-
Vow to block deceptive AI content
-
Microsoft commits to joining local election situation room
-
Tech giants admit gaps in local-language moderation, voter education
____________________________________________________________________
From Juliana Taiwo-Obalonye, Abuja
Microsoft and Google have pledged to support efforts to monitor and tackle election misinformation ahead of Nigeria’s 2027 general elections, with Microsoft committing to provide a representative for a locally organised election situation room if invited.
The commitment came amid concerns raised by election stakeholders, civil society organisations and researchers over the capacity of technology platforms to detect deceptive artificial intelligence-generated content, moderate harmful material in Nigerian languages and respond swiftly to election-related threats.
Speaking during an interactive session at the AI and Democracy Forum in Abuja, Dave Leichtman, Senior Director, Global Elections at Microsoft, said the company remained committed to combating deceptive AI use in elections and would work with the Independent National Electoral Commission (INEC) as the authoritative source of electoral information.
Leichtman said Microsoft would use INEC’s final list of candidates, once released, to restrict image generation involving the listed individuals, while maintaining measures to address misleading election content and the unauthorised use of candidates’ likenesses.
“We are going to take the finalised list from INEC when it is finalised,” he said, explaining that the commission’s information would guide the company’s measures.
He said Microsoft and other technology companies had signed the Tech Accord to Combat Deceptive Use of AI in Elections in 2024 and that, although the accord had formally expired at the end of that year, the company continued to maintain its commitments.
According to him, the company has mechanisms for reporting misleading election content and the unauthorised use of political candidates’ likenesses, with provisions for removing such content.
Leichtman also gave an assurance that Microsoft would participate in an election situation room established locally to monitor threats in real time, provided the company received an invitation.
“Microsoft is 100 per cent willing to send somebody to a situation room locally. We will do that,” he said, adding that he had personally represented the company in similar arrangements in the United States and Canada.
The forum was organised by Yiaga Africa in partnership with INEC, Microsoft, TikTok, the Centre for Journalism Innovation and Development, Dataphyte, the MacArthur Foundation, the International Foundation for Electoral Systems, the German Agency for International Cooperation and other partners.
The discussions focused on the responsibilities of technology companies in protecting electoral integrity as Nigeria prepares for the 2027 general elections.
Adewolu Adene, Google’s Head of Government Affairs and Public Policy for West Africa, said the company was working to improve its ability to recognise harmful content in Nigerian languages, while acknowledging that language diversity and insufficient training data remained significant challenges.
He said Google was collaborating with civil society organisations and other stakeholders to improve the contextual understanding of local languages and strengthen the ability of its artificial intelligence systems to identify abusive and potentially harmful expressions.
Adene, however, cautioned against assuming that technology companies had already developed perfect solutions to the problem.
“It would be deceitful to stand across you and say that it is perfected and we have the perfect model and solution,” he said, describing the work as ongoing.
He also disclosed that Google had received recognition from the National Information Technology Development Agency (NITDA) for its reporting on the implementation of the code of practice.
The technology companies faced questions from participants over the effectiveness of their content moderation systems in detecting election-related threats concealed in local languages, live broadcasts and seemingly harmless online conversations.
Iveli Ube, Executive Director of the Centre for Redefining Alternative Civic Engagement for Africa, said online gender-based violence and other forms of digital abuse frequently occurred in languages other than English.
She asked how the platforms incorporated local expressions and cultural contexts into their moderation systems, particularly as political campaigns and online conversations intensify ahead of elections.
Rosemary Joines of Digital Africa Research at Saint Kilab also raised concerns about the exploitation of live broadcasts and comment sections to spread election misinformation.
She cited examples of users who initially communicated in apparently harmless language before switching to content intended to mislead or manipulate audiences.
Joines asked how the platforms monitored such patterns and whether they proactively studied the methods used by individuals seeking to circumvent their moderation systems.
Responding, Adene said the scale of content uploaded to platforms such as YouTube made the task particularly demanding.
He said YouTube receives about 500 hours of video content every minute, requiring a combination of artificial intelligence systems, human moderators and user reports to identify potentially harmful material.
Adene said Google was working to improve its systems’ ability to recognise local expressions and identify abusive content, but stressed that technological tools alone could not address every emerging threat.
He said reports from users could help platforms identify patterns of suspicious activity and develop a better understanding of how their services were being exploited.
Leichtman said Microsoft operated threat intelligence and threat analysis teams that monitored cybersecurity threats and disinformation narratives, including those associated with foreign state actors.
He explained that the company worked with civil society partners to identify misleading narratives and reduce the likelihood of harmful disinformation being surfaced through its search engine and artificial intelligence platforms.
He said the company also sought to make authoritative information more accessible to users.
“For elections information, INEC is the authoritative source. You need to know when, where and how to vote,” he said.
Leichtman added that Microsoft would work to ensure that relevant information from INEC was proactively surfaced to users seeking election-related guidance.
Participants also questioned the companies about how they would verify electoral information supplied by the commission, given the possibility of revised lists and other official updates.
Leichtman said Microsoft would rely on INEC’s final candidate list when it was ready, rather than treating preliminary lists as final.
The discussions also highlighted concerns about the use of AI chatbots to influence voters’ choices by recommending particular candidates.
Leichtman said Microsoft’s policy was to provide information about candidates and compare their positions on issues without recommending one candidate over another.
“We will tell you about the candidate if you ask. We will compare candidates’ stances on issues, but we will not recommend a candidate over another,” he said.
Adene similarly described AI tools as instruments that should support informed decision-making without taking away the agency of voters.
He said technology platforms should help users assess information and understand issues, rather than determine electoral choices for them.
On mechanisms for redress, Adene said users and law enforcement agencies had channels through which they could report harmful content for review.
He explained that law enforcement agencies could use dedicated reporting channels to alert platforms to suspicious accounts, recurring patterns and potentially harmful activity.
Users could also flag content they considered abusive or misleading, although he cautioned that a report did not automatically guarantee immediate removal.
Depending on the circumstances, he said, platforms could remove specific content, restrict its distribution or block access to a channel.
Leichtman also urged users to report inaccurate responses generated by AI chatbots, explaining that such feedback helped Microsoft identify errors and improve its systems.
He cited the feedback mechanism on Microsoft’s Copilot platform, including its thumbs-down option, as a channel through which users could flag incorrect responses.
On data protection, Adene said Google operated access controls and permissions governing the information collected and processed through its products and services.
He urged users to understand the permissions they granted when using digital platforms, while emphasising the importance of compliance with applicable data protection requirements, including Nigeria’s data protection framework.
Leichtman said Microsoft, Google and other platforms also had different policies on political advertising, with some companies prohibiting it on their services.
The forum also heard concerns about the accessibility of digital election information and AI-related voter education in rural communities, particularly in northern Nigeria and areas affected by insecurity.
Participants from Bauchi and Borno States said many citizens lacked adequate information about voting, digital technologies and AI, while security challenges and displacement continued to complicate access to electoral information.
A participant working with internally displaced persons in Borno State questioned how technology companies could reach communities where insecurity restricted physical access and digital literacy remained limited.
Leichtman acknowledged that Microsoft’s programmes for underserved communities in Nigeria were not as extensive as he would like them to be.
He said the company supported technology skills development and had expanded broadband access to 50 million people across Africa over the preceding five years.
However, he admitted that the scale of the challenges meant technology companies could not reach every community on their own.
Adene said Google relied on networks of civil society organisations, academics, researchers, community leaders and local media practitioners to develop culturally relevant educational resources.
He cited an initiative in Kaduna State involving the development of AI learning materials using animation and Hausa language as an example of how local knowledge could be incorporated into technology education.
He urged stakeholders to extend such initiatives beyond urban centres and work with community-based organisations to reach populations that were often excluded from digital policy discussions.
The participants also raised questions about the availability of accurate polling-unit information on Google Maps, asking how the company was collaborating with local authorities to ensure voters could locate their polling units, including those in remote areas.
Adene said the availability and reliability of underlying data, as well as regulatory approvals, affected the company’s ability to provide such information.
He explained that Google generally relied on available and verifiable data and would not publish information without securing the necessary permissions and authorisations.
In his closing remarks, Leichtman cautioned stakeholders against using the term artificial intelligence as a blanket description for different technologies, arguing that doing so could unnecessarily heighten public anxiety.
He said AI could refer to several distinct applications, including chatbots, generative images, back-office tools and systems used to process information.
He noted that some electoral processes had relied on technologies such as fuzzy matching of names and optical character recognition for years, long before the recent surge in public attention to generative AI.
“If we are just doing optical character recognition on forms, we should say that. We shouldn’t say we are using AI; we should say we are digitising forms,” he said.
Leichtman urged election authorities, technology companies and civil society organisations to communicate clearly about the specific technologies being deployed and the purposes they serve.
Adene, in his closing remarks, said the public debate should not focus exclusively on technology but also on the people using it and the responsibilities that accompany its use.
He said the same tool could be used constructively or harmfully, depending on the intentions and actions of the person wielding it.
His message to voters, election officials and law enforcement agencies was to retain human judgement and responsibility when using AI tools.