Michail Bletsas
MIT Media Lab
slide 1 of 16  

Michail Bletsas is a Research Scientist and the Director of Computing at MIT's Media Lab where he has been working since 1996. He was one of the founding members of the "One Laptop Per Child" initiative where he made significant contributions to the design and implementation of OLPC's pioneering "XO" laptop as well as its Internet connectivity. He co-founded Velti S.A., a very successful mobile advertising and marketing company and has consulted for many companies in many different capacities. Over the years, he has implemented broadband access networks using cutting-edge technologies including one of the earliest ADSL testbeds and various wireless technologies. Mr. Bletsas has been a frequent keynote speaker in international conferences and has been engaged in many civic activities. He holds an undergraduate degree in Electrical Engineering from the Aristotle University of Thessaloniki and an MSc degree in Computer Engineering from Boston University.

Abstract

Digital marketers are obsessed with the last click, advertisers with the last ad that precedes conversion. The traffic jam you're in must be due to an accident. Health insurers' pay-for-performance schemes are based on strict adherence to guidelines during one visit to the doctor. Disasters are blamed on the events that trigger them. A very large deficit is attributed to a president in office. A couple getting divorced believe their last fight is the cause. We as humans are obsessed with finding simple, preferably linear, cause-effect relationships in everything. Yet we know that things are very, very rarely as simple as we would like them to be. Our urge to simplify has been shaped by evolutionary pressure and is probably well-adapted to slow-changing, relatively simple environments. In today's world, simplifying heuristics can be counter-productive. One universal issue we face when dealing with complex systems is that of attribution: how to assign figures of merit to the multiple touchpoints that may have contributed to an outcome? Giving all the credit to the last touchpoint is a simplifying heuristic which makes no sense in a wide range of situations. The marketing funnel should be called marketing maze. The traffic jam you're in may be due to no single reason -there may not even be a trigger. Performance in healthcare should be an ecosystem concept, and not be reduced to one visit in a vacuum. Disasters are usually the result of many interacting factors working in synergy. A president in office is probably only partly responsible for a country's deficit. And you're not getting divorced just because of a nasty argument last week. We need a different framework for dealing with these situations and move beyond the usual culprit/scapegoat mindset. A machine learning approach known as reinforcement learning provides a powerful new lens to augment our causal intuition. Concrete examples from advertising, healthcare and risk management will illustrate how this could become a revolutionary tool.

Alessandro Vespignani . Geoffrey West . Kathleen C. Engel . Ren Y. Cheng . Eric Bonabeau . David Lazer . Adam Price . Hamid Benbrahim . Cesar A. Hidalgo . Marta Gonzalez . Sokol Celo . Michail Bletsas . Greta Meszoely . Stoyan Tanev . Charles Worrell . Kun Zhao . Dany Bahar . Brian Peltonen .