Learning with a purpose: The balancing acts of machine learning and individuals in the digital society

The Internet transformed some of the most basic processes in our society, such as trade, payment, and communication.  We now have more access to products, services, and opinions than we ever had, but at the same time, our behaviour is tracked more closely than ever. For example, large online retailers offer hundreds of thousands of products, and can readily observe in great detail how each consumer interacts with any them.  They can also rapidly deploy individual-level, in-vivo, randomized online experiments at population scale to test concepts, insights and communication approaches which can lead to better services and products. However, there are often billions of possibilities, such as product-consumer combinations for product recommendations. The scale and complexity of these experiments create amazing challenges.

Thus, firms face balancing acts. For example, they need to constantly choose between profiting from what they already know about consumers (such as the genres of movies already watched) and learning more about the same consumers (such as by recommending a movie of an untested genre). Consumers are also facing their own balancing acts. In the digital society, we inevitably leave digital footprints but we have some discretion in terms of how much information we want to keep private. Typically, a consumer that is more open to sharing her preferences is also exposed to higher risks, but at the same time she can get better access to products and services she needs.

In this talk, Professor Liberali will show how advances in machine learning and reinforcement learning can alleviate these challenging balancing acts. After providing some background information, he will briefly describe how these methods are helping firms and consumers, illustrating with his own work. Then he indicates key exciting areas for future research.  He will conclude this address by illustrating the implications for marketing science and prescriptive analytics more generally.

Professor Liberali’s address will be preceded by a two-day Workshop on Multi-Armed Bandits and Learning Algorithms. Participants from all around the world actively working in the development and application of multi-armed bandits and learning algorithms in various disciplines will attend. This forum will particularly encourage discussions of the approaches that have evolved in computer science, management science, operations research and statistics.