Integrating Symbolic and Sub-symbolic Reasoning
This paper proposes a way of bridging the gap between symbolic and sub-symbolic reasoning. More precisely, it describes a developing system with bounded rationality that bases its decisions on sub-symbolic as well as symbolic reasoning. The system has a f
- PDF / 357,175 Bytes
- 10 Pages / 439.37 x 666.142 pts Page_size
- 41 Downloads / 274 Views
Department of Philosophy, Linguistics and Theory of Science, University of Gothenburg, Gothenburg, Sweden 2 Department of Applied Information Technology, Chalmers University of Technology, Gothenburg, Sweden [email protected] 3 Department of Applied Information Technology, University of Gothenburg, Gothenburg, Sweden [email protected]
Abstract. This paper proposes a way of bridging the gap between symbolic and sub-symbolic reasoning. More precisely, it describes a developing system with bounded rationality that bases its decisions on subsymbolic as well as symbolic reasoning. The system has a fixed set of needs and its sole goal is to stay alive as long as possible by satisfying those needs. It operates without pre-programmed knowledge of any kind. The learning mechanism consists of several meta-rules that govern the development of its network-based memory structure. The decision making mechanism operates under time constraints and combines symbolic reasoning, aimed at compressing information, with sub-symbolic reasoning, aimed at planning. Keywords: Autonomous agent · Bounded rationality Symbolic reasoning · Sub-symbolic reasoning
1
·
Survival
·
Introduction
Symbolic reasoning connects linguistic statements via syntactic rules, whereas sub-symbolic reasoning connects sensory concepts via association links [6]. These forms of reasoning have been studied since antiquity, e.g. by Euclid [2], who designed systems for axiomatic reasoning, and by Aristotle [10], who investigated associative as well as axiomatic reasoning. These forms of reasoning are closely related to James’ division into associative and symbolic reasoning [4] and Kahneman’s dichotomy of System 1 and System 2 processes [5]. The ability to do symbolic and sub-symbolic reasoning and to combine the two seems to be an essential feature of human intelligence [6]. In contrast, AI systems rarely support more than one of the two processes. For example, neural networks and reinforcement learning systems support sub-symbolic, but usually c Springer International Publishing Switzerland 2016 B. Steunebrink et al. (Eds.): AGI 2016, LNAI 9782, pp. 171–180, 2016. DOI: 10.1007/978-3-319-41649-6 17
172
C. Stranneg˚ ard and A.R. Nizamani
not symbolic reasoning, while automatic theorem provers and logic-based systems are the other way around. In particular, deep networks are good at recognizing faces or evaluating go-positions, but not at arithmetic, while automatic theorem provers have the opposite strengths. Several cognitive and agent architectures combine symbolic and sub-symbolic reasoning to varying degrees. Examples include Soar [8], ACT-R [1], OpenCog [3], AERA [14], and NARS [17]. Some of the architectures with this capacity are hybrid systems with juxtaposed subsystems operating on separate knowledge bases. Certain others are not fully autonomous, in that they depend on engineers for manually preparing the system for new domains, e.g. for updating the set of production rules. Despite the progress made, the following quote by Yoshu
Data Loading...
