Semantics and symbol grounding in Turing machine processes

Semina Nr 16 Scientiarum 2017 s. 211–223 DOI: http://dx.doi.org/10.15633/ss.2492 Anna Sarosiek Semantics and symbol grounding in Turing machine processes The symbol grounding problem is one of the most important questions encountered in Artificial Intelligence research. The problem is very old and repeatedly discussed.1 It received special attention from Steven Harnad.2 There are many articles attempting to describe the way to solve this problem.3 Despite this, works on Artificial Intelligence still face obstacles to proper embedding of symbols of artificial systems. The difficulty of the symbol grounding problem is that we cannot accurately determine how to depict the content of a representative element in a symbolic system – how sensory data affects the whole of symbolic representations. Likewise, we do not know how sensory 1 See A. Cangelosi, A. Greco, S. Harnad, Symbol Grounding and the Symbolic Theft Hypothesis, in: Simulating the Evolution of Language, ed. A. Cangelosi, D. Parisi, London 2002; S. Harnad, The Symbol Grounding Problem, “Physica D: Nonlinear Phenomena” 42 (1990) iss. 1, p. 335–346; K. MacDorman, Cognitive Robotics. Grounding Symbols through Sensorimotor Integration, “Journal of the Robotics Society of Japan” 17 (1999) no. 1, p. 20–24. 2 See Professor Stevan Harnad, http://www.ecs.soton.ac.uk/people/harnad (22.11.2017). 3 See e.g. L. Steels, The Symbol Grounding Problem Has Been Solved, so What’s Next, “Symbols and Embodiment: Debates on Meaning and Cognition” (2008), p. 223–244; M. Taddeo, L. Floridi, Solving the Symbol Grounding Problem: a Critical Review of Fifteen Years of Research, “Journal of Experimental & Theoretical Artificial Intelligence” 17 (2005) iss. 4, p. 419–445. 212 Anna Sarosiek data influences the totality of symbolic representations. The problem is related to cognitive science, but it directly influences the process of creating intelligent systems. Such systems have to relate directly to the real world in order to develop cognitive skills and thus perform intelligent tasks. Meanwhile, the existing symbolic systems (irrespective of their intelligent or autonomous actions) do not use symbols belonging to them, but ones owned by their human architects. An artificial symbolic system is the result of a double translation of the world. Firstly, the external symbols are translated by a programmer into the language of human symbols. Secondly, they are transformed into a formal language, which can be implemented in the machine. That is exactly the issue we are dealing with in the problem of grounding. The symbols of the artificial system are not related to the world directly, but indirectly through the mind and the language of human maker. Therefore, symbols processed by the machine are semantically foreign to its system. They are not connected neither to the world nor to the internal machine environment. The creators of the contemporary computational theory were mathematicians and logicians.4 In their work, they did not consider physical and cognitive processes. It was much later that their successors understood there was a need to consider the problem of computation as related to the problem of the mind. For decades, the field of Artificial Intelligence has developed in a computational paradigm, which has been associated with computer science since its inception. 1. Question of computation One of the keystones of this area remains the Church-Turing Thesis.5 The Church-Turing Thesis states that any physical problem for which there is an effective algorithm of its solution can be solved by a Tu4 The major researchers were Alonzo Church, Alan Turing, Kurt Gödel, John von Neumann. 5 See B. J. Copeland, The Church-Turing Thesis, in: The Stanford Encyclopedia of Philosophy, ed. E. N. Zalta, https://plato.stanford.edu/entries/church-turing (28.11.2017). Semantics and symbol grounding in Turing machine processes 213 ring machine. Within formal operations, all logical inferences, computations, and proofs can be made by the machine. However, there are still open questions: do living organisms “calculate” things? Are there things that are not computable in the formal sense? Church-Turing Thesis generalization for physical system assumed that everything a physical system can do can also be done by computation. This physical Church-Turing Thesis takes two versions: weak and strong, depending on whether all physical systems are assumed to be formally equivalent to computers, or whether all physical systems are computers. This distinction provides the essential problem of computationalism – the thesis that cognition is only a form of computation. Analogously, this thesis comes in a weak and strong form. There are also alternative views on the features of computations. One of them presents a conviction that there are states not formalized yet, while the other assumes that there are states that cannot be formalized at all. It seems that in the latter case one should completely reject computationalism and assume that the Church-Turing Thesis is false. Such a solution would be trivial, because the CHT is supported inductively by the computational theory.6 This rejection would also result, on the one hand, in abandonment of considerations concerning the nature of symbolic systems, and on the other, in renunciation of research on intelligent systems. Consequently, the assumption of equality between computations and cognition allows us to better understand the nature of both and recognize the significant differences between them. Likewise, adopting the thesis as potentially true assumes that grounding of symbols of an artificial system in the natural world is possible, and the symbol grounding problem may be solved. In computers, the equivalent of cognitive processes are computations. Computations are the interpretation of symbol processing. Symbols are objects that are manipulated on rules based on their shapes.7 Symbols in an artificial system are interpreted as having See B. J. Copeland, The Church-Turing Thesis, op. cit. See S. Harnad, Computation Is just Interpretable Symbol Manipulation; Cognition Isn’t, “Minds and Machines” 4 (1994) iss. 4, p. 379–390; A. M. Turing, Com6 7 214 Anna Sarosiek a specific and immutable meaning. If we assume, in relation to the Church-Turing Thesis, that computations are universal, everything can be subjected to a formal interpretation. Every event, every relationship and every object of the world can be represented with symbols used by a Turing machine. However, we distinguish consistently computational processes from different kinds of things, e.g. mental states, emotions, or feelings. None of those can be implemented (so far) through symbolic systems in a computing machine. The reason is the way living organisms are grounded. 2. Meaning and grounding Symbol grounding is strongly connected with meaning. That makes the symbol interpreted in a particular way. In an artificial system, the interpretation of (...truncated)