As Nigeria prepares for the 2027 elections, artificial intelligence is emerging as a tool for real-time election monitoring and verification. During the Osun governorship election, MundX Analytics used AI to track polling-unit results uploaded to INEC’s IReV portal and flag possible inconsistencies. In this interview with BusinessDay, Martin Uwakwe, AI expert and tech policy advocate at MundX, discusses the Osun experience, the potential and limitations of AI in election monitoring, threats from AI-generated disinformation, and how technology can strengthen electoral transparency ahead of 2027. TAOFEEK OYEDOKUN brings excerpt:
MundX used AI to monitor the Osun governorship election in real time. What exactly were you monitoring, and what did your system reveal that conventional election observation may have missed?
What we did was monitor the election results coming in from all the polling units across the state.
At the end of voting at the polling units, INEC uploads the EC8A polling-unit result form to the IReV portal. What we did was use AI to extract, in real time, all the results and accreditation details from the EC8A forms and run them through a set of validation checks.
We looked out for overvoting, anomalies, arithmetical errors and other inconsistencies, then collected the results and presented the information to the public.
So, as the results were coming into the portal, people could get a sense of what the results looked like and the direction the eventual outcome was likely to take.
Your platform relies on polling-unit result forms uploaded to IReV. How did you collect, process and independently verify those results? What safeguards did you put in place to prevent false or manipulated data from entering your system?
The IReV portal is a public portal. Anybody in the world can access it and obtain the data. What we are using AI to do is to perform that process at scale and in real time. Usually, election observers and candidates would go to the IReV portal after the election, download the results and begin analysing them manually. We do all of that in real time and at scale using technology and AI.
There are a lot of things we do before we authenticate our result sheets because we understand the risks involved in using technology and AI for a use case like this.
We have checks and balances and validations to make sure that whatever results we obtain are as accurate as possible and as close to the official INEC declarations as possible.
The first thing we do is use what we call a bimodal consensus approach. We use two independent AI models to extract the results and votes. The first AI model extracts the votes and the second also extracts the votes. We then compare the two results. If both match, the result is validated.
We check the individual scores of all the political parties against the total vote count to ensure they align. We also check that the number of valid votes plus rejected ballots equals the total votes cast.
We check that valid votes do not exceed the number of accredited voters and that the number of accredited voters does not exceed the number of registered voters.
All these things happen in real time. In less than two minutes after a result sheet has been uploaded, we are able to carry out these validations.
By the time we finished collecting results across the state and INEC announced the official result, we found that what we had collected using AI was about 95 percent accurate.
Why were you unable to achieve 100 percent accuracy?
There are several reasons.
First, we don’t have all the polling-unit results on the IReV portal. Unfortunately, INEC has not been able to achieve 100 percent upload of polling-unit results on IReV, so there were some polling units for which we could not obtain the actual EC8A forms.
Also, because it is AI, it will most likely not be able to read a blurry EC8A form. If the result sheet is so blurry or mutilated that even a human being would struggle to make sense of it, that result will not be counted by our system.
During collation, particularly at the ward level, INEC may also cancel some polling-unit results. If those results are subsequently cancelled, that may not be reflected in the AI-driven analytics we conduct.
So these are some of the reasons we cannot achieve 100 percent accuracy. But the goal is not necessarily 100 percent accuracy. That is not our north star.
Our north star is to see the trend and direction of the election in real time and determine whether the votes cast by the people at the polling units are reflected in the final collation announced by INEC.
One of the strengths of the platform is that every result can be traced back to its original polling-unit form. Why is that traceability important?
With my work as an AI governance professional, I understand that for people to trust a system, they should be able to trace exactly how a result was obtained.
That is why, on our platform, we have a page where we explain our methodology and the checks that every polling-unit result passes through. We are able to follow every step between when we receive a result and when we provide the data on our website.
It is important that when people see a number from a particular polling unit, they can compare it with what is actually on the EC8A form.
On our website, you are not just seeing the numbers. You are also seeing the actual polling-unit result. That gives people more confidence in the system and makes the process transparent.
What happens in 2027 if polling-unit results are delayed on IReV, uploads stop or the platform becomes unavailable? What alternative sources can MundX rely on?
We have been very clear about how we obtain our results. We are not going to rely on social media results. We want to focus on the IReV portal.
The confidence we have is that IReV is now provided for in the Electoral Act 2026. The chances of the IReV portal not being used for the 2027 general election are therefore very low because there is a legal mandate for INEC to upload polling-unit results to the portal.
The Electoral Act also provides a fine of N500,000 for a polling-unit officer who fails to upload the result to IReV. We have confidence, looking at what INEC has done with previous elections, in its ability to deploy IReV.
During the Osun election, we noticed a period of about two to three hours when the IReV portal was down. During that period, there was no upload to the portal. After a while, just before midnight, we started seeing results coming in.
There are possibilities of technical hiccups and challenges. But if you are collecting results from more than 170,000 polling units across the country, a two- or three-hour delay on IReV does not necessarily invalidate the collation process, particularly when you consider that manual collation usually takes several days.
With AI, we are able to get the results by the morning after the election.
How scalable is the MundX model? Can you realistically monitor and analyse polling-unit results nationwide during the 2027 presidential election?
This is pure technology. We were able to do it in Osun. The only thing we need to do is scale our systems and technical infrastructure, which is what we are currently working on.
It is possible. It is doable. It is very expensive, but we are committed to doing it.
Let’s talk about the flip side of AI. How serious is the threat posed by deepfakes, fake result sheets and misleading political narratives ahead of 2027?
AI brings a lot of risks, especially around elections and public policy, that we have to be aware of.
Even with the work that we do at MundX Analytics, we make it very clear that we do not declare election results. We don’t have that mandate. Across our website, there is a clear disclaimer that only INEC declares the winner of an election.
Our work is to make the process transparent. People can search for their polling units, find the polling-unit results and see the collation we are doing in real time. But nobody announces election results.
I think INEC has to do a lot of data education. There are going to be fake results and fake result sheets on the internet. People will be sharing numbers, figures and data.
It is important that for every polling-unit result being shared on social media, people are able to trace it to the IReV portal.
I saw this during the Osun election. I saw some funny-looking result sheets. The moment I saw them on social media, I suspected that they were AI-generated.
Because the result sheets had the polling-unit name, ward and LGA, I went to the IReV portal and searched for those polling units. What I found on IReV was different from what people were sharing on social media.
At that point, I knew the result being circulated was likely fake or AI-generated. There is little or nothing we can do to prevent bad actors from creating this kind of content. But we need a lot of data education.
INEC and civil society organisations have a major role to play in ensuring that people are educated and know how to independently verify information they see online.
Is MundX open to collaborating with civil society organisations, observer groups, newsrooms and other technology platforms ahead of 2027?
Of course. We have already started receiving enquiries and having conversations with other civil society organisations to see how we can scale these efforts efficiently and serve the Nigerian people.
Based on what MundX observed in Osun, what are the three most important things INEC, political parties, civil society organisations and voters should do now if technology is to make the 2027 election more transparent and harder to manipulate?
For INEC, it is important to understand that, from the way we conducted the Osun election and previous elections, there are now very high expectations from Nigerians regarding technology.
We expect INEC to do everything within its power to ensure that the IReV portal remains stable throughout the election period. If INEC needs to invest in the technology and infrastructure to ensure that people can access the IReV portal without hiccups on election day, it should do so.
If there is a delay or technical challenge, no matter how innocent, it can give people room to begin to believe that the election is being manipulated. INEC has to get the technology right.
For political parties, they should understand that this is 2026. With the help of technology, the electoral process is no longer the black box it used to be some years ago.
People can now independently verify almost every step, from the polling unit to ward collation, LGA collation and ultimately the governorship or presidential level.
Knowing that Nigerians are watching and independently following the process, I expect political parties to abide by the Electoral Act 2026 and play by the rules.
For civil society organisations and observer groups, they need to invest in independent verification. They need to set up and train observers, particularly in the use of technology, so that they can independently verify whatever INEC uploads to IReV.
For Nigerian voters, it is important to understand that there will be bad actors who will take advantage of the 2027 elections to spread misinformation and cause confusion. Voters need to be discerning enough not to believe everything they see on the internet.
Finally, what should Nigerians do when they encounter election information online that they cannot verify?
If something cannot be independently verified, and you cannot go to reputable platforms such as BusinessDay and other credible media platforms to verify the information, it is your responsibility not to share or spread it.
At the end of the day, fake information hurts everybody because there can be so much information circulating that people cannot distinguish what is true from what is false.
Educate yourself and understand that there are going to be bad actors. You need to be prepared to spot them when you see them.
Taofeek Oyedokun is a correspondent at BusinessDay with years of experience reporting on political economy, public policy, migration, environment/climate change, and social justice. A graduate of Political Science from the University of Lagos, he has also earned multiple professional certificates in journalism and media-related training. Known for his clear, data-driven reporting, Oyedokun covers a wide range of national and international socioeconomic issues, bringing depth, balance, and public-interest focus to his work.


Comments
Start the conversation about this story.