The limits of Modeling in AI
Speaker: Paul Robertson , CSAIL
Date: November 30 2005
Time: 1:00PM to 2:00PM
Location: 32-G449
Host: Metin Sezgin, CSAIL
Contact: Metin Sezgin, 3 2663, mtsezgin@csail.mit.edu
Relevant URL: http://projects.csail.mit.edu/dangerous-ideas/dangerous/www/The limits of Modeling in AI
Science has been dominated by model building. Propose a model for a phenomenon that accounts for the observed phenomenon. Find things that it fails to account for (or inaccurately accounts for), propose a new model that accounts for everything that the previous model
accounted for and more... In AI we have additionally been concerned with performance (combinatorial explosion). The approach has been very successful in science in general and has been spectacularly successful in AI too. Despite AI's extraordinary success it has
fallen short of expectations in certain dimensions. Modeling will not get us the rest of the way.
In this talk I will discuss:
1. Why modeling might be good enough for other sciences but is not for AI;
2. If modeling isn't enough what WILL get us the rest of the way; and
3. Why would we consider any other approach when the current approach
has been so successful.
See other events that are part of Dangerous Ideas Seminar Series Fall 2005
See other events happening in November 2005
