Architecture and Models of Artificial Intelligence Platforms

Introduction. This article presents a brief overview of modern artificial intelligence (AI) architectures and models for use in digital platforms. This review identifies emerging trends, advantages of modern models, and promising areas of application, which are driving the development of both scientific and industrial technologies. The article provides information on the functional features and shortcomings of AI models and architectures and draws conclusions about the future direction of AI platform development. The purpose. The purpose of this article is to provide a brief overview of the main AI architectures and models, substantiating the methodological and theoretical shortcomings of existing approaches and predicting future development paths for AI platforms. Results. The main types of AI architectures used in digital platforms are briefly described. Dedicated, multi-tenant, and hybrid architectures are compared. Modern linguistic models of artificial intelligence, which are fundamental for the development of AI platforms, are listed and their main shortcomings are outlined. A list of alternative approaches to creating AI models is presented, and it is concluded that event-driven models are the most promising when used as foundational models for digital platforms. A list of companies and their approaches to digital intelligence based on event-driven models is provided, which enable not only event tracking but also event prediction. In conclusion, the article compares the key characteristics of the LLM, EDM, and SEM/RSM models that dominate modern digital platform architectures. Conclusions. The development of architectures and models for AI platforms requires addressing the existing methodological gap caused by the infatuation with large linguistic models among AI platform architects. To overcome this methodological gap, a new cognitive architecture is needed, based on the principles of "Platform Thinking," that is, a transition to a subject-event approach based on Situational-Event Models (SEM) and Role-Subject Models (SRM). Furthermore, multi-tenant configurations will likely remain an important area of development for the creation of universal AI services.