Key points are not available for this paper at this time.
In low-and middle-income countries where demand for health services outstrips the available supply of skilled labor, advances in information and communication technologies have already been shown to hold promise. While much of the mHealth literature continues to explore mature technologies such as text message and web portals, continual advancement in machine learning opens innovative new areas of exploration for public health practitioners. This paper explores one such possibility, a conversational agent, able to guide users through an HIV counseling and testing session. Using commercially available software (http://api.ai), an agent was designed and built according to the Center for Disease Control's guidelines for the provision of HIV counseling and testing in a non-clinical setting. The agent was linked to the Telegram chat client (http://telegram.org) and 10 testers were invited to participate in a simulated HIV counseling interaction. Six testers found that talking to the agent felt natural, and equivalent to chatting to a human. Seven said they would feel comfortable taking a real HIV test with the agent. Key concerns with the current agent were the use of overly formal language, the speed at which the agent responded (too fast) and the agent either misunderstanding or not understanding the tester. Positive sentiment towards the agent included the fact that testers felt like the session was more private and anonymous, and avoided the need for them to visit a public health facility and stand in a long queue to get tested.
Heerden et al. (Sun,) studied this question.