Considerations for Effective AI in Mobile Networks

Print Friendly, PDF & Email

In this special guest feature, Tom Luke, VP at Tutela, explores the different sources of data on offer to mobile network operators, and why it’s important for AI to be fed with lawful, ethical and robust big data in order to be effective. Tutela is an independent crowdsourced data company with a global panel of over 300 million smartphone users. Tom is responsible for sales, marketing and partnerships at the company, and has previously held roles at Hewlett Packard and Cisco. He holds a BSc in Computer Science and E-Business from Loughborough University in the UK.

In 2018, Gartner analysts declared that AI is “almost a definition of hype”. Its popularity can be shown in the steady and sustained growth in searches about AI in Google Trends, in the ever-rising number of articles published about its potential, and in the number of new companies offering AI-powered solutions.

It’s not a trend that has passed the telecoms industry by either, with IDC forecasting that over 60% of telecoms businesses worldwide are investing in new AI technology. One recently publicized example came from Elisa in Finland, which is using AI in a network operators center. This is likely just one of many potential use cases, and we could soon see AI applied to customer service at mobile network operators, or even to sophisticated network issue detection and optimization, similar to the fraud detection systems we see in banks that can spot anomalous activity.

However, many people are wary of AI – and for good reason. AI done wrong has led to tragic consequences in the past. Sensor malfunctions and corresponding autopilot “corrections” were likely a significant factor in the crashes of two Boeing 737 MAX airliners, whilst the first fatal crash in the US of a Tesla in ‘Autopilot’ mode was due to sensors failing to detect a hazard, meaning the AI was unable to respond. These two terrible events have one thing in common – bad data.

Mobile networks may not seem as critical as vehicle autopilots, but given the dependency of users like first responders and 911 systems on cellular networks, a similar level of care needs to be taken. However, the potential benefits from AI – when done right – are hard to overstate. So how do mobile network operators ensure they have the best possible chance for success with AI?

It comes down to ensuring you can feed your AI system with the best data available. What do we mean by best? It’s a subject of ongoing research, but comes down to the robustness of the system. A robust system needs to be certain of the completeness of its data, and also have redundancy built in. For mobile network operators, this means relying on more than just the network insights they have internally. Specifically, they need to find a way to validate internal data with real-time external information that is representative and free from bias.

Bias in mobile network data often comes down to test methodology. User-instigated speed tests, for example, provide valuable insights for consumers about their mobile performance. However, if used as a tool for measuring network performance to inform AI decisions, there is the potential for bias to affect the results – people tend to instigate tests either because their network is performing particularly well, or because it is performing badly, and seldom because they’re simply curious.

Problems also exist with field-tested measurements. Professionally collected, high-quality measurements are the most precise and detailed source of network intelligence. However, they’re extremely expensive to collect, making it hard to get 24/7 real-time information. Drive-testing is inherently confined to set locations and times, meaning the depth of information may be extremely good, but the breadth is insufficient for real-time AI applications.

Instead, mobile network operators looking to utilize AI effectively for their network need to look for a data source that is able to collect enough core measurements (such as download speeds, latency, and packet loss), round the clock and without requiring a user to decide when to run the test. The good news is that, with such a high penetration of smartphones, there is now an army of “citizen sensors” out there, and an SDK embedded into the background of applications can collect these key metrics, without bias and without impacting the user. Data streams like this provide the broadest picture of network quality regardless of time, location or user.

If a single AI system at Elisa in Finland could lead to a 50% reduction in 4G customer complaints, it’s hard to imagine the kind of efficiency and improvements that more broadly deployed AIs could have on the mobile network industry. With the right AI, fed the most representative information, mobile network operators have the best chance of being able to understand their whole network and make the right decisions for network planning and optimization. The result? Better mobile internet for consumers, reduced churn for operators, and an overall improvement of the quality of mobile networks for everyone.

Sign up for the free insideBIGDATA newsletter.

Speak Your Mind