Building cognitive networks and human trust in AI – Ericsson
An average 49 percent of artificial intelligence (AI) and analytics decision makers say they will complete their transformation journey to AI this year, according to our recent Ericsson AI Industry Lab report. As the sector with the highest AI maturity, telecom and tech enterprises will be key to driving this transformation, not just within their own organizations, but also across other sectors over coming decades.
Cognitive networks: Transforming complexities into opportunities
The ultimate goal of intent-based cognitive networks is to realize the zero-touch operation paradigm, a vision that requires artificial intelligence that is on par with human capability to reason and decide within complex dependencies and broad domains.
At Ericsson, we are leading by example. Last year, we opened our most advanced 5G smart factory yet, this time in Lewisville, Texas, demonstrating the potential of 5G in leveraging real time operational data and enabling intelligent automation. We are also driving the industry momentum toward cognitive networks by developing proof of concepts to demonstrate the value add of AI in network operations.
Already with our cognitive operations support system (OSS) proof of concept with T-Mobile US, we could demonstrate problem discovery and resolution efficiency gains by more than 100 percent and reducing the number of manual operations by 70 percent. Here, by deploying AI across the OSS platform, we have enabled the network to automatically optimize VoLTE user experience and service quality based on our customer’s specific business intent.
Beyond operational automation, cognitive network technologies will also be pivotal in unlocking advanced 5G services in new enterprise sectors, creating a foundation for service providers to grow business beyond connectivity. In an earlier post, I laid out how service providers are in pole position to monetize the emerging 5G industrial cloud era, taking the lion’s share of USD 700 billion revenue opportunities within those ecosystems over the next ten years. As a critical step in making that a reality, we piloted with ENCQOR, a Canadian government funded 5G innovation ecosystem, intent-based 5G network slice service level agreement (SLA) assurance. Here, AI and explainability techniques have proved effective in reducing network slice operational procedures by as much as 35 percent, as well as increasing the overall quality of the network slice service. In future, this will serve as the basis for service providers to compete for new enterprise opportunities at the edge.
AI that delivers trustworthy, robust, and resilient network operations
Mobile networks need to deliver mission critical use cases where reliability, availability and resilience become paramount. For example, when Hurricane Florence hit the U.S. in 2018, we worked together with Verizon to pilot live zero-touch operations for the first time. Cognitive technologies in the network made it possible for Verizon’s staff to leave for safety, while ensuring that critical operational services were maintained.
Over this past year, our networks have stood up to an even greater stress test yet. Now more than ever, people rely on mobile networks to support their daily lives, access essential services, go to work and manage their businesses. However, 5G has the potential to bring an even greater era of digital transformation to our societies and workplaces, whereby cognitive network technologies will be the only way to deliver on new opportunities presented by the exponential rise in data traffic and multiplicity of new use cases at the edge. As this transformation accelerates, the economic value of the data, goods and services that rely on the integrity of mobile networks will rise with it.
This is, critically, why the industry must continue to build human trust in AI – addressing all characteristics spanning from explainability and human oversight (at first) to security and built-in safety mechanisms.
According to our AI Industry Lab report, key monetization markets such as manufacturing, healthcare, transport and logistics are still at a very early stage in their AI maturity journey. We know that, for many in the industry, the deployment of AI presents a cultural challenge more than it does an organizational or technological challenge. The same report estimates that as many as 87 percent of industry decision makers have already found this to be the case. Capturing digitalization opportunities within these sectors will require digital infrastructure that is not only intelligent, robust and resilient; but also secure, safe and trustworthy.
This is a shared interest across our societies and national economies, one where service providers, vendors and governments all play a role. Together, we must advocate and commit to the development of standardized ethics frameworks, such as the EU Ethics Guidelines for Trustworthy AI which Ericsson has adopted as part of our own approach to AI, ensuring that our AI is lawful, ethical and robust from a technical perspective.
Based on a key design principle of trustworthy AI, we are also taking the next steps by building on this in the field using live network data, such as our 5G slice assurance proof of concept, where we are increasing our research focus on explainable AI techniques. In addition to explainable AI, other critical trustworthy aspects include security, privacy, transparency, bias, human agency oversight and safe AI which can act as a measure of an AI system’s resilience to operate in alignment to a specified intent at all times.
AI in networks today
The transition to a digitalized, data-driven operation both of network and services as well as business processes enable a higher degree of automation, performance, efficiency and insight – and already we see that, through our deployed AI use cases, it’s proving not only to reduce operating- (OPEX) and capital (CAPEX) expenditure, but also have a positive impact on average revenue per user (ARPU) and overall net promoter score (NPS). Today, we are deploying AI in the right places in the network so that our customers remain competitive and deliver value in new business contexts.
We develop our AI solutions both across our worldwide research hubs including D15 AI in Santa Clara and our Global Artificial Intelligence Accelerators (GAIA) in Sweden, US, Canada and India; as well as in the field together with our customers using real network data. In doing so, we leverage our extensive domain expertise and pre-trained algorithms that we adapt to various customer contexts and challenges. We also ensure high-performance solutions by training our algorithms based on our unique access to network node data, always in an ethical and secure way. We have many live use cases in operation today which demonstrate the significant value of such solutions, for example, to prevent revenue leakages, reduce energy consumption, optimize site engineering and optimize RAN deployment through intelligent performance diagnostics.
This is the first step on our three-phase maturity journey to fully autonomous intent-driven networks. Our focus today is to ensure that AI will be embedded in a domain-first approach in the network, as well as our strong continued efforts to improve the intelligence of existing network functionalities. In this phase, a major dependency relies on the data availability and the data-driven infrastructure which is the primary enabler of this phase and the journey to come.
The future of cognitive networks
Following completion of our first phase, our focus will shift to driving scale in intelligent control, reasoning and trustworthiness. Intelligence is widely distributed through the network and most applications within the management and business layers are intent-driven. The latter allows for automatic conflict resolution and predictive control. When it comes to human agency, we will see increased operation delegation, dynamic escalation and agile collaboration with human teams. Advanced development infrastructures are deployed with knowledge lifecycle management (LCM), advanced model LCM, flexible data detection and semantic annotation of data.
Our focus in the final phase will be to ensure that all individual technologies work harmoniously together, enabling a fully autonomous, self-learning and trustworthy network. By this stage, we expect our customers to have achieved fully intent-driven operations, with their networks having the capability to run in a zero-touch fashion. Here, decision making will be managed autonomously by the system through autonomous interference between closed loops where, as a prerequisite for this autonomy, the cognitive technologies will be widely robust and explainable. Development processes of AI modules will also be automated with automated pipeline management and LCM.
As we continue on this technology trajectory, we will remain committed to developing, piloting and deploying the latest AI technologies together with our customers in the field – laying a robust, efficient and intelligent foundation to monetize emerging enterprise and consumer 5G revenue opportunities.
Visit our AI in networks web page.
Read our white paper Explainable AI – how humans can trust AI.
Read our Ericson Technology Review article: Ensuring energy-efficient networks with artificial intelligence.
Visit Ericsson Research: Autonomous networks.
Read our Ericson Technology Review article: Adaptive intent-based networking.
Visit our cognitive technology web page.