The Rise of Cognitive Systems: Engineering AI-Driven Intelligent Platforms for the Future Enterprise

Abstract

The high rate of cognitive solution development is changing how businesses are developing, implementing, and expanding smart platforms. The paper explores the experimental literature that upholds the idea of AI-powered enterprise systems in relation to the performance of different model architectures in different data contexts. It has been shown that natural language processing models are far more successful at unstructured information processing compared to traditional models, and models like transformer-based models like BERT are uniquely better at generating a better search, document processing and conversational interface. On the other hand, more classical machine learning algorithms like boosted trees, which are highly efficient at lower complexity and higher interpretability, are still more suitable to structured enterprise data. The development of models and approaches to time modeling, such as the Temporal Fusion Transformer, evidences that AI systems can be more useful in helping to create predictions on a variety of horizons to enhance operational planning and resilience. It is important to note that the literature asserts that prediction accuracy cannot simply be adopted in an enterprise setting but explainability and governance are critical. Moreover, the idea of multimodal deep learning systems demonstrates the possibilities of heterogeneous platforms which are able to learn on heterogeneous data. In general, the results demonstrate the paradigm shift in the direction of fixed information systems to adaptable and decision oriented cognitive platforms that match the technical performance and organizational needs.