Special Session on AGI & Cog Sci
AGI-2014 Special Session on AGI and Cognitive Science
For the first time, in 2014, an AGI conference included a session specifically focused on AGI in its relation to other aspects of Cognitive Science. In this special session we explored and discussed AGI’s integration and role within Cognitive Science. We presented the following contributions:
Vadim Bulitko (University of Alberta Edmonton): Flow for Meta Control
Pei Wang (Temple University, Philadelphia), Bas R. Steunebrink (IDSIA), and Kristinn R. Thórisson (Reykjavík University/IIIM): What Should AGI Learn From AI & CogSci?
Joscha Bach (MIT/Harvard PED): Artificial General Intelligence as a Foundational Discipline in Cognitive Science
Glenn Gunzelmann (Airforce Research Laboratory): The Role of Cognitive Science in Artificial General Intelligence
Background & Vision
Artificial General Intelligence is not simply AI reloaded, but may be best understood as a sub-discipline of Cognitive Science. AGI treats cognition and the mind as computational phenomena, and either strives to replicate (and exceed) human cognitive performance, or treat human and animal minds as specific cases of a more general class of intelligent system to be explored.
In the early years of the AI field, the leading approaches were mostly characterized by an understanding of cognition as the manipulation of rather abstract representations: the integration of real-world perception and action, autonomous learning, motivation and emotion were usually outside the scope of the models. AI has since seen several broad movements, including connectionism, statistical/probabilistic learning and modeling, and robotic systems.
Today, most work in AI focuses on applications or abstract methods rather than building mind-like systems. In contrast, AGI remains aimed at an understanding of the mind, in the context the progress in computational neuroscience and new insights into affect, motivation, autonomy, perceptual processing, attention, sociality, language, and so on.AGI is not well aligned with the boundaries of existing disciplines, but must draw from many fields. Unlike most work in the other sub-disciplines of Cognitive Science, AGI is ultimately always concerned with the creation of complete working systems and the identification of general functional principles, which are abstracted as algorithms and architectures.
AGI has to be informed by our best understanding of the mind, and thus cannot afford ignorance of cognitive neuroscience, cognitive and developmental psychology, linguistics, and philosophy of mind, to name just a few of the relevant fields. On the other hand, the constructionist approach of AGI offers a methodology to formulate, integrate and test theories from various cognitive sciences as computational simulations. AGI also offers formalisms, terminology and working systems that can potentially be applied to many individual problems of the other sub-disciplines of Cognitive Science.
Special Session Chair: Joscha Bach