The Covid19 pandemic had a massive impact on daily life worldwide, especially in terms of how to contain and control the spread of the disease. Before the introduction of vaccines, many countries worldwide had to adopt socio-political and environmental interventions to reduce the number of people admitted to intensive care and deaths on the one hand and try to keep the economy and social welfare at an acceptable level on the other. Adopting such non-pharmaceutical interventions (NPIs) has provoked a considerable debate in governments to identify a sufficient balance between closing services and preserving public health.

During the first wave of Covid19, Sony CSL, in collaboration with the Complex Systems Hub in Vienna, attempted to assess the effectiveness of NPIs in light of disease prevalence data provided by international organisations. We compared four statistical analysis and machine learning approaches to determine the NPIs used in 74 countries worldwide. In particular, Sony CSL contribution was to build an AI capable of evaluating the effectiveness of NPIs and building models powerful enough to construct What-If machines. The models created by Sony CSL made it possible to provide a clear indication of the efficacy of NPIs and, at the same time, to assess the most appropriate period for their application. Some NPIs have demonstrated that they belong to the ‘the earlier, the better’ category, while others can only give a valuable contribution after a specific time.

Finally, it was also possible to assess the consequences of the late application of measures with a high impact on disease control in terms of new positives and deaths.
Full details of this work have been published in Nature Human Behaviour.