The Telecom Regulatory Authority of India (TRAI) has released a consultation paper, “Leveraging Artificial In­te­lligence (AI) and Big Data (BD) in Tele­communication Sector”. This extensi­ve­ly details many of the issues and challen­ges in­volved in the induction of AI and BD in the telecom sector, and TRAI has invited comments from stakeholders on these.

The telecom sector is already using AI. The sector generates huge data, and metadata, and it will generate even more as 5G, satellite-based internet, internet of things (IoT) and 6G in the future, changing the telecom landscape. AI provides the key to­ols for BD analysis and for achieving the scalability and security demanded by 5G networks and IoT.

There are plenty of risks and possible constraints to AI induction. Data quality, data security and data privacy are large con­cerns, for example. Large investments will be required and standards must be es­tablished for interoperability.

Managing possible biases caused by biases embedded in data or by biased algorithms will lead to another set of challenges. AI that causes regulatory breaches or legal br­eaches is another concern – this is one reason why one cannot blindly im­port an al­gorithm that works in another regulatory environment. Finally, security is a big concern.

Setting standards for interoperability and setting up an ecosystem for testing AI deployments are some other challenges. Policy must ensure that there is a smooth roll-out of the necessary AI infrastructure and a suitable framework is created for managing all this.

There’s a huge opportunity. Anand Bhaskar, managing director, service pro­vi­der business, Cisco India & SAARC, says, “Today, critical assets are digitised and connected through applications powered by AI, ML (machine learning), IoT and big data, and usage is expanding. With big data analytics, telecom service providers can perform network optimisation, netwo­rk expansion, capacity planning, investme­nt planning, operational analysis and data monetisation. According to NASSCOM, AI has the potential to add $450 billion-$500 billion to India’s GDP by 2025, acc­ounting for 10 per cent of the country’s $5 trillion vision.”

Telecom sector AI deployments are al­ready common in network operations, cus­­tomer services and infrastructure operations. AI adoption has improved network reliability and delivered better customer ex­periences while optimising costs, and hel­ped with strategic decisions. Apart fr­om meeting their own needs, telecom service providers can also provide enabling functions for other industries and for the economy in general, by leveraging AI to analyse BD for insights.

The policy setting for this space will be crucial and policy-makers must find ways to enable AI in a flexible and future-proof fashion while navigating all risks. Among the points to consider is the structure of AI regulation. Should there be a specific au­thority – perhaps even a ministry, or a statutory independent authority setting standards across the entire policy environment? Should different ministries with different needs regulate AI narrowly as it pertains to specific areas of concern? Both methods have been tried in the First World, and there is a case for adopting either.

India also has special legislative concerns with regard to BD. India does not have a personal data protection law, alth­ou­gh this is defined as a legal fundamental right. Proposed legislation has been hanging fire since 2017. After lacunae were discovered in the draft legislation, the government is trying to formulate a new draft Personal Data Protection Bill. In the ab­sen­ce of such legislation, there could be greater risks to handling BD, including the possibility of misuse, as well as the likely possibility that algorithms deployed now will have to be reworked, or completely dis­carded to comply with future legislation.

The stakeholder response to the consultation paper has been quite enthusiastic.  There’s no question that AI is essential for network management as 5G rolls out. Kai­lem Anderson, vice-president, portfolio and engineering, Blue Planet (a division of Ciena), says, “There’s an increasing need for AI and ML-powered automated networks to deliver the cloud-like network ex­periences customers now demand. It is not possible to manage 5G dynamic networks with distributed workloads, with a legacy network architecture approach. AI and ML can help quickly identify the root cause of service disruptions and ensure minimal impact to customers.”

Industry bodies such as the Cellular Op­­­e­rators Association of India (COAI) and the GSM Association (GSMA) have also responded positively. Lt General Dr S.P. Kochhar, director general, COAI, poi­nts out, “Critical decisions are based on insights churned out through data analysis. Leveraging new-age technology such as AI to analyse big data will improve competition, business processes, operations and user experience. AI use cases include AI-enabled chatbots, security detection to stop fraudulent activities, user-specific re­commendations, real-time analytics and in-network optimisation. AI coupled with 5G has the potential to revolutionise data management strategies through applications such as edge computing and network slicing. Data storage will be virtualised and distributed. Big data analysis will help ide­ntify patterns and generate insights.”

Jeanette Whyte, head of public policy for Asia Pacific at GSMA, says, “The GSMA welcomes the recently released TRAI consultation paper. It raises significant issues such as understanding the true meaning of AI, the challenges involved in the process, such as the threat to privacy, and how these can be addressed, the fra­me­work for AI ethics, an appropriate governance structure and its principles, and the need for regulatory sandboxes. The­se are important subjects needed to ensure that big data and AI can prosper in a trustworthy and sustainable way.”

There are a multitude of use cases for AI. It can transform the telecom industry by using intelligent automated systems to design, deploy, maintain and manage networks, including proactively securing networks. It can optimise resources, improving service quality and customer satisfaction.  AI can improve network design, enhance di­gital connectivity inside buildings, and optimise traffic and spectrum management. It can improve network security using algorithms to detect anomalies in the network.

In terms of customer-centric functio­ns, AI can power virtual assistants and assi­st in handling queries or complaints, or choosing tariff plans. AI can also look at data con­su­mption and phone use patterns and tariff plans and other variables to predict levels of churn and possible pain-points, and th­us help with customer retention.

Telecom networks generate and store vast data, which ranges across call details, network data, customer data. The data can range from location data to social media and net usage patterns, to health records, banking and credit card use, job applications, searches, citizen-government interactions, etc. It is possible to build a 360-de­g­ree picture of individual lives by analysing the data people generate.

Globally, the average data consumption per smartphone now exceeds 10 GB per mo­nth, and is forecasted to reach 35 GB by the end of 2026. It is also forecasted that in 2026, 5G networks will carry more than half of the world’s smartphone traffic.

AI analysis of this data raises large privacy concerns, but it is also of great commercial interest to multiple sectors. An­o­n­ymising such data and using it responsibly to gain insights without brea­ching privacy will be one of the critical challenges for AI. Cu­rrent solutions involve decrypting en­crypted data to process it, but this can ex­pose private data, leading to legal and ethical challenges as well as increased vulnerability to breaches. There are other methods of trying to preserve data privacy inclu­ding differential privacy, secure multi-par­ty computation and homomorphic encryption, and in future, corporates will probably move to a mix of these.

Telecom data analysis can also offer insights to other industries. The GSMA has reported on the usage of mobile big data (MBD) across countries to support the he­al­th­care response to Covid-19. Dur­ing the crisis, institutions and governments discovered how MBD could help track, contain and predict the spread of the virus.

There are sector-agnostic and sector-specific risks to the utilisation of AI and BD. The risks that are sector-agnostic may be dealt with under a common regulatory fra­mework, but in cases of sector-specific risks, the sector regulator may be required to develop a framework for specific risk mi­­tigation.  Various countries have different approaches and initiatives to deal with this aspect. In India, NITI Aayog has discussed this situation in its paper, “App­roach Document for India Part 1 – Prin­ciples for Responsible AI”.

The sector-agnostic risks include low quality data, data biases, issues related to da­ta security and privacy, intrusion identification and tracking and profiling of individuals or groups. For example, models may be trained on non-representative data, algorithms may be inaccurate, or biased, or there may be unethical use of AI.

In sector-specific terms, there can be risks of regulatory or legal non-complian­ce. There can also be market-related risks in specific sectors where AI usage can re­sult in herding behaviour where several or all players use algorithms that lead to similar conclusions and strategies. There can also be a lack of clarity about allocations of liability in case of damages inc­u­rred when “black box” algorithms are us­ed, since there is no human understanding of how the algorithm works.

AI adoption also involves incurring high investment costs to develop the computing infrastructure required for model development, training and integration of AI-based services. Developing the necessary infrastructure will require good policy as well as investment. NITI Aayog has pro­posed to establish cloud-based AI-specific infrastructure to support varieties of AI workloads in an approach paper, AIRAWAT (AI Research, Analytics and knoWledge Assimilation plaTform).

Skilling workforces to cope with the new demands of AI will also require tho­u­g­ht. In India, the Minis­try of Electronics and Information Technology is working on standardisation interoperability and compatibility, with its Commi­ttee on Platforms and Data for AI recommending the development of a generalised meta-data standard to enable the integration of resources such as data, tools and literature from multiple owners. The Depart­ment of Telecommu­nications has also formed an AI standardisation committee to develop various interface standards and India’s AI stack. This stack is to be structured across all sectors ensuring protection of data, data federation, data minimisation, open algorithm frameworks, defined data structures, interfaces, etc. Regulatory sandboxes have also been proposed to allow for testing and experi­mentation. Further clarity will be required in licence agreements to establish clear ow­nership of call detail records, calling patterns, location data, data usage, etc. The ow­nership rights, authority to use, transfer and delete this data are all ambiguous.

The paper lays out all these challenges and issues quite clearly. Once TRAI has processed the responses and incorporated suggestions, India could have a good road­map on how to proceed. There’s no doubt that AI will be leveraged to manage the big data generated by telecom service pro­viders. This is a huge opportunity to tur­bo­charge the economy. Setting a sound policy framework will be essential if it is to be fully realised.

Devangshu Datta