Probabilistic single function dual process theory and logic programming as approaches to non-monotonicity in human vs. artificial reasoning

Abstract

In this paper, it is argued that single function dual process theory is a more credible psychological account of non-monotonicity in human conditional reasoning than recent attempts to apply logic programming (LP) approaches in artificial intelligence to these data.LP is introduced and among other critiques, it is argued that it is psychologically unrealistic in a similar way to hash coding in the classicism vs connectionism debate. Second, it is argued that causal Bayes nets provide a framework for modelling probabilistic conditional inference in System 2 that can deal with patterns of inference LP cannot. Third, we offer some speculations on how the cognitive system may avoid problems for System 1 identified by Fodor in 1983. We conclude that while many problems remain, the probabilistic single function dual processing theory is to be preferred over LP as an account of the nonmonotonicity of human reasoning. Dual process theories (Evans 2003(Evans , 2007Evans & Stanovich, 2013;Kahneman, 2011;Sloman, 1996;Stanovich, 2011; Stanovich & West, 2000;Wason & Evans, 1975) invoke two separate cognitive systems to explain performance on a variety of cognitive tasks. These are labelled System 1 and System 2. System 1 is rapid, parallel, automatic, do not require the resources of working memory (WM), and only their final product is posted in consciousness.In contrast, System 2 is slow sequential, and analytic and makes use of the central working memory system. In particular, System 2 ''permits abstract hypothetical thinking that cannot be achieved by System 1' ' (Evans 2003, p. 454). Recently, Oaksford and Chater (2012) argued that accounting for non-monotonic or defeasible reasoning in dual process theory required that both System 2 WM representations and System 1 long term memory (LTM) representations need to be interpreted probabilistically. This position is consistent with Evans and Over (2004; see also Over, Evans, & Elqayam, 2010) adoption of probability logic (Adams, 1998) as underpinning analytic processes in System 2. But it contrasts with accounts which treat analytic processes in System 2 as underpinned by standard binary truth functional logic (Heit & Rotello, 2010;Klauer et al., 2010;Rips, 2001Rips, , 2002 Stanovich & West, 2000;Stanovich, 2011). Oaksford and Chater (2012) labelled the former approach, the single function dual process (SFDP) approach and the latter the dual function dual process (DFDP) approach. Our goal in this paper is to confront some further problems and challenges for the probabilistic SFDP approach but first we rehearse Oaksford and Chater's (2012) argument in detail. Probabilistic Single Function Dual Process TheoryBoth the dual process theory (Evans 2002) and the probabilistic approach (Oaksford and Chater 1991, 1998 developed out of a critique of the classical logicist approach to cognitive architecture (Fodor 1975;Pylyshyn 1984), which is a logical single 4 function dual process theory. The store of world knowledge in LTM consists of a consistent set of logical formulae i...

Imaging deductive reasoning and the new paradigm

“…Properly analyzed, accuracy remains the same between conditions, and believability effects are pure response biases. They argue that their modeling results, “provide support for processing theories of deduction that assume responses are driven by a graded argument-strength variable, such as the probability heuristic model proposed by Chater and Oaksford ( 1999 ).” Their results are also consistent with probabilistic single function dual process theory (Oaksford and Chater, 2012 , 2014 ). There is a clear distinction between processes based on long term memory for our beliefs about the world and processes that require working memory.…”

“…The if…then construction is used ubiquitously because it can be used to describe the various relationships or dependencies in the world, like causes, dispositions, intentions, regulations and so on, which allow us to predict what will happen next and to explain why what happened happened. The reasoning mind is likely to be very concrete constructing specific small scale models of reality in System 1, like Kahnemen’s banana-vomit example or using specific relations, and reasoning over these (Oaksford and Chater, 2013 , 2014 ; Oaksford, 2014, Submitted). These last two points make the argument that there are functions, f 2 and f 3 , that are in contention to account for the tasks that engage MeFG.…”

Logic, probability, and human reasoning

“…The mental model theory requires no special logic for the task ( pace [38][39][40][41][42]). Simulation can be performed within a protected environment -an intellectual laboratory to try out hypotheses -to model other individuals' inferences, to envisage causal interventions or counterfactual possibilities, and to explain inconsistencies [65,[68][69][70].…”

“…However, human reasoners faced with a solid fact tend to withdraw any conclusion that it contradicts. Some theorists therefore defend so-called 'non-monotonic' or 'defeasible' logics developed in artificial intelligence, which allow conclusions to be withdrawn [38][39][40][41][42].…”

Dual-process theories and consciousness: the case for ‘Type Zero’ cognition: Table 1.

“…Focusing instead on processes, there are results showing that automatic, type 1 processing can produce incorrect performance in circumstances where, given more time or less cognitive load, subjects are able to meet a normative standard by engaging in type 2 processing. (Whether this is best theorized in terms of a distinction between two different systems, rather than two modes of operation of the same system, is a further question: Osman 2004 ; Kruglanski and Gigerenzer 2011 ; Oaksford and Chater 2014 .) In short, the findings about conscious versus non-conscious representations are compatible with the findings about type 1 versus type 2 processes.…”