Symbolic and Statistical Theories of Cognition: Towards Integrated Artificial Intelligence

There are two types of approaches to Artificial Intelligence, namely Symbolic AI and Statistical AI. The symbolic and statistical paradigms of cognition may be considered to be in conflict with each other; the recent debate between Chomsky and Norvig exemplifies a fundamental tension between the two paradigms (esp. on language), which is arguably in parallel with a conflict on interpretations of quantum theory as seen between Bohr and Einstein, one side arguing for the probabilist or empiricist view and the other for the universalist or rationalist view. In the present paper we explicate and articulate the fundamental discrepancy between them, and explore how a unifying theory could be developed to integrate them, and what sort of cognitive roles Integrated AI could play in comparison with present-day AI. We give, inter alia, a classification of Integrated AI, and argue that Integrated AI serves the purpose of humanising AI in terms of making AI more verifiable, more explainable, more causally accountable, more ethical, and thus closer to general intelligence. We especially emphasise the ethical advantage of Integrated AI. We also briefly touch upon the Turing Test for Ethical AI, and the pluralistic nature of Turing-type Tests for Integrated AI. Overall, we believe that the integrated approach to cognition gives the key to the next generation paradigm for AI and Cognitive Science in general, and that Categorical Integrated AI or Categorical Integrative AI Robotics would be arguably the most promising approach to it.