Last week we traveled to Boston with a delegation from Online Dialogue to attend the two-day Conference of Digital Experimentation at MIT (Massachusettes Institute of Technology) to attend. This conference is organized annually from MIT's Social Science faculty. Participants include scientists from leading universities such as Stanford and Harvard, as well as researchers at major tech companies such as Microsoft, Facebook and LinkedIn.

Since we at Online Dialogue find the combination between data and behavioral science very valuable, we were at the front with our notebooks at the many interesting presentations and conversations with the global frontrunners in this field.

Digital research in the social sciences

One of the most important innovations in modern social science is the ability to conduct research in complex social and economic environments using digital experiments. Through randomized experiments, it is possible to validate your assumptions/hypotheses at a rapid pace. This development raises specific questions and is a major change in the way scientific research is conducted.

The goal of the Conference on Digital Experimentation at MIT (CODE) is to bring together leading researchers conducting and analyzing large-scale randomized experiments in digital environments. The audience of this conference is therefore a mixture of different disciplines, including economics, computer science and sociology. In this way, they aim to lay a foundation for a multidisciplinary research community.

Moral machine

The conference looked at the topic of digital experimentation from different angles. One example was the research of Edmond Awad On the moral machine. As our machines get smarter and smarter, there are moral choices to be made. If a self-driving car drives right into a pregnant woman and can only avoid it by hitting an old lady, what should the car do? These are issues that must be carefully considered in a short period of time.

Experimenting within a social network

In addition, it was interesting to see the problems platforms like LinkedIn run into when they experiment online. When you participate in an experiment that makes you share more content on LinkedIn, it automatically affects your followers. They see that you are becoming more active and perhaps they will share more as a result. Because you have an effect on your followers, a normal random distribution on a person level is no longer enough. They have found a solution for this: Ego Clusters.

Long-term forecasts

Also shared were several prediction models that can dramatically reduce the running time of an A/B/n experiment. At a time when it is increasingly important to learn quickly from your users, this is one of those techniques that can be used to speed up.

For Online Dialogue, the combination of behavioral sciences and data very important. In all our optimizations, we use validated methods To be able to examine behavior. This conference is therefore perfectly in line with that and definitely recommended!