Perplexity logs: on routinized certainty work and social consequences of seeking advice from an artificial intelligence
Interactions, suspended moments of symbolic interaction and relationships
A sociological neo-structural approach to social phenomena can contribute to the current discussion of artificial intelligence and the social changes that it brings to society. We start here with the difference between relation and interaction. Donati (2014) identifies two schools: relationalist (pragmatist) and relational sociologists. The relationalist sociologists equate the two concepts (Dewey & Bentley 1946, Depelteau 2015). The relational sociologists think that interactions are different from relations in the sense that the former are reciprocal actions in their dynamics, while the latter are structures in their morphostatic/morphogenetic condition (Donati 2014, Donati & Aicher 2015). Here I will define personalized relationships as both channels for social exchange of resources and as symbolic and moral commitments to the exchange partner, and impersonal interactions as channels for social exchange of resources but without symbolic and moral commitment to an exchange partner. This distinction is theoretical since, over time, interactions can evolve to become relationships, and the other way around. It is nevertheless useful here because bureaucracy is often credited with the ability to extend and scale up interactions almost indefinitely; in it, indifference to everyone but the boss is accepted and can be the foundation of functional, short-term interactional social processes (e.g., solidarity or violence). Collegiality (not to be confused with congeniality) has the ability to extend deeper personalized accountability to a small circle of peers; indifference to no one in this circle is accepted as a norm, and this attitude can be the foundation for similar social processes (personalized solidarity or violence) and collective responsibility enforced with these personalized relationships as long-term commitments. Although theoretical, this distinciton helps understand certainty work and epistemic control.
Society includes both realms of interactions and relationships. Actors mix them (Goffman 1961; Donati’s hybridization (2019)) or superpose them (Lazega's multilevel approach (2020b)). I argue that for these co-constitutions and co-evolutions of interactions and relationships to be possible, actors must be able to step back and create a moment of symbolic interaction in which they ‘decide’ to go for the personalized or impersonalized route for their social exchanges and collective actions. When actors meet and judge the appropriateness of their joint involvement in collective action, they base their judgement on what could be called ‘symbolic interactional signalling'. This expression is derived from the general idea of relational signalling (see Wittek et al. 2003 for a synthesis): we signal to each other in which category of social exchange we plan to get involved, one with commitment (close personal relationship) or another without (arm’s-length interaction). Over time, however, most people need both registers to be able to play both games one after the other, or even simultaneously. The existence of a ‘suspended moment’ of symbolic interaction (SMSI) indicates in principle that a new, different project is always possible for collective action with the same alters. Thanks to this suspended moment, a symbolic, normative/cultural dimension is built into personal relationships but also into impersonal interactions.
Thus interactions and relationships are very different, but they have suspended moments of symbolic interaction in common. Here we will consider adviceseeking - defined as “What would you do if you were in my shoes?” - as an indicator of such a suspended moment of symbolic interaction. Indeed, when an advisor provides advice, s/he provides information and a framework for interpreting this piece of information and the appropriateness of its use. An advice relation embodies this symbolic interaction in an appropriateness judgement. Thus, between impersonal interaction and personalized relationships, suspended moments of symbolic interaction and signalling can be seen as appropriateness judgements. The next section defines the behavioural assumptions that can be included in a theory of appropriateness judgement that takes into account the capacity of actors to endogenize social structure (Lazega 1992) and to contextualize their common agency.1 By endogenization I mean perception of context as structure of opportunity or constraint, and use of this perception as guidance for behaviour. I will then present an empirical illustration approaching SMSIs based on the case of a multiplex network study in a corporate law firm called Sue Grabbit and Run (SG&R; Lazega 2001). In this case I will suggest how advice-seeking and its embeddedness in local structures and cultures of multiplex blending combines and exposes symbolic interactions, interactions and relationships performing appropriateness judgements.
Here local cultures and structures are illustrated by the kind of embeddedness and multiplexity that come attached to advice relationships. In collegial organizations, for example, advice relationships blend with other kinds of interactions and relationships. In the empirical case of this firm, both personalized ‘friendships’ and impersonal cowork interactions, although rarely combined directly (one rarely mixes pleasure and business), lead to advice. Advice is easily combined with personalized “friendships” on the one hand, or with impersonal cowork interactions on the other hand. This suggests that an advice tie is the closest indicator that we can find for a SMSI. It contains the signal that members of the this organization use to tell each other in which category of social exchange they plan to get involved, one with commitment (close personal relationship), or another without (arm’s-length interaction). With any one of these possibilities, a piece of advice will be considered appropriate in the eyes of at least one or more member(s) of the
Perplexity logs 71 same firm. This person has the capacity to think with you, to share more than just a piece of information, but also the interpretation of this information, the social framework that is needed to guide action that will be approved (or not) by others. It thus contains traces of the contextualization of the message that signal the suspended moments of symbolic interaction, and with it, with or without personal commitment, to whom people are accountable (at least through their reputation) for the way they reconstruct the problem and the possible solutions.
Finally I will argue that, in such social contexts, seeking advice from an artificial intelligence (Al) emancipates actors from such a local structure and culture of appropriateness judgements, with the often heavy endogenization of structure that it requires. However, there is a price to pay for such an emancipation (i.e., for advice without key social ingredients of the appropriateness judgement). First, given this understanding of how interactions and relationships work with such suspended moments of symbolic interaction in appropriateness judgement, seeking advice from an Al will at best provide an answer reflecting what, on average, people like me do in such circumstances, not “What you would do if you were in my shoes?” Indeed, at least contemporary Al agents avoid such answers in order to escape liability. So the perplexed who do not have access to human advisors (who are, for example, inaccessible on the other side of a social boundaiy) or who do not want to seek advice from human advisors can still access this information. Perhaps it is better than nothing, and not worse than seeking advice from humans who do not necessarily provide good advice anyway (Blau 1955). But Al advice will naturalize an artefact of method (i.e., the statistical computation that led to the selection of the piece of advice that I receive from this Al); it will not share with me the framework of interpretation. I will not refer to any socially shared appropriateness judgement nor trigger this suspended moments of symbolic interaction, especially with the collective legitimacy and normative force that comes attached with commitment. How appropriate it will be for you to follow this advice thus remains an open question.
Second, along with this freedom from local rules and institutions, there comes attached what I would label the curse of recursivity. Al may one day provide even very personalized advice, but it will not forget the history of one’s advice seeking. Query after query, actors will expose their uncertainties. Knowledge of others’ uncertainties is an ingredient in the very definition of power (Crozier 1963). The more Al helps citizens construct their certainties and the more it can build each advisee's perplexity log, the more power it accumulates over these citizens. With such recursive queries and answers, the latter become weaker and weaker. An Al not being a deus ex machina but a tool in the hands of very human and corporate owners, the latter will be in a position to take advantage of these individual and collective weaknesses. Al for the perplexed could lead to algorithmic regulation and its strategic political use. Advice-seeking from Al creates perplexity logs that will help third parties build the collective certainties escaping critical and democratic accountability. Individual guidance for the perplexed through Al without the suspended moments of symbolic interaction can provide bits of answers but will also reinforce a different and discrete process at a higher, collective level: that of influence and power accumulation that will be hard to observe and check.