MEETING RECAP: AI
Council members and guests discussed AI as it applies to the telecom industry at DoCoMo’s offices.
Artificial General Intelligence is what you see most often in science fiction films, like Ex Machina, with robots that are generally skilled and capable of a wide range of intelligent action, much like a human. More realistic in the current time frame is Narrow AI, where the AI is far from as capable as a human, but on a narrowly specified task, the AI can perform as well as, or better than a human.
What distinguishes Narrow AI from previous “expert systems” is that the AI is capable of Machine Learning, and can iteratively improve itself towards the human-stated goal. In Expert Systems, the intuition must come all from the human, and the machine blindly follows the decision tree the human programmed, albeit very fast. In AI, the system will discover its own patterns in the data, will extract its own learning, and may even execute its own A/B testing. If a human defines a metric as a goal, the AI will deliver refinements toward the goal.
In telecoms, AI is primarily seeing use in:
- Fraud detection
- Call center support systems
- Chat bots for direct customer support
- Real time pricing and offers
- Churn pre-detection and reduction
- Network performance and outage prevention (SDN)
One of the key take-aways from panelists Padraig Stapleton, VP Engineering at Argyle Data (@ArgyleData), and Patrick Ehlen, Chief Scientist at Loop AI Labs (@LoopAILabs), was that Narrow AI feeds on data. It needs data to become intelligent, to use for regression analysis and test runs. Telecom is an industry rich in data, and is in a unique position to feed the AI process. Unlike self-driving, where Uber, Tesla, Google, and others need to collect immense amounts of data to make their systems work, telecom ALREADY has this data, and just needs to point the AI at the haystack and set the objective as “needles”.