The use of artificial intelligence (AI) was a hot topic at TM Forum’s Technical Action Week in Lisbon. The fast-approaching introduction of new technologies, in particular software defined networking (SDN) and 5G, are driving business transformation at an ever-increasing pace. There is a recognition that without automation, operators will struggle to provide these new offerings, let alone utilize them to their full potential.
“There is no place for Humans in Operational Processes” – Lester Thomas (Vodafone)
AI is a key part of the path towards ever-greater automation. The Forum recognized this in 2017, setting up a new initiative under former BT and Brocade exec Brian Levy. This in turn led to a “state of the industry” report on AI and Telecom in December 2017, and the launch of a new post-OSS/BSS vision under the moniker of the Open Digital Architecture (ODA).
The applications of AI for customer-centric use cases are already reasonably well defined. Primarily this is due to the work done in other sectors such as finance and retail being directly translatable to the customer layers of a communication service provider’s (CSP’s) business.
Deloitte ran an excellent session on the Tuesday morning that explored how AI is already being utilized to improve aspects like customer support through the use of NLP and chatbots. Wolfgang Maaß, Scientific Director at DFKI, highlighted the parallels with ‘smart farming’ and that the same techniques could be used in a telecoms environment.
AI in network and service management
Aria Networks led the “AI, Network & Service Management” session on Tuesday afternoon in which we discussed the current state of AI in the area together with presentations from Cenx, KDDI and TM Forum. Topics discussed included:
- Network and service optimization through the use of AI/machine learning
- What does ‘good’ look like? What do we hope to achieve with AI?
- Importance of a holistic view of networks and services
- How Machine Learning is just one aspect within the overall field of AI
- The proposed Analytics and Optimization API that will help rationalize OSS and automate processes
KDDI presented work on how deep neural networks can be used to determine the remediation actions to take in a self-healing network based on inputs from orchestration and alarm systems. Cenx talked about how service assurance can be augmented with machine learning and TM Forum’s Dave Milham spoke about the Forum’s current work in the area and its interaction with projects such as zero-touch orchestration, operations & management (ZOOM).
The maturity of customer-centric AI is reflected by the maturity of TM Forum’s Open Digital Architecture (ODA). At Action Week, discussions focused on the key principles and functional architecture of the ODA though almost all cases were driven by customer and business-layer work (for example eTOM and SID from the Frameworx project) rather than network-layer operations.
In truth, the ‘bottom’ (network centric) layers of the ODA are still early in their development, with API coverage/offerings limited so far.
AI and TM Forum’s ZOOM
Looked at from an AI perspective, the work the ZOOM project is doing on intent/policy based interaction mechanisms is a great step in the right direction. It should enable AI systems to integrate closely into multiple ODA components and effect changes without the vendor/domain-specific hurdles that must currently be overcome.
However there still seems to be some confusion on the exact meaning and distinction of ‘intent’ and ‘policy’. Some more clarity and consensus is required if this work is to achieve its potential.
The interaction of ‘closed control loops’ seemed to concern many at Action Week (especially when AI is introduced). The first step to mitigating these concerns is the discussion of potential reference architectures for these new ‘decision making boxes’ (and closed control loops), something the ETSI ENI ISG is already looking at.
A word of caution: deep water ahead…
It was notable at Action Week how much airtime in the AI stream was taken up with discussing data lakes.
There’s no doubt that the accumulation and processing of truly vast amounts of raw data is a significant technical challenge. However, without a clear understanding of how that data (or the resulting analysis) feeds the processes that act on it, we’re not even addressing half of the real issue. A bar chart isn’t a better designed network. A trendline isn’t an optimized capital expenditure plan.
It’s been highlighted, and is worth emphasizing, that the uses of AI and machine learning in telecom operations must be driven by business need first, and technology second. A data lake can’t be a destination in itself. In fact, for many applications – such as network optimization or failure analysis – it isn’t even necessary at all.
TM Forum’s Action Week was a fantastic opportunity to contribute in a collaborative spirit to moving telecom a step closer to more intelligent, automated and dynamic operations.