Logics in AI (SS 2018)

Organizer: Marco Ragni
Assistants: Lukas Elflein, Nicolas Riesterer

Objective

Logics have been developed in large variety for artificial intelligence (AI). Among them are so-called non-monotonic logics that are especially useful in dealing with new informa- tion that can contradict previous knowledge. In cognitive science there has been recently a turn towards applying such logics to model human inferences, i.e. predicting human responses. In this seminar we will focus on non-monotonic logics and some findings from psychology and ask, if it is possible to model these findings by these logics. This seminar continues the successful seminar series consisting of self-study parts (i.e., the assigned logics and the psychological phenomena) and developing and defending an own approach (e.g., showing why or why not a logic can model the inferences).

Cognitive Modeling is a research discipline at the boundary of psychology and nat- ural sciences such as computer science, which aims at explaining human behavior on a computational level. Apart from matching the observable properties of human cognition as closely as possible, cognitive modeling is invested in the advancement of a general understanding of cognition. Instead of relying solely on abstract mathematical formal- ization such as neural networks, models are supposed to offer a means of interpretation while striving for functional equivalence to the mental processes.

Please find the extended Syllabus of this semester's seminar here.
The description for your modeling task is here.

Important Dates

• April 16th, 2018, 15:00 s.t.: First meeting and assignment of topics (SR 02-017, Building 52; Technical Faculty)
• May 14th, 2018, 15:00 s.t.: Preparation, Questions & Answers
• May 22rd-23th, 2018, 09:00-17:00: Blockseminary (SR 00-010/014, Building 101; Technical Faculty)

Deadlines

• May 5th, 2018, 23:59: Proof (see Syllabus)
• May 17th, 23:59: Presentation

Topic Assignment

Final Writeup

The final assignment is to refactor the presentation into a fleshed-out written report structured similarly to a technical research report:

• (Brief) Introduction
• Theoretical Background of the Logic
• Your Modelling Implementation
• High-Level Evaluation and Comparison with Experimental Data
• Conclusions

In particular, this document is supposed to give a basic, self-contained introduction of the logic, details about the model implementation (e.g., as a flow-chart), and evaluative conclusions with respect to the canonical and remaining patterns of the Wason Selection Task. How did you arrive at your model? What are the limitations? What parts of our model can be improved upon?