Application and Ethical Review of Artificial Intelligence Technology in News Writing and Editing and Distribution

From September 15 to 25, 2024, questionnaires were distributed to 20 districts in Guangdong Province, and 836 valid questionnaires were received (56% female and 48% male). The basic profile of the respondents is shown in Table 1. Among the respondents, 3.17% were under 18 years old, 67.28% were between 18 and 30 years old, 22.58% were between 31 and 50 years old, and 6.97% were over 50 years old; 69.75% resided in urban areas, and 30.25% resided in rural areas. In terms of educational level, 59.41% have less than a bachelor’s degree and 40.59% have a bachelor’s degree or higher.

The study mainly investigated audience news receiving habits from three aspects, including total time of use, frequency of use, and time of use. In terms of total usage time, 33.57% of participants used algorithm-recommended information platforms to receive news for more than three years, 44.02% used them for one to two years, and only a few participants used them for less than half a year; in terms of usage frequency, 78.62% of participants used them every day, with 42.36% using them more than three times a day; and in terms of usage time, 67.40% of participants spent less than half an hour each time, 8.31% spent more than an hour each time, and 54.36% spent more than an hour each time. 67.40% of the participants spend less than half an hour each time, 8.31% spend more than an hour each time, 54.38% of the users do not use these apps at a specific time, and 36.72% of the users focus on using these apps after 5pm. It can be seen that in terms of participants’ use of information apps based on algorithmic recommendations, which are characterized by high frequency, short time and randomness, algorithmic news not only leads to fragmented reading habits, but also fragments users’ time.

The study categorized the education level of the respondents into elementary, junior high, high school, college, graduate, and doctoral and above education users. The analysis of variance (ANOVA) of perception and attitude towards algorithmic news by education level is shown in Table 2.

According to the ANOVA, users with different education levels have different views on the same issue, especially between users with elementary school background and users with other education backgrounds, and the sig value is less than 0.05, which indicates that there are significant differences between users with different education backgrounds on the same viewpoints, which means that education background affects users’ attitudes and perceptions towards algorithmic news. Among the seven points of view, only the last point of view of “reinforcing inequality” has no significant difference in the views of different groups, while for the other six points of view, there is a significant difference in the users’ views. It can be seen that the education factor has a significant impact on users’ attitudes towards algorithmic news.

In this study, a three-group ABT experimental design was conducted, with a total of 90 rented cloud phones representing 90 virtual users. Experimental groups 1 to 3 (referred to as “group 1”, “group 2” and “group 3”) were each assigned 30 cell phones, while the data of the control group was generated by random sampling without the participation of cloud phones. The IP addresses of all cloud phones are set in Beijing, and the hardware configuration and parameter settings are guaranteed to be the same. The experiment was conducted from September 9 to September 16, 2023, with each cloud phone running for 5 minutes every hour between 9:00 and 21:00 every day for a total of 7 days.

Group 1 simulates the situation that the algorithm system pushes news videos for users in a similar “cold-start” situation when users do not reveal their news preferences to the platform; Group 2 and Group 3 simulate the situation that users reveal their news preferences to the platform to varying degrees; and Group 4 and Group 5 simulate the situation that users reveal their news preferences to the platform to varying degrees. Group 2 and Group 3 simulate the situation that the algorithm system curates news according to personalized needs when users “reveal” their news preferences to the platform in different degrees; the control group randomly selects news videos accessible to users on the platform and measures the news value orientation of these contents to reflect the news value preferences of the industry itself.

The distribution of time difference kernel density is shown in Fig. 1, when the user’s news preference is not clear (Group I), the average time lag between the release of the news video and being tweeted is the longest (M = -76.52, CV = -3.48, Median = -13.97). And when users showed a preference for news (Group II), the mean time lag was shortened (M=-43.73, CV=-3.82, Median=-9.21).

Figure 1.

Time difference kernel density distribution map

In terms of novelty, the cosine distance kernel density distribution is shown in Figure 2, where the average cosine distance between the pushed videos is the largest and highly concentrated around 1 when the user does not specify his or her interest (M = 0.987, CV = 0.006, Median = 0.995). This scenario has the lowest degree of similarity between the pushed news videos, which can be considered as the most novel; once the virtual users show different degrees of news preference (i.e., Groups II and III), the mean cosine distance between the pushed videos changes in the shape of the distribution, and its mean starts to shrink, with the degree of fluctuation becoming larger and the tendency to become highly centralized being weakened (Group II: M=0.980, CV=0.028, Median=0.993; Group III: M=0.978, CV=0.035, Median=0.993). However, the shortest mean cosine distance between pushed videos appeared in the control group (M=0.971, CV=0.052, Median=0.991), and as can be seen in Figure 2, the highest proportion of randomly selected videos from news accounts had the highest proportion of similar topics.

Figure 2.

Cosine distance kernel density distribution

Through the rooted theory analysis of the interview data, it was concluded that there were three secondary themes under the primary theme of “environmental awareness” and 12 tertiary themes under the secondary theme, as summarized in Table 3.

The 12 tertiary theme measures of the environmental cognitive level and the values measures were correlated, and the Pearson correlation coefficient was used to detect the degree of correlation between the secondary themes and the values of the social cognitive level, and the results are shown in Table 4.

As can be seen from Table 4, circle information is significantly and positively correlated with environmental cognitive values at the 0.01 level (P = 0.000 < 0.01), group belonging is significantly and positively correlated with environmental cognitive values at the 0.05 level (P = 0.019 < 0.05), emotional contagion is significantly and positively correlated with environmental cognitive values at the 0.05 level (P = 0.041 < 0.05), and self value identity is significantly positively correlated with environmental cognitive values at the 0.01 level (P = 0.000 < 0.01), indicating that with the increase of the four index measures of circle information, group belonging, emotional infection, and self-worth identity, the more the individual’s values at the environmental cognitive level converge to the socialist core values.

As can be seen from Table 4, stereotype threat is significantly negatively correlated with environmental cognitive values at the 0.01 level (P = 0.002 < 0.01), and situational attribution is significantly negatively correlated with environmental cognitive values at the 0.01 level (P = 0.008 < 0.01), which suggests that with the increase in the metrics of the two indexes of stereotype threat and situational attribution, the more detrimental it is to the individual’s environmental cognitive level values toward goodness of development.

The deeper the individual’s perception of stereotype threat, the more unfavorable it is to the identification of socialist core values. This finding is consistent with the research hypothesis. Relevant studies have shown that stereotype threat leads individuals to behave differently than usual, as well as psychological betrayal and lack of identity. Major algorithmic news platforms, which use traffic, hot words, and interests as the main algorithmic basis, are especially prone to amplify certain types of topics and produce psychological oppression on individuals.

The above clarifies that under the premise of the application of artificial intelligence technology in news, circle information, group belonging, emotional infection, and self-worth identity are positively correlated with values at the level of environmental cognition, while stereotype threat and situational attribution are negatively correlated with values at the level of environmental cognition. The six secondary themes of circle information, group belonging, emotional infection, self-worth identity, stereotype threat, and situational attribution were analyzed by multiple linear regression with the values of environmental cognitive level as the independent variable and the values of environmental cognitive level as the dependent variable, and the results are shown in Table 5. As can be seen from the Sig. values in the table, the differences in circle information and self-worth identity are significant (P < 0.01), the differences in stereotype threat, emotional infection, and situational attribution are significant (P < 0.05), and there is no significant difference in group affiliation (P > 0.05).

In order to make the regression equation more accurate, the third-level theme of “group belonging” was deleted, and five second-level themes of circle information, emotional infection, self-worth identity, stereotype threat, and situational attribution were used as the independent variables, and the values at the level of environmental cognition were used as the dependent variables to conduct the multiple linear regression analysis again, and the results are shown in Table 6.

From the Sig. values in the table, it can be seen that the differences of circle information and self-worth identity are significant (P < 0.01), and the differences of stereotype threat, emotional infection, and situational attribution are significant (P < 0.05). Setting the values of environmental cognitive level as Y2, circle information as X1, stereotype threat as X5, emotional infection as X9, situational attribution as X11, and self-worth identity as X12, and using unstandardized coefficients, the linear regression equation is: Y2 = 26.619 + 0.531X1 - 0.291X5 + 0.172X9 - 0.185X11 + 0.535X12

From this equation, it can be seen that the positive and negative values of the coefficients in front of the respective variables are consistent with the results in the correlation analysis. According to the setup of the scale, the value interval of X1, X5, X9, X11, X12 is [2, 10], and the value interval of Y2 is [10, 50], and the equation can be used to describe the relationship between the values of environmental cognitive dimensions and these five independent variables, but it is only suitable for reference due to the different rules of assigning the values of specific algorithmic procedures.