AI trong 311 và Chatbots Dịch vụ Công dân

311 is the non-emergency phone number and app many North American cities use for service requests and information; similar systems exist elsewhere under other names. AI appears in three places. First, intake and routing. Requests arrive by phone, app, web form, text and social media, described in residents' own words: 'water coming out of the road on 5th.' Language classifiers map these to a service type, such as water main break versus clogged storm drain, extract the location, and detect duplicates when many people report the same problem. Routing matters because a misclassified ticket can sit in the wrong department's queue for days. Many cities use the Open311 standard, which gives service requests a common structure that tools can connect to. Second, answering questions. Early chatbots were decision trees with scripted replies. Newer ones use large language models, ideally grounded with retrieval-augmented generation so answers come from official pages, schedules and codes. Grounding reduces errors but does not eliminate them. New York City's MyCity chatbot, built on Microsoft's Azure AI services, was reported by The Markup in 2024 to tell business owners things that contradicted the law, including on workers' tips and housing discrimination. The city kept it online at the time with disclaimers, which drew criticism. Third, access. Cities serve residents who speak many languages, have disabilities, or lack reliable internet. Machine translation and speech tools can widen access, but quality varies widely across languages, and mistranslated legal or eligibility information can cause real harm. A common misconception is that a chatbot automatically reduces workload. If it answers badly, residents call anyway, now frustrated. Successful deployments keep the bot's scope narrow, cite sources, make escalation to a person easy, and track resolution rates rather than chat volume.