Otago Polytechnic
CROP staff ChristopherFrantz 005

Helping computers think like humans

Christopher Frantz

How can computers model human behaviour accurately? Christopher Frantz, a lecturer within the College of Enterprise and Development, says software can easily deal with black and white rules but is not well versed in navigating grey areas. These grey areas may include behavioural norms that vary between cultures.

Typically, software programmes have relied on modellers’ intuition, additional rules or randomisation techniques to help fill in these grey areas, however these may not produce accurate results, let alone realistic behaviour.

Christopher explains it like this:  Assume you have two rules – Rule 1. You have to be at work at 8am. Rule 2. You shouldn’t run over pedestrians.  What do you do if a pedestrian runs out in front of you and you are late for work?  “The choice is logical for humans as we know that being late to work is preferable to killing someone.  However, computers have no road map for how to handle this type of situation, but need to be able to deal with this in an ad hoc fashion.”

This is where fuzzy reasoning comes in.  It allows software to integrate numerous opinions. and represents human behaviour more accurately.

Christopher’s research is a unique application which uses fuzzy modelling to examine underlying social norms and behaviour which opens up a raft of interdisciplinary uses.

One example of this considered the effect of the same word said by people of differing levels of authority. How does a command from the Chief Executive compare with an instruction from a colleague? The directive subjects followed depended which opinion they perceived to be more relevant or important, an effect possibly linked to authority levels. Such insights can be fed into artificial society models to allow differentiated, human-like behaviours to be represented by computers.

 

Frantz, C., Purvis, M. K., Purvis, M. A., Nowostawski, M., Lewis, N. D. (2015) Fuzzy Modelling of Economic Institutional Rules in Alireza Sadeghian, Hooman Tahayori (Eds.): Frontiers of Higher Order Fuzzy Sets, Springer, 2015, pp.87-129.