Perspective Chapter: Artificial Intelligence in Slovak Radio Industry – The Present and the Future of Broadcasting

4.1 The creation of radio texts through generative AI

In the recent years, there has been a critical breakthrough in which systems based on artificial intelligence principles can build a sufficiently complex language model based on the learning process to generate results linguistically close to human-produced text [20]. This technology started to assist journalists in various aspects of their work, offering opportunities for creativity, innovation, and improved efficiency. Generative AI tools learn from large datasets (including images, text, audio, and video) and use that knowledge to generate new samples that are similar to the training data. They can then generate new instances of data that possess similar characteristics, structure, and distribution to the training data [21]. Content generation involves the use of AI algorithms to automatically create news articles, reports, summaries, and other journalistic content. This dimension emphasises the ability of AI systems to generate coherent and contextually relevant texts, mimicking human-like writing styles [21]. Alawida et al. [22] point out that although generative AI tools (such as ChatGPT, Gemini, or Copilot) can help with many tasks, particularly those with high levels of repetition and redundancy, they are not a replacement for human intelligence. They explain that human intelligence is capable of understanding context, interpreting meaning, and making connections. They specified the differences between human-generated text and AI-generated text as follows:

• human-generated text is characterised by the ability to convey meaning and intent, which also reflects cultural and emotional intelligence;

• human-generated text uses figurative language, idiomatic expressions, and cultural references, which vary across settings and audiences;

• AI-generated text is based on patterns and stored data, and therefore may struggle to take into account and understand cultural and social contexts;

• AI-generated text may also contain grammatical, stylistic, and word-formation errors;

• AI-generated text may not take into account or recognise language that is considered exclusive or derogatory based on social norms and other cultural circumstances.

These findings are particularly evident in texts that are generated in a language other than English. One example is an error in Norway newspaper. The AI-generated text said Norwegian football star Erling Haaland was shot [23]. In this case, it was a problem of translation, because the AI tool interpreted the English word “shoot” not as “photographed”, but “shot”. The problem of the language barrier was also confirmed by our previous research [18, 24], specifically that current AI tools cannot generate Slovak radio news fit for broadcast without further editing by a news presenter.

According to the 2023 global survey, 90% of the 105 participating news and media organisations from 46 different countries (excluding Slovakia) said they use generative AI (such as ChatGPT) in their news production phase [8]. As far as Slovakia is concerned, according to Tinák and Gáliková Tolnaiová [25], the interest in AI in newsrooms is undoubtedly growing, but the actual implementation process is happening at a slow pace, and the use of AI is still in the process of initial exploration and experimentation. Mikušová [26] adds that many Slovak journalists are unfamiliar with AI tools. Unlike global counterparts, Slovak newsrooms lack formalised policies to guide the adoption and integration of AI into their workflows. Current AI applications in Slovak journalism are limited to routine tasks, and advanced uses such as data-driven reporting or automated content creations remain unexplored. The Institute for Public Affairs stated that 57% of Slovaks reported experience with different versions of ChatGPT, 18% have used or tried CoPilot, 12% Gemini, and under 5% have experience with Claude or similar AI tools [27]. Out of the most used generative AI tools in Slovakia—using the mentioned “universal” prompt—the best rewriting of news piece from news agencies into the form of radio report was made by Claude (Table 3).

Although there are still problems with the Slovak language within the generated texts, the standard of these texts is not worse than what can be heard in the current radio broadcasts of Slovak commercial stations. The most problematic is the use of verbs in the passive voice. We assume that this problem comes from the fact that the primary language of these tools is English and not Slovak, so they cannot distinguish between the verb form in the “active voice” and the “passive voice”. The same is valid for technical expressions. Concurrently, a significant linguistical shift in the development of generative AI tools can be observed. A research from 2023 states that these tools are not applicable in the Slovak radio industry, as multiple commands were needed to generate the basic text, which would still have to be additionally edited by the news presenter [24]. However, the current results show that these (single-prompt) generated news reports—after minor and quick adjustments by the news presenter—could be used in any radio station in Slovakia.

4.2 The use of voice-based AI and its reception by listeners

Voice-based artificial intelligence is revolutionising the landscape of radio broadcasting, enhancing the way content is delivered and consumed. Advances in natural language processing (NLP) and deep learning have enhanced the fluency, and realism of AI-generated speech, raising questions about listener reception, trust, and usability [28].

Trust is crucial in radio broadcasting, as the credibility of the voice delivering information can significantly affect listener perception. Liu [29] argues that perceived credibility depends on the level of agency assigned to AI voices—whether they function autonomously or are controlled by humans. Transparent AI voice systems that clarify their artificial nature tend to foster more trust than those that obscure it. Becker et al. [30] suggest that the naturalness of an AI-generated voice plays a pivotal role in compliance and trust. Their findings indicate that more human-like voices encourage users to engage longer and perceive the system as more reliable. However, Patel et al. [31] warn that highly realistic AI voices may cause ethical dilemmas, as users may struggle to differentiate AI from human speech, potentially leading to misinformation or manipulation. Moreover, accent and pronunciation influence trust in AI voices. Pycha and Zellou [32] found that listeners rated AI voices with standard accents as more credible than those with regional or foreign accents. These findings suggest an inherent bias in listener perceptions, reinforcing the need for more inclusive and culturally diverse AI voice models.

Emotional resonance remains a challenge for AI-generated voices. While AI systems can generate speech with high accuracy, conveying emotion and nuanced intonation remain difficult. Campanilea et al. [33] found that emotionally expressive AI voices significantly improve listener engagement compared to monotonous voices. Their study emphasised that incorporating pitch modulation, rhythm variation, and prosody enhances the perceived naturalness of AI speech. According to Wang’s [34] study, AI anchors can only convey the literal meaning of the text in the transmitted handwriting, and cannot convey the deeper meaning of the language and the text, particularly in terms of flexibility in tone, pitch, and pauses. The mechanical nature of AI presenters can compromise the effectiveness of communication with the audiences, especially in large live broadcasts. These presenters may lack the necessary communication skills and in-depth thinking to engage and interact with listeners, especially during large live broadcasts.

Despite these challenges, Slovak radio Europa 2 introduced a robotic news presenter “Eva” in 2019 to deliver the news block on weekdays at 4 pm. The response from listeners was largely unfavourable, with feedback such as “It lacked a human touch”, “It failed to evoke emotion”, and “It sounded like a text-to-speech conversion via Google Translate”. In the Czech Republic, they created a position for artificial intelligence in the role of a presenter. This was done at Rádio Express FM, which is owned by Seznam.cz. They are the first in the Czech Republic to use an AI-generated synthetic voice. It comments on current events, but also on songs played on the air, and works during the night broadcast. The AI presenter was named “Hacsiko”, and the original voice was provided by the presenter of the morning show Morning Club, Bára Hacsi. Similar to the Slovak audiences, the Czech listeners reacted rather negatively to the AI presenter: “I do not like Hacsiko”, “There’s a certain artificiality”, “I’ve heard it, I have not warmed up to it yet”, “The voice is the same, but the sentences are always in the same intonation, […] boring”. One comment said that he would not have known the difference between human and AI broadcasting without prior warning, and he only saw the difference in the machine not being able to make mistakes and typos, which are “the spice of broadcasting” [11]. On the other hand, Kim et al. [35] investigated individuals’ perceptions and reactions to a weather broadcast presented by an AI newscaster compared to a human newscaster. The results showed that while people perceived a human newscaster as more credible than an AI newscaster, their reactions to the news content did not differ.

Listeners are not emotionally cold to AI-generated voices, as confirmed by our ongoing research into listener perceptions of radio programmes. During the research, participants were told that some of the voices they heard were generated by artificial intelligence. Based on electrodermal activity (EDA) data and emotion analysis, we can observe significant changes in participants’ emotional responses, suggesting that this disclosure had an emotional impact. Figure 1 shows a peak in electrodermal activity (GSR peaks) around the 10th second, which correlates with the moment when participants received the information about the use of AI, which may be related to surprise or excitement.

At the same time, an increase in emotions such as disgust and sadness is evident, suggesting that some participants reacted negatively to this information. They may have been surprised that the voices they perceived as authentic were not real. However, there was also a slight increase in levels of joy and surprise, suggesting that some participants may have found the situation amusing, possibly due to suspicion that the voices were not human, or were surprised by the quality of the synthesised voices. Engagement remained at a relatively high level throughout the research, suggesting that participants remained engaged with the content and situation until the end of the test. Despite the relatively long duration (35 minutes on average), we consider this to be a positive aspect of conducting this type of survey. This may also be of interest for future research that focuses on people’s reactions to synthesised voices versus real voices.

With the growing prevalence of AI-generated voices, ethical concerns surrounding their use have become more pressing. Mirek-Rogowska et al. [36] discuss the risks associated with AI-generated misinformation, emphasising that AI voices can be manipulated to spread false narratives. This ability to manipulate AI voices for deceptive purposes poses risks for credibility in media. Broadcasters must navigate these challenges by implementing robust ethical guidelines to ensure responsible use of AI technology, thus safeguarding their audiences from potential manipulation.

4.3 The ethical aspects of using AI in radio broadcasting

The ethical aspects of the use of AI are a major topic; in the visual domain, we encounter numerous examples of people being abused or manipulated. Similar situations are not avoided in the radio industry either. For example, the Polish radio station OFF Radio Kraków broadcast an interview conducted by an AI-generated presenter who, with the help of an artificially created voice, pretended to interview the Polish poet and Nobel Prize winner Wislawa Szymborska, who died in 2012. The radio planned to broadcast a similar interview with Polish statesman Józef Piłsudski, who died in 1935 [37]. The radio had previously fired a number of presumably redundant editors, which resulted in such mistakes, for example hallucination of AI. This may be a clear signal that manpower and skills are still needed and irreplaceable. Not only for their knowledge, but also for their ethical compass, and responsible approach to journalistic values.

While artificial intelligence (AI) has the potential benefits for human beings, it also gives rise to a number of ethical considerations that must be given due attention. As these systems are increasingly deployed across a range of sectors, AI has the potential to exert a significant impact on a number of key areas, including credit, employment, education, and competition [38]. However, without ethical considerations integrated into the design of AI algorithms, it is challenging to ensure that AI will not facilitate actions by certain actors that result in more harm than good. With the broader implementation of AI in recent years, it has been employed to undermine public trust in information, and has been implicated in perpetuating discrimination in the delivery of services and in creating unfavourable profiles of segments of the population, raising numerous other ethical concerns [39].

The absence of a standardised approach to the utilisation of AI in newsrooms has facilitated the advent of innovative and adaptable initiatives. While each news organisation has pursued its own projects, the absence of a universally accepted set of standards has resulted in the development of disparate ethical frameworks, which could result in potentially creating new challenges [40]. The publicly available generative AI tools have pre-set ethical limits; thus, they do not generate absolutely everything for the user, and do not process some inputs (prompts). Concurrently, it should be noted that it is very important how the prompts are entered. In general, however, tools that use large language model (LLM) should not provide content that promotes racism, xenophobia, anti-semitism, homophobia, sexism, or any other form of discrimination, or content that is outright illegal. Those can be topics related to the realm of crime—drugs, guns, murder, child pornography [39]. In this case, however, AI must not only know the realities of the area in which the text or information is processed, but also the legislative definition of specific areas. AI should be designed and used in a way that respects human dignity and privacy, and is not a tool to discriminate or harm individuals. Algorithms should be unbiased, and data should be managed with an eye to the presence of biases that could lead to unfair outcomes [41].

The ethical aspects of radio journalism are summarised in the internal codes of ethics of individual radio stations. This includes, for example, the formulations for citing a source, or obtaining information from multiple information relevant sources. This is an important aspect of the use of AI, that no source is mentioned when generating texts, the same applies to graphics. The ethical aspects of working with AI are also related to the legal framework, the cases mentioned above are related to copyright law, and it is debatable whether content created artificially can be considered copyrightable at all. However, there are also other cases that need to be dealt with—cloning of existing people and their voice, reviving the voice of the dead, but also plagiarism in a school environment [9]. Šantavý [42] says in the context of growing confidence in AI systems, we can and must set conditions without which the deployment of AI systems in the real world would not be possible. In general, AI systems must be: legal—comply with the required standards, laws, and regulations; ethical—meet the required ethical criteria; safe—achieve the required standards of security and robustness. The integration of AI in journalism necessitates a multifaceted approach. It requires the adaptation of existing ethical frameworks, the establishment of new standards tailored to journalism, the enhancement of AI literacy, the maintenance of editorial control, the assurance of data governance, the addressing of biases, and the fostering of industry-wide collaboration. Through these measures, the journalism industry can effectively harness the benefits of AI while upholding its integrity and maintaining public trust [12]. This was summed by Kim [40] and is displayed in Table 4.

The ethical compliance of AI journalism, as represented by the balance among transparent practices, bias mitigation, and the influence of AI, is a growing concern in academic and professional circles. The existing literature emphasises the significance of transparency in the context of AI-driven news processes, underscoring the necessity for transparency and clarity in the manner by which news is algorithmically curated. Another notable challenge pertains to the potential for algorithmic bias. The possibility that AI systems may perpetuate biases, if left unchecked, has the potential to compromise the integrity of journalism. As the role of AI in newsrooms continues to expand, scholars have highlighted the importance of maintaining editorial independence and safeguarding journalistic ethics in the context of rapidly advancing technology [12].

In this context there are listed several core principles for ethical AI use in media, in which belong:

• Ethical foundations which link AI guidelines to core journalism values like trustworthiness and accuracy.

• Transparency which means to clearly communicate AI usage, specifying its role in content production.

• Disclose AI contributions in articles or content pieces to build public trust.

• Human oversight where we can assign clear responsibility for monitoring AI systems to ensure accuracy.

• Defined boundaries with specified acceptable and unacceptable uses of AI aligned with organisational values.

• Data protection and algorithmic fairness, where address AI biases by leveraging locally relevant systems and robust oversight.

• Education and engagement in which trained staffs help audiences understand AI.

• Collaboration where we should foster internal and external partnerships to ensure ethical AI use.

• Dependency management means to clarify and plan for reliance on third-party systems, exploring custom solutions.

• Enforceability is about establishing mechanisms and consequences for guideline violations.

• Media diversity stands for preserve distinct editorial voices and prevents homogenisation [43].

Brník et al. [44] claim that the work processes in editorial rooms have changed significantly. These changes include the shortening of reports, the incorporation of infotainment elements into broadcasting, the diversification and adherence to journalistic genre standards, and shifts in news preparation and fieldwork processes, such as transitioning between editorial rooms. For instance, a presenter who previously served as an announcer is now tasked with producing news, recording material, and occasionally creating new genres, such as assemblages or reportage. AI can effectively assist with all these tasks.

The integration of AI in journalism necessitates a multifaceted approach. It requires the adaptation of existing ethical frameworks, the establishment of new standards tailored to journalism, the enhancement of AI literacy, the maintenance of editorial control, the assurance of data governance, the addressing of biases, and the fostering of industry-wide collaboration. Through these measures, the journalism industry can effectively harness the benefits of AI while upholding its integrity and maintaining public trust [12].

The only thing that remains is to wait and see how the radio industry is able to adapt to the emerging needs of ethical frameworks in the context of AI, and also whether this powerful tool can influence the future of human potential in the broadcasting environment. Listeners, who are the pillars of listenership and preference, will have the last word anyway; if they prefer humanity and intercession, artificial intelligence will not be a competitor for presenters and editors, but if listenership perception slips to the level of adjacency and below-average attention is paid to the broadcast, artificial intelligence will be abundantly sufficient.

4.4 Description of the current situation in the Slovak radio industry in relation to the use of AI in practice

The Slovak radio industry is currently undergoing a technological transformation, with artificial intelligence at the core of these changes. Although traditional radio broadcasting still dominates, a number of Slovak radio stations have started to introduce AI tools into various areas of the broadcasting process to improve their operations. These range from browsing online articles and recommending personalised content, to automated news reporting, or broadcasting content through AI-generated voices.

As mentioned above, AI was first used in Slovak radio broadcasting in December 2019, when radio Europa 2 introduced a female news presenter, “Eva”, who reported the news every weekday at 4 pm. However, it should be remembered that AI technology and the tools were not as well developed before now, and as a result, reactions to the AI news presenter were also largely negative, and the broadcaster discontinued the project fairly quickly. Since the launch of ChatGPT in late 2022, many radio stations around the world have started to implement AI tools in various areas, from playlist creation to AI DJs, or news anchors.

Slovak public service and private broadcasters have also recognised the potential of AI. One of the first broadcasters to use an AI DJ was the private station Rádio Expres. In early 2024, they introduced a night-time DJ called “Robo”. According to Marian Staráček (Innovation Specialist at Bauer Media Slovakia, which owns Rádio Express, Rádio ROCK, Rádio Melody, and Europa 2), the company’s goal was not to create a perfect copy of a human being, but only to fill the night time, which would not be hosted [45]. In addition to the AI DJ, the company is currently using AI voices to commercials. The second most listened to private station—Fun rádio—has decided to use AI in a different way. As part of a temporary image campaign, they trained AI DJs (based on the voices of real DJs) to create personalised audio messages for listeners. Roman Janajev (Creative Tech Lead of the company that worked with Fun rádio on this project) said, “The AI DJs’ comments are generated using GPT-4 with fine-tuned prompts that include the personality traits and presentation style of specific Fun rádio DJs. The result of the generated prompts is further transferred to ElevenLabs with the rehearsed DJ voices. Their model also preserves the specifics of each DJ’s diction” [46]. The programme director of Fun rádio, Marek Mikúšek, added that in general they are using AI very cautiously so far, but that they are using it mainly for tasks related to information retrieval, preparing creative texts for broadcasting, advertising texts, or creating various sound elements such as jingles or podcasts [47].

As for public service Slovak radio (as an organisational component of Slovak Television and Radio, which consists of nine stations), the first radio to include AI in its broadcasts was Radio Slovakia International (RSI). RSI broadcasts information about Slovakia in six languages (English, French, German, Russian, Spanish, and Slovak) for listeners around the world and for foreigners living in Slovakia. The newsroom has started using AI to dub interviewees into French. Kristína Hanáková (French editor) explained that in the past they used to get help from people in the newsroom or external collaborators, which was difficult to organise. However, “Mathilde and François (AI voices) are immediately available and their performance in French is hardly distinguishable from a human voice” [48]. The most listened public service radio station—Rádio Slovensko—took a different approach. It invited its listeners to send 10 words to the editing room, which were then inserted into ChatGPT to create an original 2-minute short story. These are then narrated by Slovak actors and the selected short stories are broadcast on the radio as part of the morning prime-time show [49].

In addition to the radio stations with national reach, several regional radio stations have also started to use AI tools. Trnavské rádio introduced an AI editor “Ivana Šnajder” in early 2024, whose role is to cover regional content that is no longer within the capacity of human resources. According to the newsroom, this works in practice as follows: the field editor collects material and prepares a report, which the AI then fine-tunes textually, and prepares as an audio output for the newsroom [50]. Dobré rádio, on the other hand, has included a segment in which they “take ChatGPT on air”. According to DJ Michaela Kicková, the way it works is that they ask ChatGPT a question and it then creates an answer through its AI voice, which they broadcast. At the same time, they also use it in music, for example, asking it to sing a song by Ema Drobná (a Slovak singer) in the voice of Michael Jackson [47].

On the other hand, not all Slovak radio stations are currently using AI in broadcasting. One example is Viva radio. DJ Dávid Schun said that they had tried it in their editorial room, but the generated texts were inaccurate, and it was necessary to check the facts again. He added that he could see AI being very useful in the pre-production phase of radio in the future, but only if it provided 100% accurate information [47].

Despite these advances, the introduction of AI in Slovak radio is still at an early stage compared to global trends. Ethical concerns about authenticity, displacement of jobs, and credibility of content remain topics of debate in the industry. However, as AI technology develops and regulatory frameworks evolve, its role in shaping the future of Slovak radio broadcasting is expected to grow.