Machine learning and neural networks

Present situation worldwide and in Switzerland

Artificial intelligence (AI) technologies such as machine learning algorithms and artificial neural networks can be trained to perform specific tasks with an efficiency and precision that complement or even surpass human ability. The most popular AI technologies such as deep-learning networks are loosely inspired by the way the human brain learns. They typically consist of layers of nodes (neurons) linked by adjustable weight connections (synapses). Each neuron can transmit information to another group of neurons, based on the state of the weighted input. Weights adjust as learning proceeds. The dramatic increase in computing power available in the cloud and in supercomputers, advances in AI research and the recent consolidation around large data platforms have brought AI technologies great success and public visibility. Machine learning methods are increasingly shaping our everyday life: speech recognition, translation, natural speech interaction with machines and image/facial recognition have changed the way in which individuals interact with other individuals and with businesses. However, the disruptive potential of these AI technologies rests on their integration within large data platforms. Machine learning methods are at the core of automating data acquisition, labelling and classification in large data platforms. This will increase their acceptance and greatly accelerate their uptake across all sectors and by businesses of all sizes. Disruptive changes will ensue for banks, services, retail, manufacturing, pharmaceutical and insurance companies.

Implications for Switzerland

Switzerland has played a leading role in the development of AI technologies. Owing to its outstanding ecosystem for academic and industrial research in the field of machine learning, it remains very well positioned provided research investment and a favourable environment for business and innovation are maintained. In future, research should focus on the development of robust, precise and resilient machine learning methods based on high-dimensional real data (e.g. medical or industrial data). Switzerland has the opportunity to take on a pioneering role in the definition of standards and the certification of AI systems. With the increasingly frequent use of AI, it becomes ever more important to ensure that such models are ethical and safe: they should be devoid of prejudice regarding age, gender, nationality or religion, execute their task within defined boundaries, and be safe from foreseeable attacks. In addition, the need to improve public perception of AI technologies should not be underestimated.