3.3. Chatbots characteristics
The total number of AI chatbots reviewed was 17 (Additional file 3). 14 chatbots (88.2%) conversed in English language. 4 chatbots were bilingual and in addition to English, supported French, Italian, Vietnamese and Indian languages. 3 chatbots were also multi-language. Arogya setu and COVIBOT that were implemented in India supported more than 10 common languages in India[7, 24].
13 of 17 chatbots were developed for general population and 4 cases for specific population. Most AI chatbots with general audience were developed by international organizations (such as WHO), governments and large companies (such as IBM and Amazon). For instance CoronaGO and Arogya setu were designed by Indian government to improve care delivery and help to its people during COVID-19 situation[7, 25]. AI chatbots with specific audience were developed for young people (2 cases) [26, 27], elder people (1 cases) [28], and health care providers and their families (1 case) [29].
We categorized AI chatbots based on a) their aims which reflects the chatbot role(s) in response to COVID-19, b) their applications, and c) their design characteristics which include platforms, AI techniques and chatbot-user interaction design.
3.5. Chatbots applications
The reviewed chatbots provided different services to users. We classified the chatbots applications into 8 groups (Table 1), which include 1) Information dissemination and education, 2) Self-assessment and screening, 3) Connect to health centers, 4) Combating misinformation and fake news, 5) Patients Tracking and providing services, 6) Mental health, 7) Exposure monitoring, and 8) Vaccine information and scheduling.
1) Information dissemination and education
The most common application of AI chatbots was Dissemination of information and provision of education related to COVID-19 (13 cases). These chatbots provide the various information such as symptoms, diagnosis of disease, stages of the disease, virus transmission, and availability of healthcare services and how to access them. These chatbots also provide various trainings related to preventive and care practice (such as cleaning hands, wearing face masks).
2) Self-assessment and screening
Nine AI chatbots were utilized as triage and assessment of users during the COVID-19 pandemic. These chatbots ask a series question in order to assess the possibility of COVID-19 infection without going to medical centers. Self-screening chatbots check the COVID-19 symptoms based on the suggested guidelines by governments and approved sources (such as CDC or WHO) and make recommendations if needed. For instance, the Symptoma assess the possibility of a person contracting coronavirus by receiving information such as disease symptoms, age and gender. The results of a study have shown that Symptoma was able to correctly diagnose users with COVID-19 in 96.32% of cases [31].
3) Connect to health centers
Users supporting to communicate with health centers and healthcare providers was another use cases of AI chatbots(n = 8). Some chatbots, in addition to evaluating and triaging people, were able to identify the user’s location by GPS technology. These chatbots provide information about nearest health centers to the users and make an appointment for further investigation[7, 28, 32]. Some AI chatbots were able to communicate online with doctors and health centers so that, the users can benefit from remote diagnostic and consulting services[7, 33]. SanIA, was developed in Spain, created a secure communication channel between the citizen and the health system during the COVID-19 outbreak, to maintain continuity of care by providing advice, including psychological assessment, to patients whenever they want it 24/7[34].
4) Combating misinformation and fake news
7 of 17 AI chatbots were able to combat misinformation and fake news about COVID-19. These chatbots provide users with reliable and accurate information such as up to date statistical information, best practices and health protocols[28, 35]. These chatbots monitor and identify misinformation and rumors about COVID-19 and prevent its further spread by alerting users.
5) Patients tracking and service delivery
In 5 cases of AI chatbots, it was possible to provide remote care services to people with COVID-19. These chatbots were able to continuously and daily track the patient’s physical condition and vital signs. They also answer patients’ questions 24/7 during mandated self-isolation periods and if necessary, help to communicate with health care centers and receive advice and remote care services[24, 32]. For example, COVIBOT helps the patient to take proper medication and remote consultation[24].
6) Mental health
Supporting people’s mental health during the pandemic and providing necessary services was another application of AI chatbots(n = 5). People’s mental and emotional problems such as stress, anxiety and depression were one of the problems that existed during the COVID-19 pandemic. 5 of AI chatbots were able to monitoring and Supporting people’s mental health during the pandemic. These chatbots by providing various mental health services to users, specially to patients in home quarantine, help to reduce the psychological effects of COVID-19 such as anxiety, distress and depression. For example, the Aroha was an AI chatbot that provided services related to the management of anxiety and depression caused by COVID-19 to New Zealand citizens[27].
7) Exposure monitoring
Four chatbots were deployed for monitoring exposure to the Coronavirus and providing notifications. These chatbots provide alerts and necessary information to users by assessing the status of disease spread, as well as identifying high-risk and infected locations. The Arogya Setu works on Bluetooth-based and GPS technologies and tries to determine risk based on the user’s location. It also keeps the user informed in case he/she has crossed paths with the positive COVID-19 case within 6 feet proximity[7].
8) Vaccine information and scheduling
Three chatbots were designed for providing services about COVID-19 vaccines. These chatbots provide accurate and up to date information about COVID-19 vaccines and informing users with their advantages. In addition, they provide information about vaccination centers and support to make an appointment for vaccination. These chatbot help to reduce vaccine hesitancy and increase vaccination. For example, VWise is a free text-based AI chatbot that guide and inform the public about COVID-19 vaccination in the Eastern Mediterranean Region(EMR) [36].
More than 80% of chatbots had multiple application. The information dissemination and education with the Combating misinformation and fake news (7 cases), The Information dissemination and education with the Self-assessment and screening (6 cases), The Information dissemination and education with the connect to health centers (6 cases), and the self-assessment and screening with the connect to health centers (5 cases) had the highest frequency in the combination of categorized applications.
3.6. Chatbots design characteristics
As shown in Fig. 2, reviewed AI chatbots were different in terms of design features including platforms, AI techniques and user-chatbot interaction design, which are discussed below.
3.6.1. Chatbot platforms
The AI chatbots were deployed on different platforms. Nine chatbots were mobile applications and can be installed on smartphones (Android or IOS). Six chatbots were web-based and can be available through various devices such as computers, tablets, and mobile. Three chatbots were deployed on social media including Facebook (2 cases) and Telegram (1 case). One of the chatbots, in addition to web-based, was also deployed on Facebook[27].
3.6.2. AI techniques
According to our objectives, all of the reviewed chatbots were artificial intelligence based. MIRA and Vac Chat, Fact Check chatbots were hybrid type that used Rule-based and artificial intelligence methods[26, 29]. All of AI chatbots were used Natural Language Understanding (NLU) methods to understand user input and provides the appropriate response to the user in a conversational manner. For instance, WashKaro and Chloe were used LSTM and BERT methods to understand free text input[32, 37]. One of chatbots was used Artificial Intelligence Markup Language (AIML), which is an Extensible Markup Language (XML) based method, to understand user input [7]. Some of chatbots (6 cases) were used machine learning methods such as decision tree and deep learning to identify patterns in user input, make decisions, and learn from past conversations. More than 50% of AI chatbots (9 cases) were used NLU platforms including Google Dialogflow (3 cases), Rasa framework (3 cases) and IBM Watson (3 cases). NLU platforms are prebuilt NLU models that used to development AI chatbots[38].
3.6.3. User-chatbot interaction design
AI Chatbots had different user-chatbot interaction design. In terms of input format, 13 chatbots were text-based, 1 was voice-based, and 3 cases had both text and voice options. 13 of 16 text-based chatbots were free text input and 3 cases used predefined and controlled text input. In terms of output format, 8 chatbots were able to offer information in multimedia format, including text, audio, image, and video. 5 cases were text-based and 4 cases were text and audio based. In terms of dialog style, 9 chatbots were designed to be user-initiated. In these chatbots user initiate conversation and chatbot response to user’s asks such as questions related to COVID-19. The dialog style of most preventive chatbots (9 cases) were designed in this way. In 5 cases, chatbots were initiator and the user must be answers to the chatbot's questions about things like their demographic data, medical history and COVID-19 symptoms. For example, Ada Health is a mobile-based diagnostic chatbot that asks a series of questions to users screening[33].