AI spam detection faces challenges with operator data sharing
Indian telecom operators are grappling with the effectiveness and inter-operator sharing of AI-driven spam detection signals. A discussion explored whether online platforms should fall under similar anti-spam regulations.
Telecom operators in India are facing significant hurdles in implementing and sharing AI-generated spam signals, a directive from the Telecom Regulatory Authority of India (TRAI). The core issue lies in the divergence of AI models used by different operators, leading to disagreements on what constitutes spam.
TRAI's February directive mandates operators to share AI-detected spam flags across networks within two hours. However, each operator employs distinct AI systems with varying parameters and inputs. Consequently, a number flagged as spam by one operator might not be recognized as such by another, rendering the shared lists inconsistent.
Industry participants expressed concern that shared signals are treated as definitive rather than probabilistic indicators. A flag from one network, representing only an estimate or a 'guess about a guess,' is then acted upon by another operator. The consensus was that such signals should inform, not dictate, an operator's decision, allowing them to retain judgment based on their own data and customer relationship.
The discussion also touched upon the expanding scope of regulations, questioning whether internet platforms and messaging apps like WhatsApp should be brought under TRAI's purview. Concerns were raised about the potential for data sharing with security agencies to evolve into a surveillance layer, particularly as government bodies are exempt from certain data protection laws like the Digital Personal Data Protection Act.