Using chatbots against voice spam: Analyzing Lenny’s effectiveness

A new countermeasure recently appeared to fight back against unwanted phone calls (such as, telemarketing, survey or scam calls), which consists in connecting back the telemarketer with a phone bot ("robocallee") which mimics a real persona. Lenny is such a bot (a computer program) which plays a set of pre-recorded voice messages to interact with the spammers. Although not based on any sophisticated artificial intelligence, Lenny is surprisingly effective in keeping the conversation going for tens of minutes. Moreover, it is clearly recognized as a bot in only 5% of the calls recorded in our dataset. In this paper, we try to understand why Lenny is so successful in dealing with spam calls. To this end, we analyze the recorded conversations of Lenny with various types of spammers. Among 487 publicly available call recordings, we select 200 calls and transcribe them using a commercial service. With this dataset, we first explore the spam ecosystem captured by this chatbot, presenting several statistics on Lenny's interaction with spammers. Then, we use conversation analysis to understand how Lenny is adjusted with the sequential context of such spam calls, keeping a natural flow of conversation. Finally, we discuss a range of research and design issues to gain a better understanding of chatbot conversations and to improve their effciency.