Public Willingness to Get Vaccinated Against COVID-19: How AI-Developed Vaccines Can Affect Acceptance

Vaccines for COVID-19 are currently under clinical trials. These vaccines are crucial for eradicating the novel coronavirus. Despite the potential, there exist conspiracies related to vaccines online, which can lead to vaccination hesitancy and, thus, a longer-standing pandemic. We used a between-subjects study design (N=572 adults in the US and UK) to understand the public willingness towards vaccination against the novel coronavirus under various circumstances. Our survey findings suggest that people are more reluctant to vaccinate their children compared to themselves. Explicitly stating the high effectiveness of the vaccine against COVID-19 led to an increase in vaccine acceptance. Interestingly, our results do not indicate any meaningful variance due to the use of artificial intelligence (AI) in developing vaccines, if these systems are described to be in use alongside human researchers. We discuss the public's expectation of local governments in assuring the safety and effectiveness of a future COVID-19 vaccine.

READ FULL TEXT VIEW PDF
POST COMMENT

Comments

There are no comments yet.

Authors

page 22

page 23

03/01/2021

Reflections on the Clinical Acceptance of Artificial Intelligence

In this chapter, we reflect on the use of Artificial Intelligence (AI) a...
04/26/2020

Power of Artificial Intelligence to Diagnose and Prevent Further COVID-19 Outbreak: A Short Communication

Novel coronavirus-19 (2019-nCoV or COVID-19) is by far the most dangerou...
09/02/2021

COVID-19 Vaccine Hesitancy and Information Diffusion: An Agent-based Modeling Approach

Despite the unprecedented success in the rapid development of several ef...
03/08/2021

A Framework for Enabling Safe and Resilient Food Factories for the Public Feeding Programs

Public feeding programs continue to be a major source of nutrition to a ...
03/28/2021

Acceptance of COVID-19 Vaccine and Its Determinants in Bangladesh

Background: Bangladesh govt. launched a nationwide vaccination drive aga...
05/04/2021

Towards Accountability in the Use of Artificial Intelligence for Public Administrations

We argue that the phenomena of distributed responsibility, induced accep...
This week in AI

Get the week's most popular data science and artificial intelligence research sent straight to your inbox every Saturday.

1 Introduction

The global race for a COVID-19 vaccine is on [6]. Researchers are developing vaccines against the novel coronavirus at a record speed [9], and many of these drugs are currently under trial. In the current pandemic, however, the hardships are not exclusive to vaccine development but also embrace its worldwide manufacturing, distribution [56], and acceptance.

Alarmingly, protests against unpopular public health policies, such as national-wide lockdowns, have been fueling anti-vaccine movements [5]. Misinformation, which has led to vaccination hesitancy during the Ebola epidemic [64], is also predominant in the current pandemic [18, 70]. Vaccine refusal against COVID-19 could lead to a longer-standing pandemic and deaths that could be prevented by widespread vaccination. These public policies, which often conflict with human rights, are necessary to the containment of the disease [12] and can be justified by the disease’s threats to public health, for instance, under the Siracusa Principles. These movements must not be an obstacle in controlling the novel coronavirus, given the importance of future vaccination in the disease’s eradication [15].

Epidemiological models have indicated that a vaccine is the only solution that could stop the virus [34]. Human trials for coronavirus vaccines have begun worldwide, and scientists expect vaccines will be made available to the population in mid or late 2021 [58].

Adopting artificial intelligence (AI) can shorten the vaccine development process, either by reviewing possible matches for already available drugs in the market and repurposing it for coronavirus or by assisting in the development of a new one [78]. However, AI systems have also raised various moral and legal questions regarding the assignment of responsibility for their actions [21, 8]. Deploying these systems in life-or-death scenarios like medicine could raise the stakes and lead to more complicated gaps. Therefore, we raise the question of whether the inclusion of AI could modify public acceptance of a COVID-19 vaccine.

Early results indicate that 64% of Americans would get a COVID-19 vaccine in May 2020 [10]. Understanding the public’s willingness on this subject is critical, as vaccines are often the ultimate solution to lessen the burden of disease [15, 69]. It could also allow researchers and policymakers to better prepare for vaccine hesitancy, possibly caused by misinformation [64]. The introduction of AI in biotechnology in vaccine development requires a re-evaluation of people’s perception of newly developed vaccines.

The objective of this study is, therefore, to explore the public perception of AI-developed vaccines, with a case-study on the vaccine against COVID-19. We also address some ethical issues related to AI-developed vaccines by analyzing how people attribute responsibility for the consequences of these new drugs. Our findings suggest tailored approaches to the development of biotechnology-related policies.

2 Background

The COVID-19 Pandemic

COVID-19 is a newly discovered infectious disease with respiratory symptoms, caused by SARS-CoV-2 [65], which has been declared a pandemic by the World Health Organization (WHO) in March 2020 [70]. The disease was first found in Wuhan, China, in late 2019. Since then, it has spread across the globe infecting, at the time of writing, over 6,000,000 people and leading to the death of more than 350,000 patients.

While estimates of the disease’s reach and death rate have constantly been changing during its continuous spread, early data suggests that certain groups, such as elders and patients of pre-existing medical conditions, are more likely to fall severely ill from COVID-19. For instance, researchers have estimated a 20 times higher death rate for those aged over 60 when compared to their younger counterparts 

[62].

No specific treatment or vaccine for COVID-19 has been approved so far. As containment measures against the disease, the WHO and national health organizations all around the world have been promoting public policies, such as quarantines, that have led some countries to lockdown their entire population at home. These policies are expected to impact the world in the most diverse aspects harshly; for instance, the International Monetary Fund (IMF) has declared that the world will most likely face the most significant economic recession since the Great Depression [7].

Protests around the world are currently emerging to fight against these governmental decisions [2]. These unpopular policies aim to contain the spread of the virus while a vaccine is being developed. Countries that have faced the novel coronavirus earlier in the pandemic are starting to lose some restrictions [31], such as enforced lockdowns while promoting more lenient policies such as social distancing among their citizens. Doing so, however, has led to second waves of infections in some countries [76].

The COVID-19 pandemic has also created a so-called infodemic [71, 18], in which a plethora of false information regarding COVID-19 is rapidly being shared online. Studies indicate that misinformation about Ebola has negatively influenced vaccine acceptance during its epidemic [64]. The WHO has also reported how misinformation has led to unfounded controversies about the safety of the HPV vaccine [67]. Research analyzing the reach of anti-vaccine movements in social media demonstrated how these groups could dominate over those pro-vaccine online [25]. These results, therefore, indicate that the COVID-19 infodemic might contribute to vaccination hesitancy and refusal, an alarming note in the current pandemic.

Vaccination and Artificial Intelligence

Vaccine development is a complex process involving many steps, including preclinical, clinical post-licensure development. It also requires an integration of information concerning 1) pathogen life-cycle and epidemiology, 2) immune control and escape, 3) antigen selection and vaccine formulation, and 4) vaccine preclinical and clinical testing [11].

Vaccination has been successful in controlling diseases [48]. The WHO estimated that more than 17.1 million lives had been saved from measles due to widespread vaccination from 2000 to 2015 [66]. Philanthropists have invested millions of dollars in the development of new vaccines to combat diseases that still affect certain areas of the world, such as malaria in Africa [29].

Nevertheless, vaccine hesitancy has been ranked in the top-10 2019 health threats by the WHO [68]. Anti-vaccine movements still contribute to an increase in vaccine-preventable outbreaks and epidemics by promoting hesitancy. These harmful fronts sway vaccine acceptance and are often influenced by conspiracy theories [26], lack of trust in the health system, past experiences with vaccination, and other factors [17].

Much research has been devoted to understanding the acceptance of vaccines. For instance, scholars have used both quantitative and qualitative methods to explore vaccine decisions [32, 41, 16]. Hesitancy to get vaccinated can be caused by a broad range of factors, including the compulsory nature of vaccines, unfamiliarity with vaccine-preventable diseases, and lack of trust in corporations and public health organizations [54]. Previous work has also indicated differences in vaccine acceptance across cultural and ethnic groups [73, 13]. Regarding the past 2009 H1N1 pandemic, scholars have also addressed factors influencing vaccine decisions within the general population [50] and health care workers [74].

Widespread vaccination can result in herd immunity, under which a specific population becomes resistant to the disease if a majority of its citizens develop immunity against it [1]. Vaccine refusal could lead to a longer-standing disease, causing many more deaths than if vaccination was widespread. Therefore, the development of a COVID-19 vaccine is urgently needed to overcome the crisis in the pandemic situation.

AI has entered the field of medicine, in which much of its research has been dedicated to the diagnosis and treatment of diseases [47, 36, 40]. Regarding the COVID-19 pandemic, AI is currently being used to diagnose cases from CT scans and symptoms [35, 79, 24], to predict the spread of the virus [77], and in many other areas. For instance, an AI-based search system has also identified an existing drug that could effectively combat the disease [52].

Vaccine development is also employing AI systems. For instance, SAM (i.e., Search Algorithm for Ligands), an AI program designed to identify compounds to improve the human immune system, has independently developed a highly effective flu vaccine [42]. While AI could shorten the drug discovery process and improve the global health system by identifying new and more effective vaccines, it could also arguably contribute to the spread of anti-vaccine movements. Scholars have discussed how the general public might develop trust towards AI systems differently in comparison to their human counterparts [72], and therefore employing AI in the vaccine development might increase vaccine hesitancy and refusal, making more difficult the fight against vaccine-preventable diseases.

Responsibility Issues for Artificial Intelligence’s Actions

The deployment of AI systems in various sectors of society has raised numerous ethical, legal, and moral questions. Researchers have proposed the concept of Responsible AI [14], in which developers are to take responsibility for all steps of the development and deployment of their systems. Scholars, however, have also discussed how doing so might not be viable given the unpredictability and complexity of self-learning and autonomous AI systems [21].

The concept of responsibility has various related meanings [60, 63], all of which with its requirements for attribution [59]. Scholars have previously discussed whether AI and other stakeholders could satisfy these conditions, concluding that no entity fulfills all requirements for responsibility attribution [38, 33].

Similarly to the problem of many hands [61] in attributing responsibility to collective agents, responsibility attribution for AI actions suffers from the issue of “many things” [8]: AI is a collection of various entities, technologies, and smaller interacting structures that makes the assignment of responsibility more complex. In medicine, an area where consequences are often a matter of life and death, successfully ascribing responsibility for mistakes and negative outcomes is an important step towards developing safe and trustworthy systems.

In the area of medicine, scholars have shown that the public assigns responsibility and punishment for negative outcomes to both AI and human doctors, although to different extents [37]. Autonomous cars are also blamed, to a lesser degree, than human drivers in the occurrence of accidents [3].

Previous research has also proposed the existence of various moral and legal gaps arising from the deployment of AI systems. The responsibility gap, for instance, is created by the lack of capacity to predict the behavior of self-learning machines by the manufacturer or operator [44], which results in these entities not being able to be held morally and legally responsible for outcomes. Similarly, the accountability gap arises from a distance between the operator/manufacturer and the machine, making difficult the attribution of “causality to either the physical person or company that is behind the (electronic) agent” [28]. Finally, the punishment gap is a public contradiction in which the general public desires to punish AI agents for their wrongful actions even though people are aware that doing so is not feasible in the current legal landscape [39].

3 Research Questions

This study asks the following research questions (RQ):

  • RQ1: How willing are people to get vaccinated against COVID-19 during the pandemic?

  • RQ2: How does willingness to get vaccinated change depending on the recipient?

Understanding public willingness to accept an AI-developed vaccine can contribute to the fight against future anti-vaccine movements, so public officials and health care workers can better adapt to when these new vaccines are available. In our study, we include AI into the development of vaccines in two different forms: an AI system 1) independently identifying a compound for a COVID-19 vaccine and 2) assisting (human) researchers in the process of developing it. While a lot of research has been devoted to developing autonomous AI systems (e.g., autonomous vehicles), they are often still deployed in collaboration with human agents (e.g., Tesla’s autopilot requires human input). Vaccine development would be no exception to the rule, and deploying these systems would at first require human oversight. RQ3 raises the question of whether public willingness to get vaccinated might be influenced by the employment of AI in vaccine development.

  • RQ3: Does including AI in the process of developing a COVID-19 vaccine change people’s willingness to get vaccinated?

News coverage influences public perception of various and disjoint issues [19], such as attitudes towards African Americans [20], perceptions of educational institutions [27], and political campaigns [51]. Medicine, and more specifically, vaccination, can also be influenced by news reports. For instance, there is evidence that local newspapers have previously influenced vaccine acceptance [43]. Previous work has also shown that parents who have read or heard negative information regarding vaccines tend to delay childhood vaccination [57].

Hence, in this study, we raise the question of whether the approach used by news media to introduce the effectiveness or side effects of a future COVID-19 vaccine can influence public willingness to get vaccinated.

  • RQ4: Does willingness to get vaccinated against COVID-19 change based on how the effectiveness and possible side-effects are introduced?

Finally, we take a descriptive approach and aim to understand how the public assigns responsibility for the consequences, positive or otherwise, of the development of vaccines in RQ5. Inspired by previous work, we address responsibility as blame (or credit) and punishment (i.e., liability) and question to what extent people assign these variables to entities involved in the development of vaccines, such as the AI itself, researchers, and the company who developed the AI. In this study, we also tackle awareness, i.e., knowledge of one’s actions and consequences, as an important aspect of responsibility attribution. Knowledge has been proposed as one requirement for blame assignment [59], while awareness is an important aspect of civil and criminal liability (e.g., foreseeability in civil law, mens rea in criminal law).

  • RQ5: How are responsibility and awareness attributed to the entities involved in the development and deployment of the COVID-19 vaccine?

4 Methodology

Study Design

We address our proposed research questions via a cross-sectional study design, which employed a between-subjects design, in which participants were presented two different news-like vignettes out of nine (33) combinations. It is important to note that the survey was conducted in the midst of the pandemic. This study has been approved by the first authors’ institution review board.

Survey

The survey started by briefly introducing COVID-19 and the number of confirmed cases and deaths at the time of the study. The participants were then asked three questions addressing whether they had any close contact with COVID-19 suspected or confirmed patients. We also measured the participants’ personal concerns regarding the disease. For each participant, we averaged their answers to four questions designed to quantify people’s concern regarding the personal consequences of COVID-19 and their perceived likelihood of infection. The participants reported their level of concern using a slider in the range of -1 and 1. The proposed scale reported an acceptable level of internal consistency (Cronbach’s =0.71). The questions are presented in Appendix A.

We quantify the participants’ willingness to accept a COVID-19 vaccine by first presenting a stimulus similar to a realistic newspaper article (see Appendix B). The proposed vignettes introduced the beginning of trials of a vaccine against the novel coronavirus. We designed three versions of articles that solely differed in who was the entity that led to the development of the vaccine.

In the first vignette, the vaccine was completely developed by a collaboration of human researchers from various institutes; we call this treatment group human-developed (vaccine). In the other two proposed stimuli, an AI system was introduced in the development of the vaccine. The human-AI collaboration scenario introduced SAM (Search Algorithm for Ligands), the AI program responsible for the first AI-developed vaccine [42]. In this article, SAM was the entity that identified a set of compounds that were effective against COVID-19. Human researchers, working alongside SAM, distinguished the best one out of the set and synthesized the vaccine. The final article also introduced SAM; the scenario, however, presented the AI program as the entity who found the most effective compound and explicitly posited that the human researchers only synthesized the vaccine. We address this final scenario as AI-developed (vaccine).

The participants were randomly assigned an article. After reading it, the participants were asked how willing they were to get this vaccine to 1) themselves, 2) their child, and 3) their elders. The respondents were also asked to what extent they were worried about getting this new vaccine. All questions were presented in random order, and respondents answered them using a slider in the range of -1 and 1.

As an attention check question, the participants were then presented a question asking them who was able to find a compound for the COVID-19 vaccine, according to the article. The participants picked the best answer out of four options: a team of researchers, an AI program, a team of researchers with the help of the AI program, and the general public. We discuss how this attention check was used to remove inattentive responses below.

The survey subsequently presented another randomly selected news-like vignette to the participants. The news article introduced a future scenario in mid-2021, in which a COVID-19 vaccine has finally been approved and made available to the public. Each stimulus framed the effectiveness or side effects of the vaccine in a different manner. The positive news article introduced the vaccine as highly effective in developing immunity against the novel coronavirus. The article explicitly compared the effectiveness of this vaccine to the average effectiveness of childhood vaccines according to the WHO111“Most routine childhood vaccines are effective for 85% to 95% of recipients.” Available at https://www.who.int/vaccine˙safety/initiative/detection/immunization˙misconceptions/en/index2.html. On the other hand, the negative news-like stimulus discussed how there had been some reports of mild side effects, such as headaches and fever, on recipients of this new vaccine. The vignette concluded by stating that no complications had been reported and that the vaccine was the most effective way to combat the disease. We also designed a control stimulus, which excluded any mentions to the effectiveness or the side effects of the vaccine.

After reading the second vignette, the participants were asked the extents to which they would assign responsibility and awareness to various entities involved in the development and deployment of the vaccine. Participants assigned to the human-developed vaccine scenario were presented three entities to which they were asked to attribute these variables: the researchers who developed the vaccine (shown in plots as Human), the government (Gov), and the health care worker who administers the vaccine (Wrkr). In the case of a human-AI collaboration or a AI-developed vaccine, respondents were additionally asked to assign responsibility and awareness to SAM (AI), its developing company (Comp) and its main programmer (Progr).

Had participants been assigned to a negative stimulus, they were asked the levels of blame, awareness, and punishment they would assign to all entities for the side effects of the vaccine. The participants assigned to a positive and control stimulus, however, attributed credit (i.e., praise) and awareness for the effectiveness and development of the COVID-19 vaccine, respectively. Entities were presented one at a time and in random order to all participants. The responses were recorded on a scale of -1 to 1 using a slider. Finally, participants were asked the same willingness- and worry-related questions from the initial stimulus.

Finally, we used the Medical Maximizer-Minimizer Scale (MMMS) [55] to measure to what extent the participants were “medical maximizers” or “minimizers”. ”Medical maximizers” are those individuals predisposed to seek health care for minor problems, while “minimizers” would rather avoid health care treatment unless extremely necessary. Responses to all ten questions reported a high level of internal consistency (Cronbach’s = 0.86). After answering all questions, the participants were asked demographic questions, such as their age and political views.

Participants and Recruitment

We conducted our survey on Prolific, a survey sampling firm, in mid-April 2020. We recruited 630 participants from the US and UK. Prolific allowed us to recruit participants representing current sex, age, and ethnicity US and UK demographic distributions.

Figure 1: Cumulative distribution of average personal concern regarding COVID-19. The vertical lines indicate the division between categories of those less, moderately, and very concerned about the disease.

Respondents were removed from the analysis had they failed the attention check question or did not spend enough time reading both stimuli. Participants were considered to fail the attention check had they chosen the general public as the entity who identified the vaccine compound. We also discarded respondents who did not indicate the participation of an AI program even though they had been assigned a AI-developed or human-AI collaboration scenario. Finally, we removed participants that either spent less than half its reading time222We calculated the reading time of our articles using https://wordcounter.net/. or more than six minutes reading each of the news-like stimuli. Our final dataset consisted of 572 responses, out of which 287 and 285 respondents are US and UK residents, respectively.

We categorize participants into three groups: those less concerned about the disease (little concern), those very concerned about it (much concern), and finally, those in the “middle of the road” (moderate concern). We divide participants into three quantiles of personal concern based on their responses to the concern-related questions. Figure 

1 shows the cumulative distribution of responses and their categorization. Throughout this paper, we discuss people’s personal concerns using these defined categories.

We employ a similar method to categorize respondents between “medical maximizers” and “minimizers”. Respondents were divided into three categories of similar range. For instance, respondents with an MMMS value of less than -0.333 were categorized as “medical minimizers”; we consider “maximizers” those who reported an MMMS response of over 0.333.

Data Analysis

We analyze our data by conducting multi-way ANOVA tests for addressing the existence of statistically significant differences across treatment groups, and Tukey honestly significant difference (HSD) tests for posthoc pairwise comparisons between them. Interaction terms between treatment groups are included in all significance tests with the exception of the demographic analysis. All plots, unless otherwise stated, present the group’s means and their respective 95% confidence intervals.

5 Results

RQ1-2: Willingness to Get Vaccinated Against COVID-19

Our first research question broadly addressed the overall willingness to accept a vaccine to combat the current pandemic. Participants of this study were marginally predisposed to accept a new vaccine against the novel coronavirus, with a reported median and mean willingness of 0.170 and 0.143, respectively.

Figure 2: Willingness to accept a COVID-19 vaccine depending on the recipient. Figure a) is shows vaccine acceptance as a function of the respondents’ level of concern regarding the novel coronavirus. Figure b) shows willingness depending on whether the participants is a medical minimizer, maximizer, or in between.

Figure 2a shows a lower overall willingness to vaccinate children (=31.731, <.001) regardless of the respondent’s level of concern regarding COVID-19 (=0.617, =.650) or their predisposition to seek health care (=0.183, =.947). The participants revealed a lower acceptance of the vaccine for children when compared to their willingness to vaccinate themselves (=0.227, <.001) and their elders (=0.240, <.001). The participants, however, indicated no difference in their inclination to get personally vaccinated against COVID-19 and the predisposition to take their elders to get the vaccine (>.05). We did not find any differences between willingness of US and UK participants.

The participants’ level of concern regarding the novel coronavirus is positively correlated with the overall willingness to accept a newly developed COVID-19 vaccine to all recipients (=20.799, <.001). Those more concerned about the disease were more predisposed to get vaccinated in relation to those less concerned about it (=0.221, <.001) or in the middle of the road (=0.119, <.005). Moreover, those in the middle of the road were more willing to accept the vaccine than less concerned respondents (=0.102, <.01).

Figure 2b shows that medical maximizers also reported a higher willingness to get vaccinated (=12.853, <.001) against COVID-19 when compared to medical minimizers (=0.232, <.001) and those in the center of the MMMS scale (=0.131, <.05). Moreover, medical minimizers were also marginally less willing to get vaccinated vis-à-vis those with less extreme predisposition or aversion to seeking health care (=0.101, <.05).

Finally, we also analyzed whether there exists any difference in vaccination acceptance depending on the respondents’ previous experiences with vaccines and their contact with COVID-19. Plots of the overall acceptance of the vaccine, regardless of the recipient, based on these attributes are presented in Appendix E. The participants who have received all their countries’ required vaccines reported a higher acceptance (=26.795, <.001) in comparison to those who have only received some (=0.196, <.001) or none of them (=0.219, <.001). Moreover, having an acquaintance or themselves previously infected with COVID-19 also increased willingness to get vaccinated (=7.387, =0.083, <.01). Lastly, those who were residing in cities where COVID-19 had at least infected one resident also reported lower hesitancy towards the vaccine (=23.516, =0.183, <.001).

RQ3: Acceptance of AI in the Development of Vaccines

Figure 3: Willingness to accept a COVID-19 vaccine depending on the entity who led the development of the vaccine. Figure a) shows vaccine acceptance depending on the level of concern regarding COVID-19. Figure b) presents willingness as a function of the participant’s predisposition to seek health care.

Figure 3 shows the mean vaccination acceptance scores regardless of the recipient, depending on who was the entity that led its development: an AI program, a team of researchers, or a collaboration between both. Participants’ willingness to get vaccinated altered with the insertion of AI in the development of vaccines (=8.250, <.001). Our findings indicate an overall lower willingness to get vaccinated if the vaccine’s compound was independently found by an AI rather than solely by human researchers (=0.135, <.001).

Our results do not show any significant interaction between the participants’ level of concern regarding the disease and the entity who identified the vaccine’s compound in predicting their willingness to accept the vaccine (>.05). On the other hand, medical maximizers indicated a higher willingness to accept an AI-developed vaccine than medical minimizers (see Figure 3b, =0.286, <.05) and those in the middle of the road (=0.248, <.005).

RQ4: Framing of the Effects of the Vaccine

Our fourth research question examined whether the form used to introduce the consequences of the COVID-19 vaccine modifies their responses to how willing they would be in accepting it in the future. We present the outcomes of the vaccine in two different forms: discussing its high effectiveness against the novel coronavirus or reporting some side effects of the vaccine. As a method of dealing with a possible subject-expectancy effect caused by asking similar questions in sequence, we also design a control group in which the consequences of the vaccine are not explicitly introduced.

Figure 4: Assignment of blame, credit, and awareness to all entities involved in the development and deployment of the COVID-19 vaccine depending on how the consequences of the vaccine were framed.

Our results indicate a significant difference across treatment groups (=31.903, <.001). A more positive framing of the consequences of the COVID-19 vaccine increased overall willingness to get vaccinated in comparison to the control group (=0.182, <.001). Interestingly, a more negative framing also reported a positive willingness-change in comparison to the control stimulus (=0.113, <.001).

The entity who had developed the vaccine according to the first article presented to the participants influenced how much they modified their acceptance of the vaccine (=4.986, <.01). The respondents assigned to the human-AI collaboration stimulus reported a marginally higher positive change in overall willingness to get vaccinated in relation to those assigned to a AI-developed (=0.061, <.05) or a human-developed (=0.069, <.05) vaccines.

Finally, the participants’ initial willingness was strongly correlated to how much people modified their inclination to get vaccinated upon an intervention (=110.949, <.001). Participants in the lower tercile of vaccine acceptance reported a higher willingness-change when compared to those in both the center and upper terciles of willingness to get vaccinated against COVID-19 (=0.150, =0.345, respectively, both <.001). Those who reported an average willingness, i.e., in the center tercile, were more affected by the stimulus than respondent’s in the upper tercile (=0.194, <.001). Therefore, those initially less willingness to accept a vaccine are overall more likely to increase their inclination to get vaccinated (Pearson’s =-0.338, 95% conf [-0.379, -0.295], <.001).

RQ5: Responsibility for the Vaccine

Figure 4 shows participants’ level of ascribed blame, credit, and awareness to all entities addressed in our study. Even though liability could be considered one of the various meanings of responsibility [59], we address blame and praise as responsibility and punishment as liability below. Participants in this stddy assign vastly less blame for the side effects of the vaccine than credit for its development and effectiveness across all entities (=802.406, <.001). This decrease in responsibility, however, is, to a lesser extent, for the government. The overall responsibility ascribed to AI is similar to that attributed to its programmer (¿.05). The AI program is assigned marginally less blame and credit than its human counterpart (=0.086, <.001). Our results also indicate no difference whatsoever in the level of responsibility attributed to AI and humans depending on whether they independently or collaboratively identified the vaccine’s compound (see Appendix D, both >.05).

The level of awareness ascribed to all entities is overall lower in the case of negative framing of the consequences of the vaccine (=59.552, <.001). Therefore, people believed entities are less aware of the possible side effects of a vaccine than of its effectiveness (=0.115, <.001) and its development in general (=0.118, <.001). The level of awareness assigned to an AI is lower than that assigned to all entities (all <0.001). People also presumed the government should be as aware as the researchers who developed it regardless of the consequences of the vaccine (>0.05). Even though the overall awareness assigned to all entities by the human-AI collaboration treatment group was marginally lower than in the case of an human-developed vaccine (=0.084, <.001), our results suggest no significant difference between the attribution of awareness to AI and the human researchers depending on who was the entity who identified the vaccine’s compound.

Figure 5 shows to what extent participants assigned punishment for the side effects of the vaccine. Participants were only asked these questions had they been assigned to the negative treatment group.The overall punishment assigned to all entities is vastly lower than responsibility (i.e., blame and credit), with the government being ascribed marginally higher liability than all entities excluding the AI’s developing company (all <.05).

Figure 5: Assignment of punishment to all entities involved in the development the COVID-19 vaccine.

6 Discussion

RQ1: Those Concerned About COVID-19 Are More Inclined to Vaccinate Against It

Our results indicate that the personal level of concern regarding the novel coronavirus and the participants’ predisposition to seek health care influence the participants’ acceptance of the vaccine. Those more concerned about it are more willing to accept it; similarly, medical maximizers also report a higher level of willingness to get vaccinated against COVID-19. The participants who have reported a lower level of concern regarding COVID-19 demonstrated a borderline willingness to get vaccinated against it (=0.034). More alarmingly, medical minimizers indicated an even higher hesitancy towards the vaccine (=0.015). Even though the current pandemic has greatly affected the population surveyed (e.g., at the time of the study, the US was the country with the highest number of deaths), certain layers of society appear to hesitate vaccination against COVID-19 still. This trend is especially distinguishable among those participants who had not been received all nationally required vaccines or had not had any close contact with the disease.

Our results, therefore, suggest that future public policies should strongly promote the vaccination against the novel coronavirus once (and possibly before) the vaccine has been approved. Citizens who are less concerned about the disease, with history of vaccination hesitancy or refusal, and those groups less affected by the virus are less likely to get vaccinated against COVID-19. This is in line with the Health Belief Model, a classic theory which identified perceived susceptibility and severity as factors for engaging with health behavior [53]. Governments could develop health policies to promote vaccination for COVID-19 based on this finding. For example, national campaigns which increase the awareness of risk and side effects of COVID-19 might increase the perceived susceptibility and severity. These efforts could also increase compliance with preventive measures, such as social distancing, if targeted at those less concerned. For instance, an earlier study has found that younger people are less concerned about the disease and demonstrate lower compliance with these measures [45].

RQ2: People Hesitate to Vaccinate Children Against COVID-19

Our findings indicate that people are less willing to vaccinate their children in comparison to their willingness to vaccinate themselves or their elders. This finding is intriguing considering that the elders have been hit the hardest by the COVID-19 pandemic [62]. While our results indicate that people are aware that the oldest layer of society must also be vaccinated against the novel coronavirus, people hesitate to vaccinate the youngest.

We hypothesize that this result might be influenced by the false claims that children are resistant or even immune to COVID-19 [23, 49]. These rumors have not only been disseminated by the general population but also world leaders and public figures in various countries [46, 4]. Those who defend more lenient quarantine policies often argue that the youngest layer of society should not be isolated due to the lower risk of complications. Defenders of these proposals, however, neglect the possibility of spread from the youngest to the elderly, i.e., those more susceptible to complications due to COVID-19. Moreover, the burden of diseases caused by the long-term impact from the coronavirus such as cardiovascular complications [75] and mental health issues [30] could decrease the quality of life and increase healthcare cost in children for a prolonged period of time. Our findings suggest that the government and health organizations, such as local CDCs and the WHO, should also highlight the importance of and promote the vaccination against COVID-19 for the youngest layers of society as a form of control of the disease once a vaccine is available to the population.

RQ3: Human-AI Collaborative Vaccines Do Not Affect Vaccination Acceptance

Including AI systems in the development of vaccines might have an effect on vaccine acceptance. RQ3 addressed this hypothesis by asking whether a vaccine independently or collaboratively created by an AI modifies the respondent’s willingness to get vaccinated vis-à-vis vaccines solely created by human researchers. Our results indicate that an AI-developed vaccine has a lower acceptance rate in comparison to those developed by a team of human researchers. However, our findings do not show any significant difference in willingness to get vaccinated if the vaccine was developed by both an AI and human researchers collaboratively.

While the public might not be entirely trustful of vaccines solely developed by AI, our results suggest that explicitly introducing the role of these systems alongside human researchers in the development of collaborative vaccines does not modify vaccine acceptance. AI might prove to shorten drug discovery time and therefore improve the global health system [42]. Therefore, the inclusion of these systems in this area of medicine, if the humans are not yet completely removed from the process, should not contribute to vaccine hesitancy and refusal while advancing the process of vaccine development.

RQ4: Introducing the Effectiveness of the Vaccine Can Decrease Hesitancy

Explicitly introducing the effectiveness of a new vaccine has shown to positively influence willingness to get vaccinated in comparison to only reporting the availability of the vaccine to the population (i.e., the control group in our study). Our results suggest, therefore, that governments might choose to promote vaccination through an advertisement of the efficacy of a coronavirus vaccine once it has been approved.

We also examined whether a more negative stance while reporting the outcomes of the vaccine could negatively influence people’s willingness to get vaccinated. Our results indicate a marginal positive overall willingness-change in comparison to the control group. We hypothesize that this is caused by the form of how the stimulus was designed. First, the consequences were not extremely serious or harmful; we explicitly addressed the side effects saying that they did not lead to any complications and disappeared after a few hours. Second, we also reported at the end of the articles that this vaccine was the most effective and safest way to combat the pandemic, so our intervention would not promote anti-vaccination feelings. Moreover, the participants assigned to the human-AI collaboration scenario reported an even higher positive change in vaccine acceptance in such treatment group.

Our results suggest that those who are initially less willing to vaccinate report a greater positive change in their acceptance of the vaccine. Therefore, public promotion campaigns, especially if focused on advertising vaccines’ effectiveness against the novel coronavirus, could influence those less willing to get vaccinated, an important step towards stopping the pandemic.

Finally, a human-AI collaborative vaccine has shown to be malleable in terms of public willingness. We thus posit that people are open to human-AI collaborations in this field, particularly if such vaccines are proved to be safe (i.e., approved). Alongside our results indicating that participants’ initial acceptance of the vaccine did not differ between vaccines solely developed by humans and those collaboratively created, our findings indicate that including AI into the development of vaccines could be extremely beneficial to the process without much public hesitancy or backlash.

RQ5: Awareness-Lacking AI Is Praised for Its Vaccine and the Government’s Role in Approving Safe Vaccines

Participants in this study assigned a high level of credit to the AI system for its development and effectiveness. In terms of blame, moreover, AI was attributed to similar levels to that of human researchers had both of them been involved in the process. In agreement with previous work [39], the participants regard AI systems as responsible for their actions, although they are not to be considered aware of their actions. Even though marginally less responsibility is attributed to these systems in comparison to their human counterparts, their blame or credit for the outcomes of the vaccine is similar to those ascribed to their main programmers.

The government and the team of researchers involved in the development of the vaccine are ascribed to the most awareness. The government, even though it is assigned less credit for its development and effectiveness, is attributed to similar levels of blame and awareness to the humans who developed the COVID-19 vaccine. Moreover, the government, alongside the AI’s developing company, was assigned the most punishment for its side effects. Therefore, our results indicate a public attribution of responsibility to the government in this process, especially if the vaccine has any side effects. This exemplifies the governmental role in testing and approving safe and effective treatments for the current pandemic.

Our results suggest that people attribute responsibility for the actions of an AI to the system itself, as well as to other entities involved in the development and deployment of these systems. This supports the proposal of an ”extended agency theory,”  [22]

where responsibility and “moral agency would be (jointly) distributed over both human and technological artifacts,” a previously proposed response to the various responsibility gaps raised by the deployment of AI 

[21].

7 Concluding Remarks

The current study indicates three main findings that suggest how governments might want to focus their COVID-19 vaccination policies in the future. First, the participants exhibited a lower willingness to vaccinate children in comparison to accepting it for themselves or their elders. Second, those less concerned about the disease or more doubtful about seeking healthcare (i.e., “medical minimizers”) are significantly more hesitant about the vaccine. Citizens with previous history of vaccination hesitancy and communities less affected by the virus also report a lower willingness to accept the COVID-19 vaccine. Finally, our results suggest that promotion campaigns that highlight the effectiveness of the vaccine against the pandemic might increase willingness to get vaccinated, especially to the ones who are initially more hesitant about it.

The inclusion of AI in the process of development of new vaccines could influence public willingness to get vaccinated, possibly fueling harmful anti-vaccination movements. Our results indicate that while a vaccine whose compound was independently identified by an AI could lead to marginally higher hesitation, a human-AI collaboration does not modify public willingness to get vaccinated. Our results also suggest that human-AI collaborative vaccines are more malleable in terms of acceptance after reports of the vaccine’s approval. Therefore, the inclusion of AI systems, alongside human stakeholders, in this area could highly benefit the development of new vaccines without an increase in vaccination hesitancy or refusal.

The study participants assigned high levels of praise and blame to the AI systems involved in the development of a vaccine, although they are attributed to a lower degree of awareness regarding the vaccine’s outcomes. Our findings also suggest the public assignment of a role to governments in making sure that vaccines are safe and effective after approval alongside other human stakeholders. The government, however, is not highly praised for its development but only blamed or punished for their side effects.

Studies addressing public willingness to vaccinate against COVID-19 should not be restricted to ours, but also be expanded in terms of populations, size, and reach. Vaccination acceptance might also change throughout the pandemic. For instance, misinformation could lead to vaccine refusal, as found during the Ebola epidemic [64]. It is important to note that these results might not be generalizable to other vaccines, as the case for the novel coronavirus is, by definition, biased due to the current pandemic situation. Future studies should address how our findings generalize to other diseases with different levels of prominence. Additionally, we have included AI into the development of vaccines by its role in identifying its compound, as it has been done in the past [42]; however, results might be dependent on an AI’s specific role, an essential line of study is not only vaccine development but all areas where AI is currently being deployed.

In conclusion, our study indicates the government’s crucial role in the approval process of the COVID-19 vaccine. The public believes that the government should be aware of the vaccine’s consequences and blamed if they are not ideal; however, it should not be praised for its development. Tailored health policies to increase the perceived susceptibility and perceived severity could be helpful in increasing the number of people willing to accept a COVID-19 vaccine. These campaigns should especially promote vaccination among children, as people are less willing to vaccine the youngest layer of society against the novel coronavirus. A possibility would be to focus these promotion efforts on highlighting the effectiveness of the vaccine against the pandemic.

References

  • [1] R. M. Anderson and R. M. May (1985) Vaccination and herd immunity to infectious diseases. Nature 318 (6044), pp. 323–329. Cited by: §2.
  • [2] D. Andone (2020) Protests are popping up across the us over stay-at-home restrictions. CNN. Note: https://tinyurl.com/yadnscsb. Accessed 3 May 2020. Cited by: §2.
  • [3] E. Awad, S. Levine, M. Kleiman-Weiner, S. Dsouza, J. B. Tenenbaum, A. Shariff, J. Bonnefon, and I. Rahwan (2020) Drivers are blamed more than their automated cars when both make mistakes. Nature human behaviour 4 (2), pp. 134–143. Cited by: §2.
  • [4] Bello (2020) Jair bolsonaro isolates himself, in the wrong way. The Economist. Note: https://tinyurl.com/rnydo33. Accessed 22 April 2020. Cited by: §6.
  • [5] N. Bogel-Burroughs (2020) Anti-vaccination activists are growing force at virus protests. New York Times. Note: https://tinyurl.com/y7rvkb6q. Accessed 3 May 2020. Cited by: §1.
  • [6] E. Callaway (2020) The race for coronavirus vaccines: a graphical guide. Vol. 580. Cited by: §1.
  • [7] S. P. Chan (2020) Coronavirus: ’world faces worst recession since great depression’. BBC. Note: https://tinyurl.com/uownqpu. Accessed 14 April 2020. Cited by: §2.
  • [8] M. Coeckelbergh (2019) Artificial intelligence, responsibility attribution, and a relational justification of explainability. Science and engineering ethics, pp. 1–18. Cited by: §1, §2.
  • [9] J. Cohen (2020) With record-setting speed, vaccinemakers take their first shots at the new coronavirus’. Science 31. Cited by: §1.
  • [10] M. Consult (2020) How the coronavirus outbreak is impacting public opinion. Note: https://tinyurl.com/y75dcy6k. Accessed 13 May 2020. Cited by: §1.
  • [11] A. L. Cunningham, N. Garçon, O. Leo, L. R. Friedland, R. Strugnell, B. Laupèze, M. Doherty, and P. Stern (2016) Vaccine development: from concept to early clinical testing. Vaccine 34 (52), pp. 6655–6664. Cited by: §2.
  • [12] L. Dearden (2020) How the uk coronavirus lockdown is affecting your human rights. Independent. Note: https://tinyurl.com/y6wl8a2h. Accessed 3 May 2020. Cited by: §1.
  • [13] D. Determann, E. W. de Bekker-Grob, J. French, H. A. Voeten, J. H. Richardus, E. Das, and I. J. Korfage (2016) Future pandemics and vaccination: public opinion and attitudes across three european countries. Vaccine 34 (6), pp. 803–808. Cited by: §2.
  • [14] V. Dignum (2019) Responsible artificial intelligence: how to develop and use ai in a responsible way. Springer International Publishing. Cited by: §2.
  • [15] D. Draulans (2020) ‘Finally, a virus got me.’ scientist who fought ebola and hiv reflects on facing death from covid-19. Science. Note: https://tinyurl.com/yaxf7ycj. Accessed 10 May 2020. Cited by: §1, §1.
  • [16] E. Dubé, D. Gagnon, M. Ouakki, J. A. Bettinger, H. O. Witteman, S. MacDonald, W. Fisher, V. Saini, D. Greyson, et al. (2018) Measuring vaccine acceptance among canadian parents: a survey of the canadian immunization research network. Vaccine 36 (4), pp. 545–552. Cited by: §2.
  • [17] E. Dubé, M. Vivion, and N. E. MacDonald (2015) Vaccine hesitancy, vaccine refusal and the anti-vaccine movement: influence, impact and implications. Expert Rev. Vaccines 14 (1), pp. 99–117. Cited by: §2.
  • [18] V. A. Earnshaw and I. T. Katz (2020) Educate, amplify, and focus to address covid-19 misinformation. In JAMA Health Forum, Vol. 1, pp. e200460–e200460. Cited by: §1, §2.
  • [19] H. Gene Zucker (1978) The variable nature of news media influence. Annals of the International Communication Association 2 (1), pp. 225–240. Cited by: §3.
  • [20] F. D. Gilliam Jr and S. Iyengar (2000) Prime suspects: the influence of local television news on the viewing public. American Journal of Political Science, pp. 560–573. Cited by: §3.
  • [21] D. J. Gunkel (2017) Mind the gap: responsible robotics and the problem of responsibility. Ethics and Information Technology, pp. 1–14. Cited by: §1, §2, §6.
  • [22] F. A. Hanson (2009) Beyond the skin bag: on the moral responsibility of extended agencies. Ethics and information technology 11 (1), pp. 91–99. Cited by: §6.
  • [23] IBS (2020) Facts before rumors campaign webpage. Institute for Basic Science. Note: https://tinyurl.com/yb3oog5a. Accessed 22 April 2020. Cited by: §6.
  • [24] A. Imran, I. Posokhova, H. N. Qureshi, U. Masood, S. Riaz, K. Ali, C. N. John, and M. Nabeel (2020) AI4COVID-19: ai enabled preliminary diagnosis for covid-19 from cough samples via an app. arXiv preprint arXiv:2004.01275. Cited by: §2.
  • [25] N. F. Johnson, N. Velásquez, N. J. Restrepo, R. Leahy, N. Gabriel, S. El Oud, M. Zheng, P. Manrique, S. Wuchty, and Y. Lupu (2020-05-13) The online competition between pro- and anti-vaccination views. Nature. External Links: ISSN 1476-4687, Document, Link Cited by: §2.
  • [26] D. Jolley and K. M. Douglas (2014) The effects of anti-vaccine conspiracy theories on vaccination intentions. PloS one 9 (2). Cited by: §2.
  • [27] S. Kim, J. P. Carvalho, and C. E. Cooksey (2007) Exploring the effects of negative publicity: news coverage and public perceptions of a university. Public Relations Review 33 (2), pp. 233–235. Cited by: §3.
  • [28] B. Koops, M. Hildebrandt, and D. Jaquet-Chiffelle (2010) Bridging the accountability gap: rights for new entities in the information society. Minn. JL Sci. & Tech. 11, pp. 497. Cited by: §2.
  • [29] I. Kottasova (2015) World’s first malaria vaccine, backed by bill gates, gets a green light. CNN. Note: https://tinyurl.com/wrwyzhv. Accessed 14 April 2020. Cited by: §2.
  • [30] A. Kumar and K. R. Nayar (2020) COVID 19 and its mental health consequences. J Ment Heal 8237, pp. 1–2. Cited by: §6.
  • [31] L. Kuo (2020) China gets a glimpse of life on the other side of coronavirus. The Guardian. Note: https://tinyurl.com/slke533. Accessed 14 April 2020. Cited by: §2.
  • [32] Y. Lama, G. R. Hancock, V. S. Freimuth, A. M. Jamison, and S. C. Quinn (2020) Using classification and regression tree analysis to explore parental influenza vaccine decisions. Vaccine 38 (5), pp. 1032–1039. Cited by: §2.
  • [33] J. Lee (2019) Can we impose responsibilities on artificial intelligence?: to seek accountability-oriented ethics for artificial intelligence. Korean journal for the philosophy of science 22(2), pp. 79–104. Cited by: §2.
  • [34] K. Leung, J. T. Wu, D. Liu, and G. M. Leung (2020) First-wave covid-19 transmissibility and severity in china outside hubei after control measures, and second-wave scenario planning: a modelling impact assessment. The Lancet. Cited by: §1.
  • [35] L. Li, L. Qin, Z. Xu, Y. Yin, X. Wang, B. Kong, J. Bai, Y. Lu, Z. Fang, Q. Song, et al. (2020) Artificial intelligence distinguishes covid-19 from community acquired pneumonia on chest ct. Radiology, pp. 200905. Cited by: §2.
  • [36] R. Li, W. Zhang, H. Suk, L. Wang, J. Li, D. Shen, and S. Ji (2014) Deep learning based imaging data completion for improved brain disease diagnosis. In International Conference on Medical Image Computing and Computer-Assisted Intervention, pp. 305–312. Cited by: §2.
  • [37] G. Lima, M. Cha, C. Jeon, and K. Park (2020) Explaining the punishment gap of ai and robots. arXiv preprint arXiv:2003.06507. Cited by: §2.
  • [38] G. Lima and M. Cha (2020) Responsible ai and its stakeholders. arXiv preprint arXiv:2004.11434. Cited by: §2.
  • [39] G. Lima, C. Jeon, M. Cha, and K. Park (2020) Will punishing robots become imperative in the future?. In Extended Abstracts of the 2020 CHI Conference on Human Factors in Computing Systems, CHI EA ’20, New York, NY, USA, pp. 1–8. External Links: ISBN 9781450368193, Link, Document Cited by: §2, §6.
  • [40] S. Liu, S. Liu, W. Cai, S. Pujol, R. Kikinis, and D. Feng (2014) Early diagnosis of alzheimer’s disease with deep learning. In 2014 IEEE 11th international symposium on biomedical imaging (ISBI), pp. 1015–1018. Cited by: §2.
  • [41] A. Marcu, H. Rubinstein, S. Michie, and L. Yardley (2015) Accounting for personal and professional choices for pandemic influenza vaccination amongst english healthcare workers. Vaccine 33 (19), pp. 2267–2272. Cited by: §2.
  • [42] S. Masige (2019) Australian researchers just released the world’s first ai-developed vaccine and it could prevent another horror flu season. Business Insider. Note: https://tinyurl.com/y462rvns. Accessed 14 April 2020. Cited by: §2, §4, §6, §7.
  • [43] B. W. Mason and P. D. Donnelly (2000) Impact of a local newspaper campaign on the uptake of the measles mumps and rubella vaccine. Journal of Epidemiology & Community Health 54 (6), pp. 473–474. Cited by: §3.
  • [44] A. Matthias (2004) The responsibility gap: ascribing responsibility for the actions of learning automata. Ethics and information technology 6 (3), pp. 175–183. Cited by: §2.
  • [45] R. C. Moore, A. Lee, J. T. Hancock, M. Halley, and E. Linos (2020) Experience with social distancing early in the covid-19 pandemic in the united states: implications for public health messaging. medRxiv. Cited by: §6.
  • [46] S. O’Kane (2020) Twitter won’t remove irresponsible elon musk tweet about coronavirus. The Verge. Note: https://tinyurl.com/u8vy3f6. Accessed 22 April 2020. Cited by: §6.
  • [47] S. L. Oh, Y. Hagiwara, U. Raghavendra, R. Yuvaraj, N. Arunkumar, M. Murugappan, and U. R. Acharya (2018) A deep learning approach for parkinson’s disease diagnosis from eeg signals. Neural Computing and Applications, pp. 1–7. Cited by: §2.
  • [48] W. A. Orenstein and R. Ahmed (2017) Simply put: vaccination saves lives. National Acad Sciences. Cited by: §2.
  • [49] J. Powder (2020) COVID-19 myths vs. realities. Johns Hopkins Bloomberg School of Public Health. Note: https://tinyurl.com/uaunuu5. Accessed 22 April 2020. Cited by: §6.
  • [50] S. C. Quinn, S. Kumar, V. S. Freimuth, K. Kidwell, and D. Musa (2009) Public willingness to take a vaccine or drug under emergency use authorization during the 2009 h1n1 pandemic. Biosecurity and bioterrorism: biodefense strategy, practice, and science 7 (3), pp. 275–290. Cited by: §2.
  • [51] J. W. Rhee (1997) Strategy and issue frames in election campaign coverage: a social cognitive account of framing effects. Journal of Communication 47 (3), pp. 26–48. Cited by: §3.
  • [52] P. Richardson, I. Griffin, C. Tucker, D. Smith, O. Oechsle, A. Phelan, and J. Stebbing (2020) Baricitinib as potential treatment for 2019-ncov acute respiratory disease. Lancet (London, England) 395 (10223), pp. e30. Cited by: §2.
  • [53] I. M. Rosenstock (2005) Why people use health services. The Milbank Quarterly 83 (4). Cited by: §6.
  • [54] D. A. Salmon, M. Z. Dudley, J. M. Glanz, and S. B. Omer (2015) Vaccine hesitancy: causes, consequences, and a call to action. Vaccine 33, pp. D66–D71. Cited by: §2.
  • [55] L. D. Scherer, T. J. Caverly, J. Burke, B. J. Zikmund-Fisher, J. T. Kullgren, D. Steinley, D. M. McCarthy, M. Roney, and A. Fagerlin (2016) Development of the medical maximizer-minimizer scale.. Health Psychology 35 (11), pp. 1276. Cited by: §4.
  • [56] K. Sheikh (2020) Find a vaccine. next: produce 300 million vials of it.. New York Times. Note: https://tinyurl.com/yd6jk2u6. Accessed 3 May 2020. Cited by: §1.
  • [57] P. J. Smith, S. G. Humiston, T. Parnell, K. S. Vannice, and D. A. Salmon (2010) The association between intentional delay of vaccine administration and timely childhood vaccination coverage. Public health reports 125 (4), pp. 534–541. Cited by: §3.
  • [58] L. Spinney (2020) When will a coronavirus vaccine be ready?. The Guardian. Note: https://tinyurl.com/wme5puu. Accessed 14 April 2020. Cited by: §1.
  • [59] I. van de Poel and M. Sand (2018) Varieties of responsibility: two problems of responsible innovation. Synthese, pp. 1–19. Cited by: §2, §3, §5.
  • [60] I. Van de Poel (2011) The relation between forward-looking and backward-looking responsibility. In Moral responsibility, pp. 37–52. Cited by: §2.
  • [61] I. van de Poel (2015) The problem of many hands. In Moral responsibility and the problem of many hands, pp. 62–104. Cited by: §2.
  • [62] R. Verity, L. C. Okell, I. Dorigatti, P. Winskill, C. Whittaker, N. Imai, G. Cuomo-Dannenburg, H. Thompson, P. G. Walker, H. Fu, et al. (2020) Estimates of the severity of coronavirus disease 2019: a model-based analysis. The Lancet Infectious Diseases. Cited by: §2, §6.
  • [63] N. A. Vincent (2011) A structured taxonomy of responsibility concepts. In Moral responsibility, pp. 15–35. Cited by: §2.
  • [64] P. Vinck, P. N. Pham, K. K. Bindu, J. Bedford, and E. J. Nilles (2019) Institutional trust and misinformation in the response to the 2018–19 ebola outbreak in north kivu, dr congo: a population-based survey. The Lancet Infectious Diseases 19 (5), pp. 529–536. Cited by: §1, §1, §2, §7.
  • [65] C. Wang, P.W. Horby, F.G. Hayden, and G.F. Gao (2020) A novel coronavirus outbreak of global health concern. Lancet 395 (10223), pp. 470–473. Cited by: §2.
  • [66] WHO (2015) Measles vaccination has saved an estimated 17.1 million lives since 2000. Note: https://tinyurl.com/vcl6e9y. Accessed 14 April 2020. Cited by: §2.
  • [67] WHO (2019) Global advisory committee on vaccine safety, 5-6 june 2019. Note: https://tinyurl.com/yaxawvxd. Accessed 14 May 2020. Cited by: §2.
  • [68] WHO (2019) Ten threats to global health in 2019. Note: https://tinyurl.com/r7jlvbq. Accessed 14 May 2020. Cited by: §2.
  • [69] WHO (2020) WHO director-general’s opening remarks at the media briefing on covid-19 - 27 april 2020. WHO. Note: https://tinyurl.com/y9h6kryd. Accessed 24 April 2020. Cited by: §1.
  • [70] WHO (2020) WHO director-general’s opening remarks at the media briefing on covid-19. Note: https://tinyurl.com/vyvm6ob. Accessed 13 April 2020. Cited by: §1, §2.
  • [71] WHO (2020) WHO novel coronavirus (2019-nCoV) Situation Report 13. Note: https://tinyurl.com/r7jkmwv. Accessed 07 May 2020. Cited by: §2.
  • [72] M. Winikoff (2017) Towards trusting autonomous systems. In International Workshop on Engineering Multi-Agent Systems, pp. 3–20. Cited by: §2.
  • [73] L. P. Wong and I. Sam (2010) Factors influencing the uptake of 2009 h1n1 influenza vaccine in a multiethnic asian population. Vaccine 28 (28), pp. 4499–4505. Cited by: §2.
  • [74] S. Y. Wong, E. L. Wong, J. Chor, K. Kung, P. K. Chan, C. Wong, and S. M. Griffiths (2010) Willingness to accept h1n1 pandemic influenza vaccine: a cross-sectional study of hong kong community nurses. BMC infectious diseases 10 (1), pp. 316. Cited by: §2.
  • [75] T. Xiong, S. Redwood, B. Prendergast, and M. Chen (2020) Coronaviruses and the cardiovascular system: acute and long-term implications. European Heart Journal. Cited by: §6.
  • [76] S. Xu and Y. Li (2020) Beware of the second wave of covid-19. The Lancet. Cited by: §2.
  • [77] Z. Yang, Z. Zeng, K. Wang, S. Wong, W. Liang, M. Zanin, P. Liu, X. Cao, Z. Gao, Z. Mai, et al. (2020) Modified seir and ai prediction of the epidemics trend of covid-19 in china under public health interventions. Journal of Thoracic Disease 12 (3), pp. 165. Cited by: §2.
  • [78] H. M. Yassine and Z. Shah (2020) How could artificial intelligence aid in the fight against coronavirus? an interview with dr hadi m yassine and dr zubair shah by felicity poole, commissioning editor. Expert review of anti-infective therapy 18 (6), pp. 493–497. Cited by: §1.
  • [79] C. Zheng, X. Deng, Q. Fu, Q. Zhou, J. Feng, H. Ma, W. Liu, and X. Wang (2020) Deep learning-based detection for covid-19 from chest ct using weak label. medRxiv. Cited by: §2.

Appendix A

Our survey participants were asked the following four questions as a measure of their personal concern regarding COVID-19. The questions reported an acceptable level of internal consistency (Cronbach‘s =0.71). The responses to these questions were averaged for each respondent and participants were categorized into three groups identified by the distribution terciles, as explained in “Data Cleaning” Section.

  1. How likely are you to become ill with COVID-19?

  2. How severe do you think the economic consequences of COVID-19 will be to you?

  3. If you become infected with COVID-19, how likely do you believe it is that you might fall severely ill or die from it?

  4. How concerned are you about the COVID-19 transmission?

Appendix B

The participants were presented a realistic news article addressing who was the entity that identified the compound used in the COVID-19 vaccine. Figure 6 shows the AI-developed, human-AI collaboration, and human-developed vignettes presented to the respondents. Each participant was randomly assigned to a news article.

(a) AI-developed vaccine
(b) Human-AI collaboration
(c) Human-developed vaccine
Figure 6: News-like stimuli presented to participants before the first set of willingness and worry questions. The articles differ in who was the entity who led the development of the COVID-19 vaccine.

Appendix C

The respondents were presented to another randomly selected news article differing in how the effectiveness and side effects of the COVID-19 vaccine were framed. Figure 7 shows the positive, negative, and control articles presented to them.

(a) Positive framing introducing the high effectiveness of the vaccine
(b) Negative framing describing reports of side effects due to the vaccine
(c) Control framing which does not explicitly discuss the vaccine’s effectiveness or side effects
Figure 7: News-like stimuli presented to participants before the second set of willingness and worry questions. The articles differ in how they frame the consequences of the vaccine.

Appendix D

Figure 8 shows the extent to which the respondents attributed responsibility and awareness to AI and human researchers depending on who was the entity who identified the compound used in the COVID-19 vaccine. While AI systems are attribute lower awareness and marginally less responsibility, our results suggests no difference between treatment groups.

Figure 8: Assignment of blame, credit, and awareness to the AI and the human researchers who developed the COVID-19 vaccine.

Appendix E

Figure 9 shows willingness to accept a COVID-19 vaccine depending on the respondents’ previous experiences with vaccines (left), whether there had been COVID-19 cases in their city (middle), and whether an acquaintance or themselves had been infected by the novel coronavirus (right).

Figure 9: Acceptance of the COVID-19 vaccine depending on the respondents’ vaccination history and contact with the novel coronavirus. All plots present willingness to vaccinate regardless of the recipient.