, this characteristic makes it hard to fully understand how some predictions are generated. At Adnovum, we have identified ML security as an area where the industry is lacking awareness. Additionally, the elements to evaluate these threats are often out of the scope of typical security assessments, since evaluating them requires a deep understanding of how ML works as well as the context in which it is run. However, it is not all bad: in many cases dealing with this new type of vulnerabilities is very straightforward.
How can ML help companies leverage their available data?
Ireneu Pla: There are two main ways in which organizations can make full use of their data. The first one is enhancing analytics. The aim is to extract new information to better understand contexts and generate new actionable insights. While humans are very good at understanding context, a suitable algorithm can sometimes find novel and meaningful patterns. The second main way is to use ML to digitalize steps that are not automatable with typical software. These cases may arise from novel findings revealed by analytics. Notwithstanding, it is also possible for companies to benefit from ML without relying on their own data. Existing models and cloud AI offerings allow implementing solutions that significantly improve or enable new processes.
This interview was initially published in the ICT Yearbook 2022 of Netzwoche.
Conversational AI webinarWatch recording