Holding power to account through generative AI

It is a year since IMS began supporting the development of an artificial intelligence news presenter with our Zimbabwean partner CITE. Since then, Alice has made waves in the global media development community – drawing into focus the fast-moving technology shifts shaping journalism.

The sandy driveway and disused swimming pool on the grounds of the Bulwayo-based offices of CITE belie the pioneering technology developments happening within. “We have developed programmes with Alice where I interact with her as a fellow presenter,” said Zenzele Ndebele, managing director at CITE about their weekly podcast, This week on CITE. “It allows us to hold power to account in new ways. I can ask Alice about key facts and figures that have been published by leaders and that we want to challenge without getting the same criticism we would have had previously.”

Last year, CITE introduced the use of artificial intelligence in the newsroom as an innovative way to engage and grow its audiences. Following a mapping report of AI use cases in newsrooms and knowledge exchange between media in Africa, they were supported by IMS’ Digital Publishing Lab to evaluate a number of opportunities and AI use cases – from discriminative artificial intelligence based tasks such as newsroom assistants, coding and sorting, to generative use cases – including a series of consultancy meetings to understand how AI would be adopted into its daily news production. As a result, CITE chose to produce an AI-generated news reader by the name of Alice to carry out different programmes, including the Brief News Bulletin, Rate Your Councillor and Meet Your Candidate. CITE also managed to use AI to be more efficient in news distribution by using a scheduling tool to plan a timetable that would be used to distribute news content through CITE’s social media platforms.

The news reporters gather news and write it and Alice then reads the top three stories selected by the editor. They combine a number of tools, including video editing tools Flexclip and Synthesia, to package the stories and have Alice read them through the text to speech function. The stories are logged into the video editing tool with subtitles. Alice has an X handle @Aicitezw, from where she has received invitations out for a coffee.

Gains to newsroom production

A daily news bulletin was previously not possible without Alice because of the time and number of people involved in the production. With the introduction of AI, only one person is in charge of producing the news bulletin. The time saved is estimated at around three hours per week, as it takes less time than using people; CITE can record the weekly podcast programme in 10 minutes. The introduction of the news bulletin also opened up CITE to engaging with audiences with disabilities through video and audio formats. The Brief bulletin includes subtitles, which makes it easy to read if they cannot listen to the bulletin. Rate Your Councillor, an election programme proposed for Alice to present, aimed to provide residents with the chance to evaluate the performance of their councillors on feedback meetings, accessibility and campaign promises; the programme promoted oversight and accountability and facilitated the practice of demanding quality service and stewardship from elected officials. A total of 19 videos were published and presented by Alice. Alice presented Meet Your Candidate, which aimed to foster electoral and political accountability while also seeking to empower voters to participate effectively in the electoral process. Contesting candidates from Bulawayo, Midlands, Matabeleland North and Matabeleland South were given five minutes to present their manifestos to the electorate and tell the voters why they should vote for them. Incumbents had to tell the electorate what they achieved in the previous term. The five-minute programmes were published on all of CITE’s media platforms. A total of 49 candidates from various political parties were profiled.

Learning on the go

The AI avatar that CITE used was changed after feedback from an audience on the AI presenter’s relatability to the ordinary person. Despite opting for a female presenter out of well-meaning gender representation intentions, they received some criticism that it meant women’s jobs were more easily replaceable by AI than men’s. Among the foremost challenges is the accurate replication of local accents and dialects—a nuanced undertaking that demands careful consideration and refinement. Interdisciplinary work is under way on how to develop local language modules, which is an extensive problem across Africa and many regions where IMS works. Workarounds include rewriting scripts with incorrect spellings, using existing large language models as a base and then grounding the data to adapt them further and experimenting with voice cloning. They have gone on to develop a male presenter, Vusi, for news casting, with further nuance and adaptation to the appearance of the avatars.

The impact of Alice has been felt globally. The CITE team has been invited to various international fora to give presentations and participate in panel discussion on AI-powered newsrooms, including in Jordan at ARIJ, in the POLIS Journalism AI report as a case study, in Baraza Media Lab in Kenya, on radio stations and by other media in, for example, Rwanda. Several articles have been written including at WITS centre for journalism. Zenzele has also been selected as a trainer in the JournalismAI Academy for Small Newsrooms targeting small newsrooms in sub-Saharan Africa.

Institutional capacity building and monetisation

Such is the pioneering development pace at which the team is developing and reflecting on their institutional policies and guidelines to incorporate these new AI functions. Locally they have been discussing with other media houses as well as at the media commission level in Zimbabwe to impact policy development within broader ICT frameworks. IMS is also working with CITE to harness the strategic and business opportunity.

“Alice brought in her first paid work as we used her as part of a sponsored content offer to the Zimbabwean road agency,” said Zenzele. “They wanted Alice to explain what they had spent the budget on and it worked more effectively than them doing a presentation. It was an explainer as to how they spend their budget.

“We are also working with a parking company for Alice to explain what you get clamped for in the city, as this is a big problem. We built the opportunity by first developing social listening on the topic so that we could demonstrate there was a problem, and then we could bring Alice in as a solution.” CITE is exploring other opportunities with local universities, companies and councils.

Audience perception of Alice

An audience study was carried out about Alice by Prof Mphathisi Ndlovu. This qualitative study used in-depth interviews and focus group discussions to gain insights on audiences’ perceptions of Alice. These interviews were conducted in both online and offline spaces. The 50 participants included journalists, journalism students, journalism educators and other individuals. CITE’s social media manager also provided insights on the work of Alice. To make sense of audience perceptions of the AI news presenter, user comments on Twitter, Facebook and Instagram were also analysed. Over 200 tweets and Facebook comments were analysed. The scope of study was May 2023 to August 2023. The aim of the study was to generate deeper insights on audiences’ reception of Alice. Alice has garnered a positive reception, particularly among younger audiences, affirming the potential of AI as a complementary tool rather than a replacement for human journalism.

Some participants thought the news was read by a human, despite being labelled and introduced as generative artificial intelligence. These users rebuked Alice on Twitter for the mispronunciation of local names as if they were addressing a “real” person. Other participants added that they need “human expressions” in newsreaders, but Alice “lacks human command”. Despite having imitating capabilities, AI machines lack cultural sensitivity, as is evident in the failure of Alice to pronounce specific words. This shortcoming has triggered expressions of annoyance and disappointment from some audiences who perceive Alice as undermining local cultures. According to Prof Ndlovu in the audience study: “To understand this concern with Alice’s mispronunciation of local names, it is important to note the politics of language in Matabeleland. There is a concern in Matabeleland that local languages such as IsiNdebele are marginalised in the country. Alice’s mispronunciation evokes memories of the perceived and realities of linguistic and cultural subjugation of local people.”