Generative AI

Source: arxiv.org

The Challenge

Generative AI introduces profound organizational shifts that render traditional IT adoption frameworks obsolete. Most prior scholarly discourse focuses narrowly on machine learning algorithmic architectures or domain-isolated use cases without considering overarching socio-technical ramifications. Consequently, organizations and Information Systems researchers struggle to contextualize generative AI within business workflows, leading to uncoordinated technological adoption, severe misalignments between technical capabilities and operational governance, and an acute lack of structured research agendas within Business and Information Systems Engineering (BISE) disciplines.

Core Findings

Feuerriegel et al. present a structured taxonomy conceptualizing generative AI across mathematical foundations, hierarchical systems, and socio-technical dimensions. The authors delineate the fundamental mathematical divergence between discriminative and generative modeling before decomposing implementations into a tripartite view: the model-level (foundational deep neural architectures like diffusion models and transformers exhibiting emergence and homogenization), the system-level (infrastructural integration, data pipelines, and RLHF fine-tuning), and the application-level (domain-specific interfaces). Grounded in socio-technical systems theory, the framework maps generative AI onto seven core BISE domains, including business process management, digital leadership, decision analytics, and human-computer interaction, underscoring systemic risks such as hallucinations, algorithmic bias, and labor substitution.

Strategic Takeaway

For academic leaders and organizational strategists, this conceptual framework bridges technical machine learning principles and socio-technical enterprise implementation. Leaders must treat generative AI not as a static IT tool, but as an evolving socio-technical agent requiring tripartite alignment: foundational model governance, system-level infrastructure integration, and contextual application management. The paper offers actionable guidance for updating business process management, redefining digital leadership roles, and designing resilient human-AI collaboration interfaces while navigating operational disruption, IP exposure, and intellectual labor reallocation.

Deep Dive Q&A

How does generative modeling differ mathematically from discriminative modeling?

Discriminative modeling learns conditional decision boundaries to classify data points into distinct classes, whereas generative modeling infers joint probability distributions to generate entirely new, synthetic data samples that mimic the underlying training distribution.

What is the tripartite model-, system-, and application-level view?

The model level encompasses the core neural architecture and foundation models; the system level includes supporting infrastructure, training pipelines, and user interfaces; and the application level addresses practical contextual implementations solving specific end-user problems.

What core foundation model properties affect enterprise adoption?

Foundation models exhibit emergence, where capabilities arise implicitly without task-specific training, and homogenization, where a unified consolidated model powers an extensive variety of downstream organizational applications.

Related Research & Implementation Playbooks

Cross-disciplinary frameworks and practical blueprints connected to this insight.