When an automated state system strips a pensioner of their monthly grant, who steps into the breach? When a police facial-recognition camera misidentifies a township resident as a criminal suspect, where does the victim file an appeal?
As African governments rush to automate public services, these questions are no longer hypothetical. Ministers present machine-learning software as a cure for administrative decay. Yet behind the rhetoric of digital modernisation lies an acute constitutional dilemma: the erasure of state accountability behind proprietary code.
During a recent discussion on The Angle Podcast Ep 51 | Who is accountable when an algorithm harms someone?, public policy specialist Dr TK Pooe, professor of Statecraft and Foresight at the African School of Governance in Kigali, cut straight to the political core of this crisis: “We have politicians who drive on broken roads with blue lights while visiting poverty during funerals without fixing the structural reality. If we are serious about statecraft, our benchmark must be absolute accountability. When you introduce technology into public administration, you cannot use code to hide from your constitutional obligations.”
However, research from the Tayarisha African Centre of Excellence in Digital Governance at the Wits School of Governance working paper series, AI and Digital Governance Politics and Management: A View from the Global South, demonstrates that automating public services without institutional capability simply automates inequality.
Public sector governance specialist Kenneth Riany tackles what harm comes from letting algorithms manage public resources in his research paper, Algorithmic Bias in AI Governance: Strategies for Promoting Data Sovereignty in Kenya’s Public Service Delivery.
Riany’s paper exposes how computational rules reproduce social marginalisation through an investigation of automated systems deployed across Kenya, like GovBot on the e-Citizen platform and Kenya Revenue Authority tax profiling. He notes that governance deficits precede technological failures and when public sector algorithms are trained on skewed, urban-centric datasets, they produce systematic allocation bias. In tax enforcement, automated profiling aggressively targets informal micro-enterprises while high-value corporate evasion slips past undetected.
The Kenyan experience shows that algorithmic harm is not a technical glitch. It is a structural transfer of power from vulnerable citizens to opaque software models. When an automated platform denies an informal trader a municipal trading permit or flags a rural clinic for reduced medical supplies, the affected citizen encounters an administrative wall. The bureaucrat claims innocence by pointing to the screen, while the private software vendor disclaims public liability behind intellectual property clauses.
Their mutual abdication of duty is supported by a procurement trap. Across Africa, cash-strapped state departments buy proprietary software suites off the shelf from foreign technology vendors. The state purchases the finished product but often lacks the technical competence to audit the code, deploy optimally and provide first- and second-level support internally.
In his working paper, E-Governance and the Kerala Model: Unpacking the Changing Dynamics of Dependency and Discretion, public policy scholar Ashwin Varghese examines how dependency on external vendors destroys public accountability. Analysing administrative reforms in the Indian state of Kerala, Varghese documents how relying on external corporate providers created a severe “design-reality gap” that disrupted welfare distribution.
Varghese argues that dependency on external service providers strips public officials of discretionary judgment. When departments outsource software development, they end up with systems misaligned with citizen needs. Kerala responded by rejecting corporate lock-in and established in-house technical capacity within the civil service to build public technologies using free, open-source software.
Kerala trained police officers and administrative personnel through the Digital University of Kerala to code their own civic tools. The approach made sure that public servants understood how decisions were generated and preserved human discretion alongside direct legal responsibility. By contrast, African bureaucracies remain trapped in vendor dependency. Instead, many purchase black-box platforms that leave officials unable to explain why an algorithm made an adverse decision against a citizen.
The danger deepens when automated governance merges with security infrastructure. In their critical study, Governing AI from Below: An Ubuntu-Anarchist Critique of State-Led Digital Transformation in Africa, researchers James Maisiri, Leroy Maisiri and Blessing Mbalaka document how centralised digital tools expand state surveillance rather than citizen empowerment. They found that state-led digital transformation frequently mirrors colonial patterns of centralised social regulation. They also found that from automated biometric tracking in Zimbabwe to private algorithmic surveillance networks operating across South African suburbs, technology has become an instrument of social control rather than public service.
Maisiri and his co-authors warn against allowing political elites to co-opt African philosophical concepts. “Ubuntu is frequently weaponised as a moral slogan by political leaders to demand compliance,” they argue. “Without radical, bottom-up democratic oversight, state algorithms will consistently prioritise elite interests over community well-being.”
Advocates of public sector automation will consistently invoke efficiency as a defence. They argue that algorithmic scoring eliminates petty bribery while also clearing heavy administrative backlogs across chaotic municipal offices. Human bureaucrats can be corrupt and biased. Replacing an obstructive clerk with an automated rules engine appears, on paper, to protect public funds.
In the South African context, however, the country’s constitutional reality collapses this technocratic defence. Section 33 of the Constitution guarantees every citizen the right to administrative action that is lawful and procedurally fair. When an algorithm executes a decision, procedural justice requires that the affected person receives intelligible reasons for the outcome. A proprietary algorithm whose internal logic cannot be audited by the court violates this constitutional guarantee.
In this looping debate, automated code becomes a convenient evasion of state duty. An algorithm is not a legal persona; it cannot be cross-examined in a court or voted out of office. Direct legal and political liability rests entirely with the state organ that chose to deploy the code.
Furthermore, this sort of political evasion carries even more severe material consequences. As Pooe emphasised on The Angle Podcast, artificial intelligence is not an ethereal cloud phenomenon; it requires heavy physical infrastructure. Global technology firms are building power-hungry data centres across Gauteng and the Western Cape in South Africa which consume scarce municipal water for cooling systems and pull megawatts from an unstable electrical grid.
“We are in a country facing acute water scarcity and energy strain,” Pooe warned. “Now foreign tech monopolies want our water and our electricity to power data centres that run foreign models, while our own public universities lack the funding to build sovereign public infrastructure. What is the national interest in that trade-off?”
Africa cannot afford to sleepwalk into algorithmic governance. If the state deploys automated systems to allocate welfare or police streets, it must establish clear statutory liability. National legislation must mandate algorithmic impact audits and guarantee accessible human appeal channels.
When code inflicts harm, the state remains the perpetrator. Outsourcing administrative decisions to machines does not dissolve the social contract, it tests whether our constitutional protections can survive the digital age.
This is the second of five stories from the series. It is derived from the podcast: https://www.youtube.com/watch?v=Wd6oWpcikrs&t=19s
Earl Joseph is the founder of The Digital Afrikan, a digital media company in Johannesburg. He is also an award winning journalist and commentator on African technology policy, governance and institutional transformation based in Johannesburg.
Stanley Moloto is the editor, host and producer of The Angle Podcast which focuses on the African digital innovation ecosystem. He is a digital entrepreneur, web developer and co-founder of The Digital Afrikan.