Overview

This excerpt is from NMR. The editors comment that: Suchman made a deep critique of AI, that planning and symbolic manipulation is flawed as models of human intelligence. Suchman argues instead that human reasoning is not based in planning, but rather, action is based on embodiment. Suchman claims that plans are stories used to justify and explain these actions. This critique influenced Philip Agre and David Chapman, who explored AI in consistency with Suchman’s argument.

Notes

The preface, “Navigation” opens with a discussion of Trukese and European navigators. The European navigator operates in accordance with the plan at all times, while the Trukese navigator instead operates only with an objective in mind, using circumstance and conditions to alter his course. The metaphor can inform three possibilities: One, that how to act is purposefully learned, and is different across cultures. Suchman follows to counter this with the counterargument that all activity, even the most analytic is fundamentally embodied. Two, one might argue that planning is used instrumentally, depending on experience or expertise. But this seems to imply that the Trukese navigator would not get anywhere. Three, Suchman’s critique, is that all purposeful actions are situated in their circumstances: we act like the Trukese, but talk like the Europeans. Rather, plans are an ad-hoc resource for the action. This metaphor is an excellent representation of Suchman’s critique, but it also exposes some other qualities that may easily slip by. One is that it was Eurpoean thinking, heavily based in Cartesian dualism, that led to the development of computers and AI. The absence of non-western thinking pushes other forms of reasoning and philosophy into the background in common electronic models of cognition and interaction.

Suchman begins by discussing Turkle’s research on computers as collaborative objects. Computers are reactive, linguistic, and internally opaque: this leads to design challenges, especially with accountability. Computers seem to reason, but the manner of that reasoning is concealed. On Automata: cognitive science has pushed a symbolic metaphor of cognition, with AI and computers being the logical receptors. Cognitive science emphasizes the detachment of rationality from embodiment, and supports the abstract symbolic reasoning pervasive in AI today. Suchman discusses the linguistic metaphor for interaction in HCI. This emphasizes interaction as a dialogue, compared with dialogue between two people. In such case, as Dennett argues, mutual opacity makes intentional explanations much more powerful. Thus, the opacity of computer invites an intentional stance. From the design perspective, artifacts do and should try to explain themselves, but this is muddy water when it comes to intentionality. Opacity, especially in certain untrusted situations can place a user at odds with an treacherous and untrusted world. Obviously, this is not the dominant perception of computers or computation, rather they are seen as extensions or objects. Interaction is not a dialogue, so much as commands and filtering. From an AI perspective, though, the concern is natural.

Suchman finishes off the chapter with a comparison of the computer as an “artifact designed for a purpose” versus “an artifact having purposes”. The former, which is the instrumental approach, evokes embodied and situated reasoning, with the computer as an adaptable tool. The latter approach is that of AI, which (as far as it is instrumental) treats the computer as an intelligent device, which engages with the user reflexively, and is not *usable* in the literal sense. However, again, this form of reasoning gets convoluted when applied to games, which, having no extradiagetic goals, have no purpose. Games are most engaging when reflective and automated, but simultaneously evoke an intense state of situation via their immersive character.