Skip navigation
Like what you’re reading?

AI as a service: How AI applications can benefit from the network

  • Artificial Intelligence (AI) is already a key part of 5G networks, and is increasing in use and importance as the networks evolve.
  • Future 6G networks could also expose AI capabilities to applications through user-friendly APIs.
  • Read this blog post about the concept of AI as a service – a promising way to unlock a wide range of novel use cases where AI applications benefit from using the network as a platform.

Senior Researcher, Networks

Senior Researcher, Network management and automation

Principal Researcher, Networks

Senior Researcher, Cloud systems and platforms

Senior Engineer-Research, AI

Hashtags
AI as a service

Senior Researcher, Networks

Senior Researcher, Network management and automation

Principal Researcher, Networks

Senior Researcher, Cloud systems and platforms

Senior Engineer-Research, AI

Senior Researcher, Networks

Contributor (+4)

Senior Researcher, Network management and automation

Principal Researcher, Networks

Senior Researcher, Cloud systems and platforms

Senior Engineer-Research, AI

The introduction of 5G Advanced is marked by the extensive integration of artificial intelligence (AI). AI is currently used within the network to assist with complex tasks in various domains. For example, the network data analytics functions (NWDAF) in the core network (CN) can be trained with machine learning (ML) prediction models to accurately forecast user equipment (UE) mobility patterns. Radio access network (RAN) functions can harness AI to automate Multiple input, multiple output (MIMO) energy management.

With the journey toward 6G, we envision the network evolving into an AI-native platform that not only utilizes AI internally but also exposes its AI capabilities to support a wide range of applications. We refer to this concept as Artificial Intelligence as a Service (AIaaS).


The idea is that the network, functioning as an AIaaS provider, can offer pre-built AI models, datasets, algorithms, and tools that applications can readily access through application programming interfaces (APIs). The primary advantage of AIaaS lies in enabling applications to leverage AI functionalities without app developers having to construct and manage their own AI infrastructure. Consequently, the network transforms into a platform that fosters innovation for a variety of use cases.

But what exactly is AIaaS, and how can the mobile network play a crucial role in supporting the needs of applications?

To shed light on this topic, we’ll first introduce the role of AIaaS in the future 6G network. Then, we’ll present a use case application for AIaaS. Finally, we’ll examine the relationship between AIaaS and API aggregators and conclude by discussing future research directions.

AI as a Service – An overview

Before explaining the AIaaS concept, it’s crucial to clarify a fundamental component of it: machine learning operations (MLOps).

MLOps refers to the processes and functionalities for building, deploying, operationalizing, and observing ML-based systems. Typically, support for MLOps is provided through a set of MLOps tools that can be implemented within the communications service provider (CSP) domain. (See Figure 1).

Figure 1: Architecture of a future 6G network

Figure 1: Architecture of a future 6G network with internal exposure of the MLOps toolset(s) to any of the network domains in arrow (1) and external exposure of AIaaS to applications in arrow (2)

The MLOps toolset(s) can be internally exposed to any of the network domains within the CSP domain. Arrow 1 is an example of exposure to the RAN to support the training of AI-enabled network functions. The toolset(s) can also be made available to applications running on top of the CSP network. We refer to the latter as MLOps-aaS.

We view AIaaS as a superset of MLOps-aaS. AIaaS includes all MLOps-aaS APIs plus additional APIs. These APIs enhance the network's exposed capabilities, further reinforcing the vision of the mobile network as a platform.


AIaaS APIs are realized by API orchestration of the MLOps toolset (1) and of network functionalities in network domains such as RAN, CN, and Management. The AIaaS box in Figure 1 includes not only API orchestration but also other functionalities such as authentication and access management. This way, the AIaaS box realizes higher-level AI services that can be exposed to the application domain.

It's essential to note that an application can utilize MLOps-aaS also directly from a hyperscale cloud provider (HCP) — (3) in Figure 1. It might even use a combination of (2) and (3). Three primary reasons for an application to opt for (2) instead of (3) are as follows:

  1. The CSP can offer specific services (APIs) using network insights that the HCP cannot provide.
  2. There could be trust issues between the application and the HCP, such as data privacy concerns.
  3. The CSP itself might be an HCP, making (2) and (3) essentially equivalent.

In the next section, we focus on scenario 1) and outline how the network can be utilized to deliver innovative AI services. We illustrate how the CSP can offer added value compared to HCPs, covering examples like data privacy, device mobility information, and support for quality of service (QoS).

An example application using AI as a service: From robots to cobots

Robots evolving into cobots is one of the main Hexa-X consortium use cases for 6G (download this pdf to read the full description of this and other use case families). The use case centers on integrating collaborative robots (cobots) to operate alongside human workers. For instance, cobots in a factory setting collaborate to perform specific industrial automation tasks like smart picking and packing. AI significantly enhances the capabilities and efficiency of this use case.

AI algorithms play a pivotal role in enabling robots to perceive and understand their surroundings, thereby improving their adaptability and responsiveness during human interaction. ML techniques empower robots to learn from human feedback and optimize their performance progressively. Furthermore, AI supports predictive maintenance by enabling algorithms to analyze sensor data, identify potential issues, and schedule proactive maintenance.

These functionalities may exist enclosed within the robot application. However, it raises the question if such robots can benefit from AI services provided and exposed by the network. Below, we provide two examples based on Figure 2.

Figure 2: AIaaS for the cobots use case

Figure 2: AIaaS for the cobots use case

  1. The AIaaS APIs handle requests that neither the existing HCP offerings nor the cobot itself can address. For instance, the cobot application requests the creation and execution of a custom AI/ML model tailored to its specific use case and use the inference through the exposed API. Such a model may be designed to predict the cobot’s location, which may rely on a combination of data sources, including sensing data from within the CSP domain derived from the network’s antennas for sensing; GPS or other location sensor information from the cobot’s application; or, cobot UE mobility events.
  2. The AIaaS APIs receive the application's requirements and fulfill them by triggering relevant functions within the CSP domain. The specific functions invoked vary based on the requirements but may encompass computational offload or sensing capabilities or activate an orchestrator to provision additional resources or reconfigure specific network functions. For example, QoS guarantees for bandwidth and latency may trigger the setup of new bearers or the allocation of RAN and CN resources at the network edge to enable low-latency connections.

Generic aspects of AI as a Service

Generalizing from the robot evolving into cobots use case, we envision a list of API families characterizing AIaaS.

Figure 3 - API families for AIaaS

Figure 3 - API families for AIaaS

The MLOps-aaS API family offers a set of tools and services to manage the complete life cycle of machine learning models. For instance, it provides the necessary tools and environment for training, deployment, and monitoring of models.

The exposure of the network services API family includes functionalities such as sensing and compute offload.

The exposure of the network data API family equips applications with diverse network-related data and analytics, including mobility events describing movement patterns, network load levels aiding adaptation to changing conditions, performance insights, and energy consumption details.

The lifecycle management of the customer models API family oversees the full lifecycle of customer-specific models. This includes training application models using a combination of network and application data and exposing APIs for model inference.

The QoS APIs may be used in combination with other API families. Take the example where an AI/ML model of an application requires a certain maximum latency during inference. In such a scenario, the application may package a model within a container, request compute capability from the network service API family to run the container on, and then request a maximum latency through the QoS API.

Offering AIaaS necessitates highly flexible APIs. API orchestration plays a pivotal role in integrating existing network APIs into AIaaS APIs. This orchestration involves managing and coordinating multiple APIs to collaboratively achieve specific objectives efficiently. It requires integration of functionalities and data from different APIs, often sourced from various providers or sub-systems, like RAN or CN, to create a unified and cohesive solution. We envision the execution to be automated and triggered by application-originating requests.

AI as a service in the context of Ericsson’s global network platform

As explained above, a single CSP can offer tailored AIaaS APIs to meet its customers' specific needs. This level of customization can also be extended to CSP federations, enabling sharing of the same set of APIs across multiple CSPs. Integration of services from CSPs, HCPs, and potentially other actors establishes a global network platform. This approach consolidates various services into a comprehensive and unified network platform.

The Global Network Platform is depicted in Figure 3. Arrow (4) represents the aggregated platform exposure of AIaaS from CSPs, HCPs, and Ericsson, where Ericsson serves as the Global Network Platform provider.

Figure 4: Future 6G network architecture, AIaaS, and GNP

Figure 4: Future 6G network architecture, AIaaS, and GNP

The integration of AIaaS into Global Network Platform, alongside the evolution of future 6G network platforms, has the potential to advance the deployment and utilization of AI services in a network-centric ecosystem. By effectively integrating aggregators and enhancing network capabilities, a robust and scalable infrastructure can seamlessly offer AIaaS. This will pave the way for transformative use cases and unlock new opportunities for collaboration and innovation in the AI landscape.

Summary

The advent of 5G Advanced is characterized by the widespread adoption of AI. In the journey toward 6G, we envision the network evolving into an AI-native platform that not only utilizes AI internally but also exposes its AI capabilities to support a wide range of applications. We refer to this concept as AIaaS. AIaaS encompasses a suite of API families, including MLOps, exposure of network services and data, life-cycle management of customer models, and QoS. Orchestration of these APIs ensures effective implementation and integration of AI capabilities within the network ecosystem.

AIaaS envisions the network as a provider of pre-built AI models, datasets, algorithms, and tools accessible to applications through user-friendly APIs. The additional value brought by a CSP includes data privacy, device mobility information, and QoS support. By eliminating the need for applications to establish and maintain their AI infrastructure, AIaaS positions the mobile network as a catalyst for innovative use cases.

The incorporation of AIaaS within the global network platform presents an opportunity to propel AI service deployment and utilization, fostering a dynamic environment for innovation and collaboration. Such an environment holds promise for unlocking novel use cases where AI applications benefit from using the network as a platform.

Learn more

Read the white paper 5G Advanced: Evolution towards 6G

Learn more about our research journey to 6G

Learn more about Ericsson’s Global Network Platform architecture

Explore telecom AI

Explore AI in networks

The Ericsson Blog

Like what you’re reading? Please sign up for email updates on your favorite topics.

Subscribe now

At the Ericsson Blog, we provide insight to make complex ideas on technology, innovation and business simple.