Dynamic Algorithmic Service Agreements Perspective

TL;DR

This paper proposes the Dynamic Algorithmic Service Agreements (DASA) framework to improve human-AI interactions by addressing limitations of traditional Terms of Service agreements, emphasizing adaptability, co-constitution, and feedback mechanisms.

Contribution

It introduces the DASA framework as a novel approach for self-regulation of AI systems, enhancing interaction and cooperation with human users beyond static ToS models.

Findings

DASA enables dynamic feedback loops between users and AI systems.

Illustration of DASA in a recommender system context.

Addresses sociotechnical concerns like change, co-constitution, and friction.

Abstract

A multi-disciplinary understanding of the concepts of identity, agency, relationships, interactions, and information could help us develop mitigation strategies for responsible human-algorithmic systems in the field of AI. It is imperative for us to question the use of the Terms of Service (ToS) agreements model in the context of algorithmic systems, specifically AI systems that make decisions which affect people and their livelihoods. In this position paper, we identify five areas of concern in traditional ToS agreements by drawing on studies of sociotechnical systems in Science and Technology Studies - accommodating and enabling change, co-constitution, reflective directionality, friction, and generativity. We aim to address these ToS shortcomings and propose components of a novel Dynamic Algorithmic Service Agreements (DASA) framework. The DASA could be employed as a self-regulation framework while also enabling additional feedback loops between people and algorithmic systems. Rich interaction frameworks could enable us to better negotiate and cooperate with AI systems towards accomplishing the real-world goals we use them for. We illustrate the DASA framework in the context of a Recommender System used in the curation of real and synthetic data. We do not intend for the DASA framework to replace the ToS model, but instead think it will provide practitioners with an alternative point of view for the design of dynamic interaction interfaces for AI systems that account for human identity and agency.