Comparing Computer Models Solving Number Series Problems

Inductive reasoning requires to find for given instances a general rule. This makes inductive reasoning an excellent test-bed for artificial general intelligence (AGI). An example being part of many IQtests are number series: for a given sequence of numbers the task is to find a next “correct” successor number. Successful reasoning may require to identify regular patterns and to form a rule, an implicit underlying function that generates this number series. Number series problems can be designed along different dimensions, such as structural complexity, required mathematical background knowledge, and even insights based on a perspective switch. The aim of this paper is to give an overview of existing cognitive and computational models, their underlying algorithmic approaches and problem classes. A first empirical comparison of some of these approaches with focus on artificial neural nets and inductive programming is presented.

• Type: Journal
• Year: 2015
• Where: AGI
• Authors: Ute Schmid, Marco Ragni
• Keywords: AGI 2015, Artificial Intelligence