Welcome back everyone!
The opening colloquium speaker for the fall semester is Kieran Marray. Kieran is a Laidlaw Scholar from St Catherine’s College, University of Oxford. He has been visiting the Center for Social Complexity over the summer to do research in complexity economics supervised by Professor Rob Axtell.
Kieran’s talk entitled “FORTEC: Forecasting the Development of Artificial Intelligence up to 2050 Using Agent-Based Modelling” (abstract follows) is scheduled for 3:00 p.m. on Friday, August 31, in the Center for Social Complexity suite located on the 3rd floor of Research Hall. The talk will be followed by a Q&A session along with light refreshments.
This session will be live-streamed on the newly created CSS program YouTube channel: https://www.youtube.com/channel/UC7YCR-pBTZ_9865orDNVHNA
For announcements regarding this and future streams, please join the CSS/CDS student and alumni Facebook group: https://www.facebook.com/groups/257383120973297/
For a list of upcoming and previous seminars, please visit: https://cos.gmu.edu/cds/calendar/
We hope to see you on Friday, August 31.
FORTEC: Forecasting the Development of Artificial Intelligence up to 2050 Using Agent-Based Modeling
The past decade has been characterised by massive leaps forward in fields like machine learning and computer vision. Will it last? How should we think about this problem? I shall argue that agent-based modelling can be used to forecast the emergence of innovation in AI and should be able to do so better than the current models. I shall do this as follows. Firstly, I shall review previous attempts, specifically Osborne and Frey (2013) and McKinsey (2017). However, I shall show that the methodology underlying these approaches is flawed due to its reliance upon expert opinion. This can be reduced by simulating the individual firms and their actions in an empirically realistic manner. So, I shall attempt to do this, building an agent-based forecasting model (FORTEC) from the bottom up. It will be applied to the set of firms and research labs in the US working on AI to derive the first empirical predictions for the ability to automate nineteen different types of task from now until 2050. The model is far from perfect, though; it is intended simply as a first approximation for others to build upon. Therefore, I shall finish by laying out some ways in which this framework could be applied or improved, from forecasting innovations in other high-technology industries such as photovoltaics to simulating non-American firms.