author
Brief introduction
Liu Na
He is a postdoctoral fellow at the School of Foreign Language and Culture of Communication University of China, and a researcher at the Pomegranate Seed Research Institute Chinese of Communication University of China
Li Qian
Lecturer, School of Journalism and Communication, Beijing Normal University
Liu Qian
Associate Professor, School of Journalism and Communication, Beijing Normal University
Wu Ye
Professor, School of Journalism and Communication, Beijing Normal University, Computational Communication Research Center, Beijing Normal University (Corresponding Author)
body
Article
1. Background
In recent years, emerging information technologies represented by artificial intelligence (AI) are driving the journalism industry to turn to intelligence. As the "latest form and advanced stage" of AI technology application in the media field, AI anchors have become a hot spot for the government, academia and industry. In 2021, the State Administration of Radio, Film and Television issued the "14th Five-Year Plan for the Development of Science and Technology for Radio and Television and Online Audiovisual", which clearly stated: "Promote the widespread use of virtual anchors in the production of news broadcasts and other programs, innovate program forms, and improve the efficiency and intelligence of production and broadcasting." "With the support of technology and policies, AI anchors are widely used in different news scenarios, and the practice of using AI anchors as the main body of communication has become a revolution in the field of news broadcasting, which will have a profound impact on the media industry pattern, especially media practitioners.
AI anchor refers to "intelligent media products that are driven by AI technologies such as big data processing and learning, virtual synthesis and clones, and human-computer interaction, and are responsible for broadcasting tasks in the media". Before the widespread application of AI anchors, live anchors were the key communicators in the hosting and broadcasting links, and were known as the spokespersons of the media and the audience. In the digital age, even if the functions of information intermediary and process connection of live anchors have been partially reconstructed or replaced, they are still a key part of media such as radio and television, connecting a wide range of social life spaces, and playing an important role in determining the communication effect. Therefore, can AI anchors play the same functions and roles as real anchors, and how does the audience perceive and evaluate the news content broadcast by AI anchors? is the focus of this article.
2. Literature review
(1) Theoretical basis
AI anchor is a type of machine, and the audience's perception and evaluation of it belongs to the category of human machine communication (HMC). Different from radio, television, telephone and other media, AI anchors move from behind the scenes to the foreground, and directly establish sensory linkage with the audience through the lens of visual and auditory symbols, no longer just as a mediator for information transmission, but as a "other" as the sender or receiver of information to participate in the entire communication and interaction process. With the advancement of technology, the boundaries between humans and machines are gradually blurred, and communication is no longer limited to people, but also between humans and machines.
Media equation theory and the CASA paradigm help to understand this new type of human-machine relationship. In Media Equivalence, Clifford Nass and Byron Reeves argue that individual interactions with computers, radio and television, and new media are based on social and natural responses, just as they do with people in real life. CASA, an important paradigm of media equivalence theory, further points out that human-computer interaction will apply social norms such as reciprocity and stereotypes to human interactions, and even if people realize that the media is completely different from humans and that this response is inappropriate, they will unconsciously apply the thinking patterns learned in interpersonal communication. This unconscious behavior may be associated with premature cognitive commitments, and to elicit such an unconscious social response, the media needs to present enough social cues for people to classify them as social actors and respond accordingly.
Initially, the CASA paradigm and media equivalence theory were mainly applied to explain human-computer interactions. For example, humans tend to rate computers they interact directly with more positively, and less negatively when they don't. People tend to classify computers as teammates or non-teammates based on their consistency with their personal identity, which in turn affects their evaluation and acceptance of opinions. In addition, computer sounds are also a strong social cue. Experiments have found that audiences are able to identify the personality tendencies of computer voices and produce corresponding social-emotional responses: audiences who hear computer voices exhibit similar personalities to their own tend to perceive computers as having a stronger sense of social presence.
Subsequently, the CASA paradigm has also been applied to other mediums, such as autonomous vehicles, robots, and virtual humans. When the media presents social cues such as language, appearance, sound, emotion, and action, the audience will see them as social actors and respond socially accordingly. Self-driving vehicles or robots with names are more likely to be favored and trusted by the audience, short-haired robots are considered more suitable for jobs such as repair and care, while long-haired robots are more suitable for jobs such as housework and nursing; AI teachers with real human voices instead of synthetic voices are more trusted, and human voices have more advantages than synthetic voices in audience emotion arousal and thinking cognition. Chatbots that exhibit human emotions such as empathy and enthusiasm are considered more supportive and better able to motivate audiences to respond to their requests; Robots use gestures to be more engaging and intimate; Embedded in retail websites is a virtual shopping guide that provides consumers with explanation services, which can stimulate consumers' positive emotions and purchase intentions.
It can be found that anthropomorphism such as human characteristics, behaviors, and emotions are very important social cues in related studies. Of course, anthropomorphism is not always as realistic as possible, and studies have found that virtual humans with a lower degree of anthropomorphic appearance can trigger a sense of shared presence and social presence in the subject. According to the uncanny valley effect, as the degree of anthropomorphism of non-human objects increases, human affection for them increases, decreases, and increases. In the beginning, the higher the fidelity, the more positive emotions there are; And when the fidelity reaches a certain value, the favorability will suddenly fall to the bottom, resulting in disgust and horror; And as the similarity increases, the favorability increases again. It is worth paying attention to how AI anchors can present enough social clues to achieve better interactive effects, and not fall into the uncanny valley and cause the audience to be disgusted.
(2) Literature review
With the "full penetration" of AI in the field of news communication, more and more research focuses on the audience's attitude and perception of AI-generated content. A large number of empirical studies have focused on "automated journalism" that uses algorithms to convert data into news text, especially the comparison between AI-generated news and traditional human writing. Overall, AI-generated news and human-written content are on par with audience attitudes and perceptions, with only minor differences in some aspects. Some studies have found that AI news writing is better than human writing in terms of audience perception professionalism, credibility, and objectivity. Some studies have also found that compared with AI writing, journalists write news content that is more professional, credible, and readable. The conflicting conclusions may be due to the fact that the audience does not perceive AI news content in a single way, but is influenced by the interaction of other factors, for example, the audience perceives AI news on online media as more interesting, while news written by human journalists on traditional media is more popular. In addition, the research also focuses on "data journalism" that is "based on data capture, mining, statistics, analysis, and visualization". The results show that audiences prefer data journalism in a novel format over traditional text reporting. Other studies have focused on the audience's perception of AI-generated art content, and found that there is no significant difference in the audience's evaluation of AI-created poetry compared with that created by real people. The audience is inferior to human-drawn paintings in terms of preference, beauty, novelty, and comprehensibility.
Compared with machine writing, data journalism, and AI art creation, AI anchors appeared relatively late and have a relatively single research, mainly focusing on the topic of whether AI anchors can replace real anchors. Some scholars have pointed out the obvious advantages of AI anchors, such as unlimited work locations and working hours, especially in special periods such as the new crown epidemic, AI anchors can be on duty at any time and broadcast uninterrupted around the clock; The "second-level conversion" from text content to audio and video greatly improves the efficiency of news production and ensures the timeliness and accuracy of news. The huge and complete information reserve system supports real-time interaction between AI anchors and audiences, and provides reports and consulting services. AI anchors not only have strong capabilities, but also gradually have a "virtual personality", and interact with the audience to show the characteristics of interpersonal communication, becoming a new type of "virtual personality communication subject". It can be seen from this that the traditional anchor position will face the fact that it will be redefined. On the contrary, some scholars believe that AI streamers will still not be able to compare with human streamers for a long time. In terms of professional skills, AI anchors lack a sense of authority and have poor news credibility; In terms of non-language, the expressions and body language of AI anchors are not vivid; In terms of creation, AI cannot assume the function of an "ideological" moderator.
The empirical research on AI anchors focuses on the influencing factors of the audience's attitude towards them. Overall, the audience has limited understanding of this new media application, but the overall attitude is positive. Among them, Wang Yixi et al., based on the dual perspectives of new technological characteristics and social actors, found that technological novelty, credibility, and anthropomorphism have a positive impact on the attitude towards AI anchors. Factors such as the characteristics of the technology itself, social influence, and hedonistic motivation have a positive impact on young audiences' willingness to use AI anchors. In addition, some scholars pay attention to the audience's perception of the audio broadcast of AI anchors. The study found that when listening to reading materials, the proportion of listeners who chose AI and human anchors to read was basically the same, and there was no significant difference in liking and trust perception. Another experiment showed that there was no difference in the intention of querying news information and adopting news suggestions after listening to the weather news broadcast between the two subjects, but the credibility of human anchors was higher. An in-depth interview study found that news content, media organizations, and technical transparency are important factors affecting the audience's perceived credibility of AI anchors. In the above study, the questionnaire survey method was mainly used, and the definition and display of AI anchors were not clearly defined, so the cognitive bias of the participants towards AI anchors could not be ruled out. Or the experimental method is used to explore the audience's attitude towards some functions (voice) of AI anchors, and there is a lack of research on AI anchors in the form of video.
In general, there are few studies on AI anchors, and they are mainly based on phenomenon explanation and speculation, and empirical research, especially experimental research, is insufficient. "As an emerging communication practice, the social acceptance process of AI anchors is not yet clear." Therefore, this paper uses the control experiment method to explore the audience's perception and evaluation of AI anchors and real anchors, and experiments produce video materials for AI anchors and real anchors to broadcast news, and the audience perception is based on real and intuitive viewing experience, which is also one of the innovations of this paper.
3. Research Questions
(1) Types of anchors
AI anchors appear in the public eye as a new type of intelligent communication subject. The subject of communication is an important factor in mass communication activities, which directly affects the content, process and effect of communication. Carl Hovland and other scholars have summarized two important dimensions of media trustworthiness: professionalness and trustworthiness. Both refer to the communicator, the former refers to the ability of the communicator to be considered able to provide effective information, including the communicator's professional training, experience, and social background and values similar to the audience; The latter refers to the extent to which a communicator is perceived to be willing to provide the truth, including its reliability, impartiality, neutrality, and lack of specific motives and intentions. C. Pierce S. Peirce determined the paradigm orientation of communication semiotics from the ontology, pointing out that the meaning conveyed by signs is not static, but given by different communication subjects, and the recipient produces understanding in the process of symbol transmission, and the transmission process of signs is the result of negotiation between the two parties. In traditional media, the host, as the main body of communication, transforms the potential communication power into the actual communication effect, and the audience's evaluation of the quality of the program is greatly affected by them. What is the impact of AI anchors on the perception and evaluation of audiences, and what are the differences between AI anchors and real anchors? Therefore, the following research questions are proposed in this paper (Fig. 1):
RQ1: What is the impact of different types of communication subjects (AI anchors/live anchors) on the audience's content perception and program evaluation?
(2) Category of news
AI anchors are widely used in news production practices, covering various categories such as people's livelihood news, current affairs news, and weather forecasts. Among them, current affairs news and people's livelihood news represent two different forms of expression. In the current political news broadcast with strong policy and authority, the anchor assumes the important role of the "mouthpiece of the party" and conveys objective facts and political positions, so the announcer's expression is required to be accurate and objective. The people's livelihood news is more life-oriented and close, and most of the anchors choose to "talk the news" instead of "broadcast the news", with a relaxed and natural style and full of emotions. Previous studies have shown that news categories affect audiences' perception of the credibility of machine news, but does the same conclusion apply to AI-powered news? This paper selects two different types of news content, serious and objective current affairs news and friendly people's livelihood news, to examine the audience's perception and evaluation. Therefore, this paper proposes a second research question (Figure 1):
RQ2: What is the impact of news categories (people's livelihood news/current affairs news) on the audience's content perception and program evaluation?
(3) Platform attributes
At present, not only official media such as CCTV and Xinhua News Agency have launched AI anchors, but individuals can also independently customize AI anchors through technology companies and broadcast the generated news works on self-media platforms. With the diversification of media forms, the perception and evaluation of the audience are not only affected by the characteristics of the source and the content of the information, but also by the media or channel. Therefore, this paper takes official media and self-media as one of the variables to examine the impact of different broadcast platforms on audience perception and evaluation of AI anchors. The third research question is raised in this paper (Figure 1):
RQ3: What is the impact of broadcast platform attributes (official media/self-media) on the audience's content perception and program evaluation?
Audience attitudes towards different communication subjects can vary depending on the type of news and the platform on which it is broadcast. This paper proposes a fourth research question:
RQ4: Are the audience's content perception and program evaluation affected by different news categories and different broadcast platform attributes among different communication subjects? That is, is there an interactive relationship between the three independent variables of communication subject, platform attributes, and news category?
Cognitive dimensions such as trustworthiness and likability were shown to be important dependent variables. In addition, the sense of news substitution is an internalization process for the audience to receive news content, so that the audience can actively seek the development elements of the news, generate emotional resonance, and better understand the news. Therefore, this paper divides the dependent variable into two parts: content perception and program evaluation. Content perception includes credibility and news substitution, while program evaluation is divided into liking and overall quality perception. In summary, this paper proposes the following research model diagram (Fig. 1):
Fourth, the initial establishment of the party's emotional system and its influence
In this study, a between-subject factor experiment was conducted with 2 (anchor type: AI/real person), ×2 (news category: current affairs news/people's livelihood news), ×2 (platform attributes: official media/self-media). In the experiment, the subjects filled in questionnaires after watching different videos, and scored in terms of content perception (credibility, news substitution) and program evaluation (liking, overall quality).
(1) Experimental subjects
In this study, a random sampling method was adopted, and the questionnaires were distributed through the "Questionnaire Network" (https://www.wenjuan.com). A total of 548 questionnaires were collected, and 497 valid questionnaires were excluded from invalid samples and samples that failed to pass the operation test. As shown in Table 1, the gender ratio of the participants was about 2:3. The age distribution ranged from 18 to 70 years old, and the proportion of young and middle-aged people aged 18 to 40 was 88.33%. 93.76% of the participants had a college degree or above.
(2) Design of experimental materials
In this study, 8 different presentation methods of video stimulation materials were produced: live anchor + current political news + official media; Live anchor + people's livelihood news + official media; Live anchor + current political news + self-media; Live anchor + people's livelihood news + self-media; AI anchor + current political news + official media; AI anchor + people's livelihood news + official media; AI anchor + current political news + self-media; AI anchor + people's livelihood news + self-media.
1. Material selection
Choose a live streamer. First of all, avoid the anchor's own popularity is too high, causing the subject to be affected by past impressions. Second, reach the level of professional news anchors and have rich work experience. Combined with comprehensive factors such as popularity, social recognition, and business ability, this study selected a news anchor who has worked in the news column of provincial media for 10 years as the recorder of the live anchor's video.
Select the AI anchor. First of all, a virtual synthetic image that can represent the current level of development of AI anchor technology. Second, it is the same gender as the real anchor and has a similar visual age. Based on the above factors, through the screening of AI anchors on CCTV, Xinhua News Agency, Baidu AI, iFLYTEK, Qingbo and other platforms, Xiaoxuan, the AI anchor developed by iFLYTEK, was finally selected as the experimental object. Xiaoxuan uses AI technologies such as speech synthesis, image processing, and machine translation to create it, and its visual presentation can represent the current level of technological development of AI anchors, and the image and temperament are also consistent with the experimental materials.
Select a news topic. First of all, exclude recent hot news topics to avoid the participants being influenced by existing reports or public opinion. Second, controversial topics should be excluded to avoid participants being distracted by existing attitudes. Thirdly, the experimental materials selected one current political news and one people's livelihood news. Fourth, the theme is consistent. The theme was determined to be "Qinghai Environmental Protection". Qinghai Daily's "Provincial Party Secretary's Field Investigation on the Ecological Protection of Qinghai Lake" was selected as the current political news material, and China Environment Daily's "Protecting the Beautiful Qinghai's "Face Responsibility" was selected as the people's livelihood news material. Finally, according to the text content of the two reports, two professional announcers from traditional media rewrote the oral transcript in line with the reporting standards. Among them, there are 370 words in the current affairs news and 416 words in the people's livelihood news.
Video rendering. First of all, all 8 videos presented by the combination of different factors were produced into a unified studio space. Second, according to the different attributes of the platform, the official media station is marked as CCTV, and the self-media is marked as bilibili. All use platform logos with high audience familiarity. Third, the video format is MP4 to ensure that the video size and resolution are consistent. In current political news, the broadcast time of the AI anchor is 1 minute and 32 seconds, and the average speech speed is 4.02 words per second; The broadcast time of the live anchor is 1 minute and 30 seconds, and the average speaking speed is 4.11 words per second. In Minsheng News, the broadcast time of the AI anchor is 1 minute and 36 seconds, and the average speaking speed is 4.33 words per second; The broadcast time of the live anchor is 1 minute and 40 seconds, and the average speaking speed is 4.16 words per second. The host's costumes, shooting scenes, studio backgrounds, station logo design, and manuscript content in the videos of real anchors and AI anchors are all consistent.
2. Distribution of materials
Embed the created video in the online questionnaire and set it to appear randomly. Participants can click the play button to watch it, and only after watching the full video can they enter the questionnaire to fill in the questionnaire.
3. Measurement scales
This study tests the audience's perceived credibility, news substitution, liking, and overall quality evaluation of the video. A 5-point Likert scale was used for all variables.
Reliability. The proposed credibility measures were used: "accurate", "true", "credible" and "authoritative". Scale reliability α=.89.
News substitution. Yang Wei, Yang Wei, Guo Zhongshi, "News Content, Understanding and Memory: A Mental Model for Interpreting Controversial Event Reports", Journal of Journalism and Communication Research, No. 11, 2016. The scale proposed by "When you watch the video just now, the picture of the news story can easily appear in your mind"; "You can pay attention when you watch the news video just now"; "This report makes you feel empathy"; "You want to know the ending of this news story". Scale reliability α=.90.
Likeability. The news content liking scale proposed by Sundar was adopted. Including "boring", "enjoyable", "interesting", "vivid", "pleasant". Scale reliability α=.90.
Overall quality. Using a direct evaluation approach, participants were asked the question: "What do you think of the overall quality of the program?" ”
5. Results
IBM SPSS Statistics 26 was used for data analysis.
(1) Demographic variables
Cross-over analysis showed that there were no significant differences in gender (X²(7, n=497)=2.00, p=.96), age (X²(21, n=497)=18.05, p=.65) and education (X²(21, n=497)=26.25, p=.20) among the eight video groups.
One-way variance and independent-samples T-test were used to analyze the differences in education, age, and gender in the dependent variables. Table 2 shows that different educational backgrounds have no significant effect on credibility, liking, substitution, and overall quality. Age significantly affects the audience's perceived credibility, liking, sense of substitution, and overall quality of the video. Gender had no significant effect on likability, immersion, and overall quality, but it did have a significant effect on credibility, specifically women perceived videos as more credible than male audiences.
The results of the correlation test (see Table 3) showed that there was a low but significant positive correlation between age and credibility, liking, sense of substitution and overall quality, indicating that the older the video, the higher the evaluation of the video. However, there was no significant difference in age between groups (see Table 2) and was not treated as a covariate in subsequent analyses. Gender was only associated with significantly low confidence and was evenly distributed across groups, so covariates were not treated in subsequent analyses. There was no significant relationship between educational attainment and the dependent variable.
(2) Results
In order to answer research questions 1-4, the multivariate ANOVA method was used to find that (see Table 4) anchor type, news category and platform attributes significantly affected the participants' content perception and program evaluation. In addition, the interaction effect shows that there is no second-order or third-order effect for anchor type, news category, and platform attributes.
(2) Results
In order to answer research questions 1-4, the multivariate ANOVA method was used to find that (see Table 4) anchor type, news category and platform attributes significantly affected the participants' content perception and program evaluation. In addition, the interaction effect shows that there is no second-order or third-order effect for anchor type, news category, and platform attributes.
1. Anchor type
The results showed that the type of anchor had a significant effect on the dependent variable. As shown in Tables 4 and 5, the type of anchor significantly affects the credibility: F(1,489)=15.09, p=.000<.001, and partial η²=0.03. Among them, the credibility of the video broadcast by the real anchor (M=4.02, SD=0.80) was significantly higher than that of the video broadcast by the AI anchor (M=3.69, SD=0.96). The anchor type significantly affected the likability: F(1,489)=17.03, p=.000<.001, partial η²=0.03). Compared with the video broadcast by the AI anchor (M=3.48, SD=1.05), the participants preferred the video broadcast by the real anchor (M=3.85, SD=0.86). The anchor type significantly affected the sense of substitution: F(1,489)=15.65, p=.000<.001, and medium effect (partial η²=0.03). The immersion sense of the video broadcast by the real anchor (M=3.91, SD=0.86) was significantly higher than that of the video broadcast by the AI anchor (M=3.55, SD=1.02). The anchor type significantly affected the overall quality: F(1,489)=16.93, p=.000<.001, and partial η²=0.03. The video quality of the live anchor (M=3.81, SD=0.80) was significantly higher than that of the AI anchor (M=3.48, SD=0.91).
2. News Category
As shown in Tables 4 and 5, news categories have no significant effect on the overall quality: F(1,489)=0.04, p=.84. There was no significant difference between the quality of people's livelihood news (M=3.64, SD=0.91) and current affairs news (M=3.63, SD=0.84). However, the news category significantly affected credibility: F(1,489) = 4.20, p=.04, small effect (partial η²=0.01). The credibility of people's livelihood news (M=3.93, SD=0.77) was significantly higher than that of current affairs news (M=3.76, SD=1.01). The news category significantly affected the liking: F(1,489)=3.99, p=.05, and small effect (partial η²=0.01). Compared with current affairs news (M=3.56, SD=1.07), the subjects preferred people's livelihood news (M=3.75, SD=0.88). The news category significantly affected the sense of substitution: F(1,489)=5.82, p=.02, and small effect (partial η²=0.01). The participants' sense of substitution for people's livelihood news (M=3.83, SD=0.87) was higher than that of current affairs news (M=3.62, SD=1.05).
3. Platform Attributes
As shown in Tables 4 and 5, the platform attributes had no significant effect on the overall quality: F(1,489)=1.01, p=.32. There was no significant difference between the video quality published by official media (M=3.69, SD=0.87) and that of self-media (M=3.59, SD=0.88). However, platform attributes had a significant effect on credibility: F(1,489)=5.21, p=.02, and small effect (partial η²=0.01). The credibility of videos released by official media (M=3.96, SD=0.79) was significantly higher than that of videos published by self-media (M=3.76, SD=0.97). Platform attributes had a significant effect on likability: F(1,489)=6.51, p=.01, and small effect (partial η²=0.01). The subjects preferred the videos released by official media (M=3.79, SD=0.87) over the videos released by self-media (M=3.55, SD=1.05). Platform attributes had a significant effect on the sense of substitution: F(1,489)=6.09, p=.01, and small effect (partial η²=0.01). The participants thought that the sense of substitution in the video released by official media (M=3.85, SD=0.86) was significantly higher than that of the video released by self-media (M=3.63, SD=1.03).
VI. Conclusions and Discussion
As the main body of communication, news anchors have undergone changes from specialization, elitism, generalization to intelligence. With the rapid development of AI, the absolute dominance of live anchors has been challenged, and whether AI anchors can replace live anchors has sparked heated discussions. From the perspective of the audience, this paper discusses the content perception (credibility, news substitution) and program evaluation (liking, overall quality) of the news broadcast by AI anchors. Different from the existing research conclusions that AI and real human identities do not significantly affect the perception of participants, this study found a significant difference: compared with real people, the audience's content perception and program evaluation of the news broadcast by AI anchors are significantly lower. The reasons for this may be as follows: First, previous studies have rarely used real experimental materials, and the audience's perception of AI anchors is only based on imagination or past experience, but this paper produces a video of AI anchors to provide intuitive feelings and judgments. Second, the broadcast of oral news depends to a large extent on the creative ability of the communication subject. In the expression and creation of the host, the movement process of spoken language can be summarized as: psychological-physiological-physical-physiological-psychological. The activity process of the first two (psychological and physiological) is the processing process of the communicator, the physical activity refers to the operation process of the communication media, and the activity process of the latter two refers to the information receiving process of the audience. At this stage, the intelligence and "situational cognition" ability of AI anchors are weak, and they have not been able to get rid of the problems of voice mechanization and simplification of expression. Thirdly, the experimental material in this paper shows the anchor's seated position, which is mainly shown through the upper body, especially the face, and the importance of eye contact is magnified. In order to avoid the stiff eyes caused by the teleprompter, professional anchors usually use a wider angle of view and a slightly downward camera setting to create a sense of equal participation and interactive realism for the audience. However, AI anchors are deficient in simulating the naturalness and liveliness of this kind of eyes, and unnatural visual characteristics sometimes trigger an uncomfortable "uncanny valley effect", which may also be one of the reasons for the significant differences in audience evaluations. Finally, the research conclusions confirm the irreplaceability of live anchors and the value of live anchors to improve their professional determination. In the new media era where the right to speak is scattered, real anchors should give full play to the irreplaceable initiative of AI, enhance the creative value of human subjects, adhere to authority and professionalism, and improve news expression. Pay attention to the temperature of information dissemination, be a regulator of social emotions, and be a watchman of fairness and justice. With the continuous development of technology, the deep integration of AI and media has become the development trend of journalism, and real anchors should change their thinking, turning the "opponent" of AI anchors into "helpers", and the cooperation between humans and machines can effectively complement each other and integrate benignly and deeply.
In terms of news categories, the credibility and sense of substitution of people's livelihood news were significantly better than those of current political news, and they were more popular with the participants. Minsheng news itself has the characteristics of being close to life and presenting stories. In the survey of news programs preferred by the audience, the top two are Minsheng News Wu. Telling people's stories in a "humane" way is undoubtedly more intimate and inspires deeper emotional resonance, which may be the main reason for its high sense of substitution, credibility and love. When it comes to platform attributes, videos identified by state media are considered more credible, more immersive, and more liked. It may be related to the type of news program selected for this experiment. The research shows that current affairs news, people's livelihood news, and cultural content are the most popular types of official media reported by the audience, and the official media is considered to have the mission of representing the mainstream ideology and disseminating public information, and has strong credibility. The advantages of official media in multiple dimensions in the results of this experiment also coincide with its advantages in news credibility, public opinion guidance and social influence.
At the theoretical level, firstly, the conclusions of this paper provide empirical support for the CASA paradigm and illustrate its applicability and limitations in explaining the phenomenon of human-computer interaction. The CASA paradigm argues that people naturally and unconsciously perceive computers or AI as social actors and interact with them socially. Although the data shows that compared with live anchors, AI anchors have lower scores in credibility and liking, their mean values are still higher than the medium level, and the gap with live anchors is small (only about 0.3), indicating that although the audience can clearly distinguish between live anchors and AI anchors, the overall acceptance and evaluation of AI anchors is still positive. This shows that while the audience prefers live streamers, it does not negate the CASA paradigm. The study found that people were indeed interacting with AI streamers in a social and emotional way, but due to a number of factors, such as the limitations of delivering social cues such as facial expressions and voice intonation, it still failed to achieve the same effect as human streamers. This result is actually a practical example of the CASA paradigm. In addition, the CASA paradigm has its limitations. The CASA paradigm focuses more on the unconscious social response of individuals in the face of machines, and it was proposed in the era when computers began to be popularized, and people's understanding of AI and AI itself was less intelligent and popular. An important implication of the CASA paradigm in the field of human-computer interaction design is that social cues can be used to make the process of interaction more "social" and "emotional", that is, closer to human interaction. However, with the rapid development of AI, some scholars have pointed out the potential problems of this anthropomorphic tendency—excessive anthropomorphism may lead people to see the machine as a companion, which can lead to disappointment or other undesirable consequences if it fails to fully meet social and emotional expectations; Maintaining certain machine properties and treating them as tools can reduce this expectation, i.e., "de-anthropomorphize", limiting its tendency to be over-anthropomorphic and retaining its machine properties. In recent years, some scholars have proposed that AI is becoming more and more widely permeated into human life, and we should jump out of "anthropocentrism", which should not only be regarded as a tool for imitating human behavior and emotions, but also as an entity with its own unique functions and limitations. This perspective is especially important in today's increasingly digital age, helping us to understand the social, psychological, and cultural impact of AI more fully, helping to lower people's unrealistic expectations of AI. Therefore, in human-computer interaction design, it is not necessary to pursue anthropomorphism excessively, but should pay attention to the functionality and efficiency of AI, and coexist harmoniously with AI. Secondly, the previous experimental studies on the application of AI news have paid more attention to the problem of "subject identity", that is, by informing the participants of different subject identities (human or AI) to form experimental stimuli, so as to obtain the participants' evaluation of AI news applications, such as AI poetry and painting creation, machine news writing, etc. As mentioned above, this study makes the audience's perception based on real and intuitive experience, rather than based on the imagination of technology, through the self-made video materials of AI anchors and real anchors, which is also one of the innovations and contributions of this paper. Both experimental studies are of great significance for exploring the human-machine relationship in the context of AI, both from the distinction of subject identity and the difference between subjects based on practice.
First, this study did not directly explain the basic technical principles of AI anchors to the participants, and the participants may have different criteria for judging the operation mechanism of AI anchors, such as whether the news release is generated by an algorithm or manually entered by humans, and questions like this will affect the judgment of the participants. Secondly, only two types of news, current political news and people's livelihood news, will be examined, and the impact of more categories and other dimensions will continue to be explored, so as to provide more theoretical perspectives and empirical support for the audience's acceptance mechanism of AI anchors. Finally, in terms of experimental materials, although the consistency of the two anchors is pursued as much as possible in terms of gender, visual age, clothing and hairstyle, it is still impossible to rule out the influence of individual perceptual differences caused by factors such as appearance and timbre. Ideally, a prototype of an AI anchor would be used for live anchor video recording. However, at present, most of the AI anchor prototypes do not have the business ability and professionalism of news anchors, and a small number of AI prototypes built according to the modeling of professional anchors are news anchors of mainstream media platforms, such as CCTV's AI Kang Hui, Xinhua News Agency's AI Qu Meng, etc., which cannot avoid the interference of the experimental results due to the influence of the anchor's own popularity.
This paper is the research result of the National Natural Science Foundation of China project "Statistical Laws and Modeling of User Behavior in Mobile News Media" (Project No.: 11875005).
Originally published in Journalism and Communication Research, Issue 3, 2024
Due to space limitations, the official account will be left with annotations, and the full version can be found in the publication
(Editors: Intern Editors Zhang Jingyi, Cui Mengtian)
References:
Liu Na, Li Qian, Liu Qian, Wu Ye: Can Artificial Intelligence Anchors Replace Real Anchors? An Experimental Study Based on Audience Perception and Evaluation", Journal of Journalism and Communication Research, No. 3, 2024.
LIU Na,LI Qian,LIU Qian,WU Ye. Can AI streamers replace live streamers? Experimental research based on audience perception and evaluation[J].Journal of Journalism and Communication,2024(3):47-61.)