Representation, Analytic Pragmatism and AI

Our contribution aims at individuating a valid philosophical strategy for a fruitful confrontation between human and artificial representation. The ground for this theoretical option resides in the necessity to find a solution that overcomes, on the one side, strong AI (i.e. Haugeland) and, on the other side, the view that rules out AI as explanation of human capacities (i.e. Dreyfus). We try to argue for Analytic Pragmatism (AP) as a valid strategy to present arguments for a form of weak AI and to explain a notion of representation common to human and artificial agents. 1. Representation in AI The notion of “representation” is at the basis of a lively debate that crosses philosophy and artificial intelligence. This is because the comparison starts from the analysis of “mental representations”. First, we move by adopting a fruitful distinction between the “symbolic” and the “connectionist” paradigms in AI [1]. This distinction is useful to highlight two different ways of explaining the notion of representation in AI. An important challenge for AI is to simulate not only the “phonemic” and “syntactic” aspects of mental representation but also the “semantic” aspect. Traditionally, philosophers use the notion of “intentionality” to describe the representational nature of mental states namely intentional states are those that “represent” something, because mind is directed toward objects. The challenge for AI is therefore to approximate to human representations i.e. to the semantic content of human mental states. If we think that representation means to connect a symbol to the object of representation we focus on the discreteness of mental representations. On the contrary, it could be plausible to focus on the interrelation of mental representations. The first corresponds to the symbolic paradigm in AI, according to which mental representations are symbols. The second corresponds to connectionism in AI, according to which mental representations are distributed patterns [2]. 























































 1
Pontifical Lateran University The task to consider the similarity between human and artificial representation could involve the risk of skepticism about the possibility of “computing” this mental capacity. If we consider computationalism as defined in purely abstract syntactic terms then we are tempted to abandon it because human representation involves “real world constrains”. But, a new view of computationalism could be introduced that takes into consideration the limits of the classical notion and aims at providing a concrete, embodied, interactive and intentional foundation for a more realistic theory of mind [3]. We would like to highlight also an important and recent debate on “digital representation”[4] that focuses on the nature of representations in the computational theory of mind (or computationalism). The starting point is the nature of mental representations, and, particularly, if they are “material”. There are authors who maintain that mental representation are material [5] others thing that thought processes use conventional linguistic symbols [6]. The question of digital representation involves the “problem of physical computation [7] as well as the necessity of the notion of representation [8] so that we only have the problem of how to intend the very notion of representation [9]. But, there is also the possibility of understanding computation as a purely syntactic procedure or to include “every natural process” in a “computing universe” [10].